Emergency Management Systems and Algorithms: a Comprehensive Survey

Article Emergency Management Systems and Algorithms: a Comprehensive Survey Huibo Bi 1 and Erol Gelenbe 2, * 1 Beijing Key Laboratory of T rafﬁc Engineering, College of Metropolitan T ransportation, Beijing University of T echnology , Beijing 100124, China; huibobi@bjut.edu.cn 2 Intelligent Systems and Networks Group, Department of Electrical and Electr onic Engineering, Imperial College London, London SW7 2BT , UK; e.gelenbe@imperial.ac.uk * Correspondence: E-Mail: e.gelenbe@imperial.ac.uk; T el.: +44-207-594-6342; Fax: +44-207-594-6274. Academic Editor: name V ersion July 10, 2019 submitted to Electronics Abstract: Owing to the incr easing frequency and destr uction of natural and manmade disasters to modern highly-populated societies, emer gency management, which pr ovides solutions to prevent or address disasters, have drawn considerable resear ch over the last few decades and become a multidisciplinary area. Because of its open and inclusive nature, new technologies always tend to inﬂuence, change or even revolutionise this r esearch ar ea. Hence, it is imperative to consolidate the state-of-the-art studies and knowledge to meet the r esear ch needs and identify the futur e resear ch directions. The paper pr esents a compr ehensive and systemic review of the existing resear ch in the ﬁeld of emer gency management fr om both the system design aspect and algorithm engineering aspect. W e begin with the history and evolution of the emer gency management r esear ch. Then the two main resear ch topics of this ar ea, “emergency navigation” and “emer gency search and rescue planning”, are intr oduced and discussed. Finally , we suggest the emerging challenges and opportunities from system optimisation, evacuee behaviour modelling and optimisation, computing patterns, data analysis, energy and cyber security aspects. Keywords: Emergency Management; Emergency Navigation; Emergency Search and Rescue Planning; 1. Introduction Emergency management is the creation of plans and strategies for emergency personals and evacuees to cope with disasters and r educe vulnerability to hazards [ 1 ]. Generally , it can be divided into four aspects, prevention, prepar edness, response and recovery (PPRR), which are originated fr om the work of US State Governors’ Association in 1978 [ 2 ] and have now been widely accepted as the fundamental components to deal with natural or manmade disasters. The attempts of managing disasters have been deeply rooted in human history and even folklor e owing to the impacts of disasters to social and economic aspects of societies. Many great disasters have been recor ded in ancient literature such the Bible story of the Deluge. In the ancient times, the operations of emergency management were mostly conducted in a unorganised, reactive manner . Until the 20 century , laws or policies wer e begun to be passed worldwide to provide ﬁnancial assistance and investment after or before a disaster strike [ 3 ], transforming the operations of emergency management to a more organised, pr oactive fashion. From the 1950s, with the nuclear thr eat during the Cold W ar as well as the development of computer technologies, many efforts have been dedicated to computer-aided civil defense programs [ 4 , 5 ], which motivated the development of the subsequent systematic emergency management studies. The current research ef forts in this area can be generally classiﬁed into two Submitted to Electronics , pages 1 – 33 www .mdpi.com/journal/electronics V ersion July 10, 2019 submitted to Electronics 2 of 33 directions: emergency navigation and emer gency search and r escue planning. Emergency navigation, which is also known as emergency evacuation planning, is the process of directing evacuees out of hazardous areas with the aid of real-time r outing algorithms or pr e-deployed static plans that are based on the pr ediction and analysis from evacuee behaviour models. Emergency sear ch and r escue planning, on the other hand, focuses on rescuing immobilised and incapacitated evacuees that are trapped in hazardous ar eas with the assistance of task assignment or r esource allocation algorithms. This paper reviews the development and applications of the two main r esearch dir ections, from both the system design aspect and algorithm engineering aspect. Owing to its open and inclusive nature, new technologies always tend to inﬂuence, change or even r evolutionise the r esear ch area of emer gency management. Hence, the current and emer ging challenges are also discussed. The remainder of the paper is or ganised as follows. W e ﬁrst summarise the research progr ess in the ﬁeld of emergency navigation in Section 2 , from the system design aspect in Subsection 2.1 and algorithm engineering aspect in Subsection 2.2 ; then we review the resear ch efforts related to emergency sear ch and r escue planning in Section 3 , including various systems in Subsection 3.1 and associated algorithms in Subsection 3.2 ; next, the resear ch trends and challenges are discussed in Section 4 ; ﬁnally , we draw conclusions in Section 5 . 2. Emergency Navigation Nowadays, the resear ch of emergency navigation aims to direct evacuees out of hazardous ar eas in a timely and safe manner with the assistance of computer -aided systems. In this section, we ﬁrst review a series of emer gency navigation systems chronologically in Subsection 2.1 , then we summarise various emergency navigation algorithms topic by topic in Subsection 2.2 . 2.1. Emergency Navigation Systems The study of emergency evacuation in conﬁned spaces, which was initially motivated by defence applications [ 4 , 5 ], has attracted much attention owing to the potential of losses in terms of human lives and property during a disaster . Accompanying the advancement of computer technologies, emergency evacuation systems have experienced a few stages: from the original human experience driven systems, to the currently booming in-situ wireless sensor network based navigation systems, towards the cloud based navigation systems which ar e still in their infancy . Due to limitations in processing power , early emergency navigation systems are commonly computer-aided information reporting systems to assist emergency managers in making decisions [ 6 ]. Associated emergency navigation algorithms at that time normally used human experience or purely mathematical models to simplify an evacuation process and seek optimal solutions. Before 1990s, resear ch in this area is very limited. The resear ch in [ 7 ] considers the evacuation planning problem as a minimum cost network ﬂow problem that converts the original building graph to a time-expanded network; by solving the time-expanded network via a linear programming algorithm, evacuees can obtain optimal r outes and achieve shortest evacuation time. The study in [ 8 ] designs a graph processing software to represent a underground mine as a graph-based evacuation network in conformity to proper ventilation requirements; each edge is assigned with a weight in terms of its distance to the source of hazard and fresh air intake, and Dijkstra’s shortest path algorithm is utilised to ﬁnd the safest paths for evacuees. The work in [ 9 ] pr oposes a traf ﬁc monitoring and analysis system to pr edict the possible trafﬁc jams for emergency planners during urban-scale evacuations; real-time traf ﬁc data ar e collected at r oadside trafﬁc counting stations and transmitted to the system via conventional telephone lines; an evacuation simulation model is used to provide the locations and timing of occurrence of potential trafﬁc bottlenecks. The resear ch in [ 10 ] presents a human experience driven emergency alarm system to facilitate emergency authorities to evacuate r esidents before the landfall of a hurricane; a “vertical evacuation” methodology is proposed to lower the evacuation time and the issuing of “early evacuation orders” is believed to be critical in reducing fatalities. T o improve the disaster r esponse ability in accidents at nuclear power plants, the study in [ 11 ] proposes V ersion July 10, 2019 submitted to Electronics 3 of 33  Emergency Navig ation Emergency Navig ationSyst ems HumanExperience DrivenS ystems Static WSNbased Systems MixedWSNbased Systems Cloudbased SystemswithWSN Cloudbased Systems with MobilePhones Emergency Navig ation Algorithms Off‐lineNa vigation Algorithms CellularAutomat a Models SocialForce Models Fluid‐dynamics Models LatticeGas Models GameTheoretic Models ComputerAgen ts basedModels AnimalAgent based Models On‐lineEmergency Navig ation Algorithms NetworkFlo w basedAlgorithms Geometric Algorithms QueueingModel basedAlgorithms Poten tial‐ Maintenance Algorithms BiologicallyInspired Algorithms RoutingPr otocol basedAlgorithms Prediction‐based Algorithms Figure 1. A tree diagram explaining the structure of Section 2 . an evacuation plan to ﬂee residents within 20 miles of the plant when a radiation leakage occurs. The work in [ 12 ] designs a real-time emer gency monitoring and response system for a nuclear power plant; the response decisions ar e based on discussions between experts at off-site emergency response centers. W ith the fast development of information and communications technology (ICT) as well as the advent of low-costing microelectronic devices, in the middle of 1990s, resear ch moved to the development of complex Emergency Cyber -Physical-Human systems to direct evacuees to exits with the aid of an on-site wir eless sensor network (WSN). Until today , most of the state-of-the-art emer gency navigation systems and algorithms are still based on static WSNs. For instance, the resear ch in [ 13 ] presents a static WSN based distributed system to compute shortest safe paths for evacuees; this system employs a two-tiered ar chitecture that contains a sensing sub-system and a decision sub-system. The study in [ 14 ] proposes a static WSN based navigation system to direct evacuees with a variant of the temporally ordered routing algorithm [ 15 ]; initially , exit sensors broadcast “initial packets” to V ersion July 10, 2019 submitted to Electronics 4 of 33 assign each sensor with an “altitude” that is positively correlated with its distance to the nearest exit (sensors that ar e nearer to the exits possess smaller altitudes than those farther sensors); when a disaster occurs, the altitude of sensors inside a hazar dous r egion will be increased and escape paths will be generated along sensors with higher altitudes to those with lower altitudes. The work in [ 16 ] utilises a self-organising WSN to guide a robot acr oss a hazar dous ar ea; by using an “artiﬁcial potential ﬁeld” [ 17 ], sensors can cooperatively generate a safe path without knowing the network topology . Compared with static WSNs, mixed WSNs which contain mobile nodes can monitor uncover ed areas of static sensors and is less prone to failures in harsh hazardous environments. Hence, some resear ch has employed mixed WSNs to build emer gency response systems. For instance, the studies in [ 18 – 20 ] have proposed a resilient emergency support system (ESS) with the aid of opportunistic communications [ 21 ]. This system consists of sensor nodes (SNs) and communication nodes (CNs). SNs can detect the hazard in its vicinity and inform the evacuees passing by of the location, while CNs are portable devices that are taken by occupants. The work in [ 22 ] presents an indoor autonomous navigation system composed of an intelligent evacuation sub-system (IES) for primary use and a opportunistic communication based evacuation sub-system (OCES) for backup purposes; both sub-systems are supported by pre-installed sensors in the building, the IES utilises static decision nodes to guide evacuees in proximity while the OCES employs mobile decision nodes carried by civilians to disseminate emergency messages and direct evacuees when the IDE malfunctions. The experimental results show that the use of the OCES can considerably r educe the number of fatally injured civilians during an evacuation process. W ith the increasing ubiquitousness of smart phones which provide powerful sensing ability and suf fer less fr om battery power limitations, many studies have integrated evacuees’ portable devices into emergency navigation systems. For instance, the study in [ 23 ] proposes an emer gency support framework that integrates a pre-deployed WSN with the existing mobile network infrastructur e to guide evacuees out of a built environment; the framework, namely “CoW iSMoN”, employs both ﬁxed sensors pre-installed in the building and mobile phones carried by evacuees to collect information and send to a quick rescue r esponse center via short-range wireless communication links or the mobile cellular network; moreover , a cognitive communication protocol that optimises both the network layer and data link layer is designed to ease the network congestion caused by the transmission of large volumes of sensory data and the degradation of communication quality during disasters. Similarly , the resear ch in [ 24 ] presents an indoor emer gency evacuation system composed of a sensor-data management sub-system and an indoor navigation sub-system; the sensor-data management sub-system gathers sensory information and can alert users and the building manager via mobile phones; the indoor navigation sub-system utilises radio beacon devices to estimate users’ position; each user carries a beacon r eceiver that can receive signals from beacons and transmit to the mobile phone via Bluetooth. The major drawback of the WSN based emergency navigation systems is the limited computing capacity , which does not allow them to compute optimal evacuation plans in a timely manner so as to forward this information to evacuees in the pr esence of time-varying hazar ds. Hence, some emer gency navigation systems have integrated cloud computing technologies that are accessible via on-site WSN, to ofﬂoad intensive computations to r emote cloud servers. For instance, the resear ch in [ 25 ] proposes a hazard surveillance system to detect unusual events in an environment and alert residents; the system is composed of a number of static sensors, several mobile sensors and an external cloud server; when static sensors detect unusually high temperatur es, mobile sensors will be dispatched by the cloud server to take snapshots and upload to the server for further analysis; if a ﬁre emergency is conﬁrmed, the cloud server will notify the residents in the vicinity to evacuate. the study in [ 26 ] proposes a multi-cloud based evacuation system that integrates an on-site WSN and remote cloud servers to calculate evacuation paths for users; when a disaster occurs, the system launches an instance for each user to compute the desir ed evacuation r oute; owing to the limited I/O capability of cloud providers, several cloud platforms are employed to launch sufﬁcient instances for evacuees and a V ersion July 10, 2019 submitted to Electronics 5 of 33 dynamic programming algorithm is used to minimise the overall latency and service maintenance cost of the system. Although the hybrid emergency r esponse systems that integrate on-site WSNs and off-site cloud servers can avoid the problems caused by the limited computing power of WSNs, the disadvantages of limited battery life-time of the WSNs, as well as the high likelihood of system malfunction during an emergency still r emain. Hence, many of the studies have replaced WSNs with smart phones carried by evacuees to build mor e ﬂexible systems. For example, the work in [ 27 ] pr esents a building ﬁre evacuation system that consists of radio fr equency identiﬁcation (RFID) sensor tags, mobile phones with RFID r eader and a back-end cloud server; since signals fr om the global positioning system are unavailable inside built environments, RFID sensor tags are used to record the temperature and location information; when a ﬁr e br eaks out, mobile phones carried by evacuees will periodically sense the RFID signals and upload to the cloud server; the cloud server will then calculate the shortest safe route for each civilian with r espect to the distance to exit and the summed temperature along the route. The resear ch in [ 28 ] proposes a smart cloud evacuation system (SCES) to post emer gency messages and plans to r esidents in a built environment; in the front end, a wireless intelligent sensor network (WISN) that integrates a WSN with smartphones are utilised to collect information; in the back end, a cloud based decision-making system is used to analyse the uploaded multimedia information (such as voice, text, images, etc.) and calculate escape paths with a 3D simulator . The study in [ 29 ] proposes an infrastructur e-less emergency navigation system to guide evacuees out of hazardous buildings with the aid of smart phones carried by evacuees and an off-site cloud based decision support system (CDSS); evacuees can locate themselves by taking a snapshot of pre-dep loyed landmarks (e.g. door signs) in the vicinity and uploading it to the CDSS for location identiﬁcation; the CDSS computes congestion-aware paths with the shortest time to exits for evacuees based on their uploaded locations; evacuees are guided to exits in loose groups with the assistance of a combination of the social potential ﬁelds algorithm and a cognitive packet network based algorithm; to reduce the likelihood for the battery power of smart phones to be drained during the energy-hungry communication among smart phones and the CDSS, a power-aware communication protocol is also presented to balance the r emaining battery power of smart phones by relaying sensory information via more ener gy efﬁcient short-range communication techniques before uploading to the cloud server thr ough 3G. T able 1. Emergency Navigation Systems System type Period Human experience driven systems 1970s - 1990s Static WSN based systems 1995 - pr esent Mixed WSN based systems 2006 - pr esent WSN & Cloud based systems 2007 - pr esent Cloud based systems & mobile phones 2011 - pr esent 2.2. Emergency Navigation Algorithms As the kernel of an emergency navigation system, many studies have concentrated on emergency navigation algorithms, which aim to guide evacuees out of hazardous areas safely and efﬁciently . Previous emergency navigation algorithms can be divided into two types: off-line algorithms and on-line algorithms. Off-line algorithms focus on optimising the design of cr owded sites and evaluating the overall clearance time for all evacuees before a disaster occurs. On the other hand, on-line algorithms aim to provide evacuation paths for evacuees in a real time manner . The literature review for the two categories of algorithms is detailed as follows. V ersion July 10, 2019 submitted to Electronics 6 of 33 2.2.1. Off-line Emer gency Navigation Algorithms Since resear ch has indicated that destr uctive crowd behaviours, such as clogging, pushing and trampling, can lead to serious fatalities [ 30 ], also owing to the absence of r eal evacuation data [ 31 ], off-line emergency navigation algorithms have been dedicated to investigate and design crowd behaviour models [ 32 , 33 ] to simulate the crowd movements in reality and pr event destructive cr owd behaviours from occurring by improving the design of built environments. The cr owd behaviour models of the of f-line emer gency navigation algorithms can be classiﬁed into cellular automata models [ 34 – 38 ], social force models [ 30 , 39 – 41 ], ﬂuid-dynamics models [ 42 , 43 ], lattice gas models [ 44 , 45 , 45 , 46 ], game theoretic models [ 47 – 50 ], computer agents based models [ 51 – 54 ] and animal agent based models [ 55 , 56 ]. Cellular automata models discretise a given structure into uniform “cells” that each cell can hold one person. This approach can pr ecisely model the inﬂuence of an individual’s physical dimensions, but is ineffective in depicting the movement speed and dir ection, due to the discr ete spatial structure [ 57 ]. The physical conditions and the movement patterns of evacuees are normally determined by a set of local rules at each cell (one drawback is that it is relatively difﬁcult to customise the physical attributes of each individual civilian). Since this model can effectively mimic the interactions between the environments and the pedestrians, many studies have utilised this microscopic model to simulate the pedestrian dynamics during evacuations in the last two decades. In these models, evacuees are consider ed as either homogeneous with identical physical attributes (e.g. gender , age, mobility , psychology) or heterogeneous individuals with dif fer ent characteristics. For instance, the resear ch in [ 58 ] utilises a homogeneous cellular automata model to investigate the exit dynamics of evacuees in a room with differ ent number of exits; the arching behaviour , which is a signature of jamming that happens when the exits are overused, is observed near the exits; a “power law behaviour ” is also found: when the exit door can evacuate mor e than one evacuee at a time, the evacuees will escape fr om the room in bursts of various sizes. In [ 37 ] a heterogeneous cellular automata model mimics the evacuation process in a r etirement house; evacuees initially belong to thr ee groups (middle-aged people, nursing staff and older people), and groups are also formed dynamically due to the follow-the-leader effect. In [ 38 ] grouping behaviours in evacuations ar e induced by intr oducing “bosons” into cells of the ﬂoor ﬁeld cellular automaton [ 59 ]; bosons are placed by evacuees as markers to increase the probability for other group members to reach some particular cells. The resulting simulations indicate that the evacuation time decreases with the incr easing numbers of groups. The resear ch in [ 39 ] ﬁrst proposed that the motion of a crowd of pedestrians are subject to “social forces”; in the social for ce model, the motion of a pedestrian is mainly affected by the destination, the repulsive forces fr om other objects (e.g. the pedestrian keeps a certain distance away fr om other pedestrians or obstacles), the attractive forces from other objects (e.g. the pedestrian is attracted by friends or window displays); a “direction dependent weight” is introduced into the model since the objects behind a pedestrian have a weaker ef fect on the pedestrian; a “ﬂuctuation ef fect” is also integrated into the model to simulate the random movement behaviours or deliberate deviations from usual motion r ules. The study in [ 60 ] combines the social force model with a counterpropagation neural network model [ 61 ] to mimic crowed behaviours in panic; the personality of evacuees is expressed as impatient and patient. the velocity and the action of evacuees is determined by the social force model and the neural network model, respectively; the neural network has four inputs: the personality of an evacuee, the deviation between the desired speed and real speed, the space on the left side of the evacuee and the space on the right side of the evacuee; the output of the network is the action of the evacuee: follow the person in front, evade the front person from the left side and evade the front person fr om the right side. The resear ch in [ 40 ] utilises the social force model to investigate the pedestrian evacuation dynamics in a room with an exit; experimental results show that if evacuees move at the low desir ed velocities, the faster the evacuees moves, the faster they will evacuate the room; however , if evacuees move at the high desired velocities, the “faster is slower ” effect, that the faster the evacuees wish to move the slower they can escape from the room, is observed and analysed. V ersion July 10, 2019 submitted to Electronics 7 of 33 Fluid-dynamics models imitate evacuee ﬂows as ﬂuids to study the density and speed adaptation during an evacuation pr ocess. Compared with micr oscopic models, the macr oscopic ﬂuid-dynamics models are better at simulating and analysing the behaviours of large crowds. For instance, the resear ch in [ 62 ] derives several equations that govern the motion of a pedestrian ﬂow from the “continuity equation” of ﬂuid mechanics in physics; the proposed equations lead to two possible regimes of a pedestrian ﬂow: the fast-moving, low-density “supercritical” ﬂow in which disturbances spread within the ﬂow and the slow-moving high density “subcritical” ﬂow in which disturbances are swept along by the ﬂow; several partial differ ential equations, that govern the cr owd behaviours of a ﬂow that contains differ ent types of pedestrian, are also studied; a pedestrian type is determined by the destination, walking speed and per ception; the analysis and experimental r esults show that pedestrians tend to reach each immediate destination in minimum time rather than arriving at all destinations in overall minimum time. The study in [ 63 ] presents a continuum model to investigate the r elation between evacuee density and walking speed during a process of evacuees leaving a corridor thr ough a door; the proposed model is based on the Lighthill-Whitam [ 64 ] and Richards [ 65 ] model that is used to simulate vehicle ﬂows; speciﬁcally , this model describes the decr ease in the outﬂow thr ough a door caused by the panic “over compression” effect of evacuees in fr ont of the door; the analytical r esults indicate that the rise of panic can dramatically decrease the outﬂow of evacuees when the door is narrow . Since a pedestrian ﬂow is a many-body system [ 66 ] that is composed of strongly interacting persons, lattice gas models that consider pedestrians as particles on the square lattices have attracted attention since 1980s. The resear ch in [ 45 ] utilises a lattice gas model to simulate the process of a pedestrian ﬂow evacuating a hall; the hall is represented by square lattices and evacuees are randomly distributed over the lattices; each evacuee can either hold still or move in four dir ections: forward, backward, left and right; evacuees move in the preferential dir ection with no back step and cannot overlap on lattices occupied by other evacuees; differ ent dynamical patterns such as arching, ﬂattening and pitting are observed in computer simulations; experimental r esults indicate that the dynamical phase transition from the choking ﬂow to the decaying ﬂow occurs at a critical time. The study in [ 46 ] employs a lattice gas model to study the evacuation time for a cr owd to escape fr om a hall thr ough a single exit; evacuees are modelled as biased-random walkers and move in pr eferential dir ections; the hall is r epresented by square lattices and each square lattice may contain up to one evacuee at a time; the spatio-temporal distribution of evacuation time of evacuees is derived from simulations; the experimental r esults show that the evacuation time of an evacuee depends highly on its initial position within the hall; the ef fect of the exit width, initial population density and ur gency level ar e also investigated in the experiments. T o explicitly model the behavioral reactions of the individuals during an evacuation process, especially the cooperative and competitive behaviours [ 67 ], many of the studies have utilised game theory to mimic the interactive decision-making and strategy-adapting among evacuees. For instance, the resear ch in [ 48 ] employs the non-cooperative game theory [ 68 ] to mimic evacuees’ exit selection process when an emergency occurs in a building with multiply exits; the pr ocedure of the algorithm consists of two steps; in the ﬁrst step, all evacuees ar e considered as a “whole entity” which aims at minimising overall evacuation time while a “virtual entity” is used to maximise the overall evacuation time by imposing the blockage inﬂuence; hence, a two-player zer o sum game is envisaged between the evacuees and the virtual entity; the optimal strategy is found when a Nash Equilibrium [ 69 ] is achieved via optimising the probabilities for the evacuees to choose each exit and the possibilities for the virtual entity to pick each exit (to generate congestion); in the second step, the decision of each individual evacuee is determined by calibrating the evacuees’ probabilistic choices in terms of evacuees’ distance to exits; this is because, in reality , an evacuee will not pick a farther exit unless the nearer exit is congested. The study in [ 50 ] utilises a game-theoretical model to investigate the competitive and cooperative behaviours during an evacuation process fr om a large single r oom with one exit; During the evacuation process, when N evacuees wish to occupy the same desired position, the conﬂict among V ersion July 10, 2019 submitted to Electronics 8 of 33 evacuees leads to a N ∗ N game; each evacuee can choose to either compete or cooperate: if all the evacuees choose to cooperate, then they will all reach the desired position; if all the evacuees are competitive, they will all be blocked at the initial position; if one evacuee choose to compete and the rest is in a state of cooperation, only the competitive evacuee can r each the desir ed position; the simulation results show that: ( 1 ) with the incr easing urgency of emergency , the cooperation among evacuees decreases; ( 2 ) higher cooperation frequency will r esult in shorter overall evacuation time; ( 3 ) the imitation behaviours among evacuees can enhance the cooperation level but reduce the efﬁciency of the evacuation process. Algorithms based on pure mathematical models have difﬁculty in fully representing and capturing the dynamics of an evacuation process. Hence, the agent-based algorithms, which normally repr esent a hazardous envir onment with a number of autonomous decision-making virtual agents, have drawn considerable attentions in recent years. One major advantage of the agent-based algorithms is the ability to evolve and learn, which can lead to unanticipated behaviours during simulations. This characteristic makes the agent-based algorithms a canonical approach to mimic the counterintuitive emergent phenomena [ 51 ]. For instance, the r esearch in [ 52 ] utilises a multi-agent framework to simulate a metro system in the case of a tunnel ﬁr e; the passengers and metr o personnel, the technological system, as well as the ﬁr e and smoke are simulated by separate agents and co-evolve in an interactive manner; an effective evacuation plan is designed by varying environmental factors, such as the number of passengers on the train, the time cost for the train driver to open the doors, etc.; with the aid of the multi-agent computer simulations which can test different scenarios, the emer gency personnel can quickly customise a rescue plan when a disaster occurs. The study in [ 54 ] presents a prototype multi-agent simulation system that can build a virtual environment with autonomous agents for safe egress analysis; the pr oposed system consists of a geometric engine that r epresent the physical environment with AutoCAD, a population generator that can produce evacuee agents with diverse age, mobility , etc., a global database which maintains all the state information of agents; an events recor der that captures the behaviours of evacuee agents, a visualiser which displays the movement of evacuees, a crowd simulation engine that is assigned to each evacuee agent to manage the individual behaviour in terms of the perception-action appr oach [ 70 ]; each evacuee agent is modelled to makes decisions based on thr ee basic conventions: instinct, experience and bounded rationality [ 71 ]; some emergent behaviours such as competitive, queueing and her ding are observed in the simulation. Owning to the scarcity of human emergent behavioural data and the difﬁculty in conducting genuine emergency evacuation experiments, the studies of emergent behaviours have largely depended on simulations. Hence, in recent years, animals have been used in escape panic experiments to study crowd evacuation. For example, the research in [ 55 ] employs mice to study an indoor evacuation process; mice wer e released into a r ectangular container (simulate a large single r oom) ﬁlled with tap water and wer e left to swim towar ds a dry platform; an exit is placed between the wet and dry ar eas to simulate the door of the single room; the ef fect of exit width and exit number on the mouse escape rate is investigated over dif ferent experimental sessions, which ar e r ecorded with a digital video camera; the experiments demonstrate some well-known behaviours of panicking crowd: when the exit width approximately equals to the size of a mouse, the dif fusive evacuation ﬂow is observed; when the exit width becomes larger , the mice evacuate the exit in bursts of differ ent sizes that yield the power -law distributions depending on the exit width. The work in [ 56 ] employs a species of Cuban leafcutter ants called Atta insularis to investigate the effect of panic-induced herding to an evacuation process fr om a two-exit room; ants are intr oduced into a circular acrylic cell with two exits symmetrically situated at left and right; in the ﬁrst experiment which simulates a normal evacuation process, when the ants are placed into the cell, the two exits ar e opened synchronously so that the ants can escape; in the second experiment which mimics an emergency evacuation pr ocess, the only differ ence from the ﬁrst experiment is to inject a dose of insect-repelling liquid to generate a panic before opening the exits; the experimental results show that ants escape from both exits in appr oximately equal proportions in normal conditions but prefer one of the exits in emergency conditions; the experiments demonstrate the V ersion July 10, 2019 submitted to Electronics 9 of 33 theoretical pr ediction that the her ding behaviour in conﬁned spaces can generate a non-symmetrical use of two identical exit doors; in addition, the observed evacuation dynamics are repr oduced with a computer model inspired by [ 30 ]. 2.2.2. On-line Emergency Navigation Algorithms Contrary to off-line emergency navigation algorithms which aim at optimising the design of crowded sites or generating evacuation plans for facility managers via developing various crowd behaviour models and computer simulations, on-line emer gency navigation concentrates on combining mathematical models [ 72 ] or algorithms [ 73 , 74 ] with underlying sensing, communication and computational devices to guide evacuees out of hazardous envir onments in a real time manner . Since on-line emergency navigation algorithms requir e real time information exchanges with the hazardous envir onment, these algorithms are usually integrated into emer gency navigation systems. W ith the development of emergency navigation systems which are detailed in 2.1 , various emergency navigation algorithms have been pr oposed such as network ﬂow based algorithms [ 7 , 75 – 78 ], geometric algorithms [ 79 , 80 ], queueing model based algorithms [ 81 – 87 ], potential-maintenance algorithms [ 14 , 16 , 88 ], biologically inspired algorithms [ 89 – 91 ], routing pr otocol based algorithms [ 19 , 92 – 94 ] and prediction-based algorithms [ 95 – 99 ]. Network ﬂow based algorithms consider the evacuation planning pr oblem as a minimum cost network ﬂow problem [ 100 , 101 ]. Commonly , this type of algorithm ﬁrst predicts the upper bound of the overall evacuation time and then convert the original building model to a time-expanded network by duplicating the original network for each discrete time unit. After that, linear pr ogramming or heuristic algorithms are utilised to compute the optimal evacuation plan. This type of approach can achieve the optimal solution but normally does not take the spreading of the hazard into consideration. For instance, the work in [ 7 ] utilises a dynamic network optimisation model to minimise the overall evacuation time and prevent “bottlenecks” from occurring in a large building; the building is repr esented by a graph model composed of nodes and arcs; the capacity of a node is determined by dividing the space area of the node by the typical space occupied by an evacuee; the capacity of an arc, which is deﬁned as the maximum number of evacuees that are allowed to traverse the arc per unit time, is determined by the passageway width; the graph model is expanded into a time-expanded network by duplicating the original graph model over T time periods, where T is determined by dividing the approximate evacuation time T e by the length of a time period (10 seconds); to reduce the computational complexity and ensur e the existence of a feasible solution, the minimum feasible building evacuation time T e is determined by the proposed bisection search algorithm; the time-expanded network is solved via a large-scale dynamic transshipment algorithm fr om the GNET program [ 102 ]. Since the search complexity of a time-expanded graph grows exponentially with the increase of the time bound T , the studies in [ 103 , 104 ] develop a polynomial time algorithm to solve the evacuation problem with a ﬁxed number of sour ces and exits; the evacuation problem is converted to a quickest ﬂow pr oblem, which aims to send a speciﬁc amount of ﬂows fr om sour ces to sinks in the shortest time; the building model is r epresented by a graph with integral transit times and capacities on the edges; the evacuees ﬂows are r epresented by the temporally r epeated ﬂows pr oposed in [ 100 ] rather than static ﬂows in a time-expanded network; the quickest ﬂow problem with multiple sources and sinks is then transferred to a lexicographic maximum dynamic ﬂow problem and can be solved by using the algorithms pr esented in [ 105 ] and [ 106 ]. Since linear programming algorithms that utilise time-expanded networks to calculate optimal evacuation plans can suffer fr om high computational cost, the work in [ 77 , 78 ] propose a heuristic-based algorithm called capacity constrained route planner (CCRP) to produce sub-optimal evacuation plans in a time-efﬁcient manner; rather than transforming the original evacuation network into a time-expanded network, CCRP employs the Dijkstra’s shortest path algorithm [ 107 ] to search only the original evacuation network and calculate the quickest routes for evacuees; CCRP ﬁrst sear ches the route with the shortest arrival time from any source node to any destination node in terms of path length, pr evious reservations and possible waiting time; then V ersion July 10, 2019 submitted to Electronics 10 of 33 it allocates evacuees to this r oute with respect to the capacity of the route; the CCRP algorithm will iterate the above two steps until all the evacuees r each the exits. These approaches can theoretically solve optimal routes with the shortest time to exits by avoiding congestion. However , to achieve shortest time to exit, evacuees must accurately follow the suggested paths and r each every node on schedule and may even wait certain time at a node to avoid congestion. This is impractical in a real evacuation process. Mor eover , these approaches suf fer from high computational complexity because the time-expanded network will contain at least ( N + 1 ) T nodes for a graph with N nodes and an upper bound of evacuation time T . In addition, as aforementioned, the spreading of the hazar d is not considered in these appr oaches. Geometric algorithms normally use a graph model to repr esent a hazardous envir onment and take advantage of the unique pr operties of geometric graphs to calculate safe egress paths for evacuees. For instance, the research in [ 79 ] adopts the localized Delaunay T riangulation method [ 108 , 109 ] to partition a wireless sensor network into triangular ar eas and construct ar ea-to-area egr ess paths with the aid of a distributed navigation protocol; each sensor , which is the shared vertex of all the adjacent triangles, maintains the node ID, hops to the exits and the sensed hazard level (temperature) of the neighbour sensors; the direction of an egress path is generated from vertices with larger hop count to the exit to vertices with smaller hop count; the safety level of a triangle area is classiﬁed into three color-coded levels (red “high”, yellow “moderate” and green “low”) by comparing the average detected temperature of the associated sensors with a pr e-set temperature thr eshold; in built environments with multiple-exits, additional wireless access points (AP) that can cover the whole environment are deployed in the vicinity of each exit to count the number of evacuees nearby , a load-dispersion algorithm is employed to distribute evacuees by limiting the number of users per exit. The resear ch in [ 80 ] proposes a WSN based emergency navigation system to guide evacuees without the aid of any pre-knowledge of sensor or user locations; the process of navigating evacuees to the exit contains three stage: ﬁrstly , a r oad map is generated as the backbone r oute; secondly , the exit is connected to the backbone r oute via a virtual power ﬁeld algorithm; thirdly , evacuees are dir ected to the backbone r oute via the virtual power ﬁeld algorithm and then follow the backbone r oute all the way to the exit; the road map is constructed via concatenating the medial axis of the boundary of any two safe ar eas; as is proven in [ 110 ], the medial axes of the safe regions, which can form continuous curve graphs, retain the topological and geometric features of the safe ar eas; in the virtual power ﬁeld algorithm, the virtual power of a point is inversely proportional to its distance from the hazard; the route fr om any point to the backbone r oute will follow the most descending dir ection of the virtual power ﬁeld; owing to the expanding or shrinking of the hazard, the danger ous areas vary during the evacuation process; hence, a local r oad map updating algorithm is pr oposed to rebuild the backbone route of the affected areas instead of reconstructing the entire backbone route when a variation of the dangerous areas is detected. However , the effectiveness of these approaches highly depends on the topology of the deployed wir eless sensor network. The change of the topology will induce redeployment and r e-calibration of these algorithms. Owing to the stochastic, highly transient and nonlinear natur e of an evacuation pr ocess, queueing models have been proven as a useful tool to capture and analyse the dynamics of evacuees. Normally , by treating signiﬁcant locations such as doorways or stair cases as “servers”, queueing model based approaches [ 111 ], which generalise the Markovian models of computer systems [ 112 ], transfer building graphs to a queueing network or a number of isolated “queues” to estimate congestion and evacuation delays. For instance, the process of pedestrians traversing a corridor or stairwell is analysed as a state-dependent process in [ 81 ], a M / G / C / C state-dependent queue model is utilised to estimate the congestion delays at corridors or stairwells and the overall evacuation time of an evacuation process; the pedestrian ﬂows ar e classiﬁed into three categories: uni-directional ﬂow , bi-directional ﬂow and multi-directional ﬂow; the r elationship between the crowd density and the mean walking velocity of evacuees in the three categories of pedestrian ﬂows are derived from [ 113 ]; the capacity of a corridor or stairwell is calculated based on [ 114 ], which indicates that the evacuee ﬂow will cease to move when V ersion July 10, 2019 submitted to Electronics 11 of 33 the population density r eaches 5 evacuees per square meter; the state-dependent service rate of the three categories of pedestrian ﬂows can be calculated in terms of the mean walking speed and the corridor capacity; ﬁnally , the time cost for an evacuee ﬂow to traverse a corridor or a stairwell can be computed by the mean value analysis (MV A) algorithm introduced in [ 115 ]. T o ensur e no corridors will be block during an evacuation process in a built environment, the work in [ 82 ] considers the evacuation planning problem as a service and capacity allocation (SCA) problem and sear ches the smallest capacity of each corridor via modelling the building as a M / G / c / c queueing network; the M / G / c / c queueing network is employed to calculate the average queue length at each corridor with the following steps: ( 1 ) the average walking speed V n of n evacuees in a corridor is calculated by the equations derived from the congestion model pr oposed in [ 116 ], ( 2 ) the state-dependent service rate f ( n ) with n evacuees in a corridor can be computed by f ( n ) = V n V 1 , where V 1 is the average speed of a lone evacuee, ( 3 ) term p n , which is the probability of n evacuees in a corridor can be calculated by the equations derived from [ 117 ], ( 4 ) the average queue length of a corridor can then be computed by L = ∑ c n = 1 n p n ; to analyse the smallest capacity of each individual corridor , the generalised expansion method [ 118 , 119 ] is used to expand the M / G / c / c queueing network into an equivalent Jackson network via adding an artiﬁcial holding node in front of each ﬁnite queue to register the blocked evacuees due to capacity limitation; After decomposing the queueing network, a local search algorithm inspired by [ 120 ] is used to sear ch the smallest feasible capacity of each queue. Similarly , the studies in [ 84 , 121 ] utilise a M / G / c / c queue model to simulate the dynamics and predict the overall evacuation time of an egress pr ocess without hazar d; rooms, corridors and stairways are modelled as queues in which the service rate depends on the evacuee density; doors, exits and gateways are imitated as queues in which the service rate depends on not only the evacuee density , but also the faster-is-slower ef fect [ 122 ] and the cr owd impatience [ 123 ]; to validate the effectiveness of the queue system, a discrete-event simulation model is implemented via the SimEvents toolbox in the MA TLAB/Simulink environment and experimental r esults show that the egr ess time of evacuees in simulations highly matches with the prediction of the pr oposed queueing model. Rather than simulating all the building components as M / G / c / c queues, the r esearch in [ 124 ] models doorways that can pass one person at a time as M / M / 1 queues; on the other hand, corridors or stairs ar e modelled as M / G / ∞ queues, in which the inﬁnite number of servers imply that no congestion occurs in corridors or stairs. Rather than using traditional closed network models which suffer from high computational costs, the study in [ 85 ] proposes a computationally efﬁcient open network model with product form to predict the congestion level at each point of interest (PoI) and the overall evacuation time with respect to average arrival and departure rates at each observation point; By assuming Poisson arrivals of evacuees at each originating location, uni-directional corridors that allow at most one evacuee to pass at a time and exponentially traversal delays at each corridor , a M / M / 1 queue model is established to mimic each corridor; hence, the average delay at a corridor can be calculated by 1 µ − λ , wher e 1 µ repr esents the average traversal time of a corridor and λ repr esents the average arrival rate of evacuees at a corridor; the average traversal time of a path can be calculated by summing the average delay of each corridor on it. Rather than considering each signiﬁcant location (such as a doorway or staircase) as an independent “queue” and then use either the limiting probabilities for the number of customers in an M/G/C/C state-dependent queueing model [ 125 ] or steady-state solutions [ 85 ] to analyse the number of evacuees at the location, the work in [ 86 ] treats all the signiﬁcant locations in the designated ar ea as a “queueing network” by considering the interaction effects of various evacuees among linked “queues”; in this study , to predict the time cost T for an evacuee to traverse a path, a G-network model [ 126 ] is employed to periodically compute the utilisation rate of each node and edge under the combined impact of a speciﬁc r outing scheme and panic behaviours; Little’s formula is then used to calculate the average delay of each node and edge; ﬁnally , T can be calculated by summing the average latency of each node and edge on it. The resear ch in [ 87 ] proposes an urban scale emer gency navigation system to guide vehicles to safe zones in the aftermath of a disaster in a latency and energy efﬁcient manner; a G-network model [ 127 ] is utilised to analyse and capture the dynamics of vehicles under the joint inﬂuence of interactions V ersion July 10, 2019 submitted to Electronics 12 of 33 among individual vehicles and the r e-routing decisions from the navigation system; by using this G-network model, the average number of vehicles and the average traversal time at each intersection or road segment can be calculated; hence, the total average delay experienced by a vehicle and the total fuel consumption in the network can be described by a goal function; ﬁnally , a gradient descent algorithm is utilised to r educe the time and fuel cost (minimise the goal function) by optimising the probabilistic choices of linked r oad segments at each intersection. Potential based algorithms normally can dynamically develop navigation paths by assigning attractive or repulsive potentials to the exits and hazards, and the evacuees move as a result of the net attraction-repulsion in various dir ections. For instance, the resear ch in [ 16 ] presents a self-or ganizing sensor network to guide users such as r obots, evacuees or unmanned vehicles out of a hazardous environment along safest paths by using the “artiﬁcial potential ﬁelds” algorithm [ 17 ]: when a sensor detects hazar d, it will broadcast emergent messages including sensor ID, number of hops to the arrived sensor ( N h ) to other sensors; when a sensor receives multiple emergent messages from the same hazardous sensor , it will keep the smallest N h ; the potential value of a sensor generated by a hazardous sensor is calculated by 1 N 2 h ; hence, the overall potential value of the sensor is computed by summing the potential value generated by each hazardous sensor; in this way , an attractive force is generated by the destination to pull the user while r epulsive forces are generated by the dangerous zones to push the user away from them; the safest path is generated by following the most descending dir ection of the potential ﬁeld; experiments are conducted on a testbed with 50 Mote MOT300 sensors [ 128 ] and the results indicate that the algorithm could successfully direct the objects to the destination; however , multiple destinations may have a negative impact on the efﬁciency of r eaching the exits as the users move under the actuation of artiﬁcial forces; Moreover , the convergence time for network stabilization is relatively long due to the effect of data loss, asymmetric connection and network congestion. The study in [ 14 ] proposes a temporally order ed routing algorithm [ 15 ] based multi-path routing protocol to route evacuees to exits through safest paths; a navigation map is manually deﬁned during the deployment process to avoid impractical paths; in the initialisation phase, each sensor is assigned with an altitude with respect to its hops to the near est exit: sensors nearer to the exits are assigned with smaller altitudes while sensor farther from the exits ar e allocated with larger altitudes; when an emergency event is detected, a sensor s i within the hazardous regions will update their altitudes by A 0 ( s i ) = m a x  A ( s i ) , A em g × 1 h 2 s i , s h + h s i , s e  , where A 0 ( s i ) and A ( s i ) repr esent the altitude of sensor s i before and after update; term A em g is a large constant, term h s i , s h repr esents the shortest hop distance between sensor s i and the hazar dous sensor s h while term h s i , s e repr esents the shortest hop distance between sensor s i and the exit s e ; a hazar dous r egion is constructed by sensors within a pr edeﬁned hop distance from the hazardous sensor; egr ess routes ar e generated from sensors with higher altitudes to senors with lower altitudes; ther efore, the update of altitudes of hazardous sensors can ensur e evacuees bypass the hazardous regions. The work in [ 129 ] extends the algorithm in [ 14 ] to a 3D envir onment and divides the sensors into normal sensors, exit sensors and stair sensors in terms of location; if no available path to exits can be discover ed, evacuees will be directed to rooftops and wait for rescue. However , multiple destinations (exits) may affect the efﬁciency of reaching the exits as the users move under the actuation of artiﬁcial forces. Moreover , the convergence time for network stabilization is relatively long due to the effect of the information synchronization, asymmetric connection and network congestion. Millions of years of evolution has made the animal foraging behaviours become near -optimal solutions of autonomous sear ch and path-ﬁnding [ 130 ], biologically inspired approaches, which ar e inspired by simple but reliable natural mechanisms, employ heuristics to sear ch optimal routes in a computationally efﬁcient manner . For instance, a feed-forward neural network model is adopted to a wireless sensor-actuator network (WSAN) for evacuation r outing in [ 90 ]; all physical nodes in the WSAN deploy a neural network with identical topology: an input layer , a hidden layer and an output layer; the input layer r eceives the latest two coor dinates of a pedestrian and a suggested direction is subsequently generated by the output layer; the neural networks are trained with a back-propagation V ersion July 10, 2019 submitted to Electronics 13 of 33 algorithm [ 131 ] in standard situations and are deactivated when an emergency happens; hence evacuees will be directed to exits over their normal walking paths; however , back-propagation algorithms suffer from slow learning rate and easily converging to local minimum; furthermore, this model cannot r eact to the spr eading of a hazard. The study in [ 91 ] employs a genetic algorithm [ 132 ] to minimise the total evacuation time, travel distance and number of congestion encountered during an evacuation process; non-domination sorting [ 133 ] is used as no priori knowledge is available to determine the weight of the thr ee goals; the initial “chromosomes” are paths found by the k -th shortest path algorithm [ 134 ] and ar e incrementally evolved to feasible solutions thr ough crossover and mutation with r espect to the path length, congestion level and hazar d intensity; as an evolutionary approach, this algorithm has advantages in solving multi-objective optimization problem (MOP) [ 135 ]; however , the computational overhead is relatively high due to the path-ﬁnding and the evolution process. The research in [ 136 ] adopts a variation of particle swarm optimization (PSO) to search routes and adjust velocity during evacuations; occupants ar e viewed as particles to search exits; once an exit is discovered, all the other particles will move towards it while keep their moving inertia to expand searching space; if more than one exit is found, particles will choose the nearest exit as the destination; nevertheless, use occupants directly to explore paths may cause sever e injuries and fatalities; meanwhile, this algorithm may induce seriously congestion and oscillation pr oblems. Inspired by the bee colony foraging behaviour , the work in [ 137 ] uses bee colony optimization [ 138 ] to displace evacuees in hazardous areas to safe areas during an emergency evacuation; hives, food sources and bees represent safe ar eas, hazar dous areas and evacuees, respectively; evacuees select a safe area with regard to “attractiveness” which is determined by the distance to the area and the distribution of people in hazardous areas; once an evacuee determines a target, it will recruit other evacuees by sharing information of the devoted area; this algorithm obtains a robust evacuation plan at the expense of relatively high communication over head. Since many of the curr ent emergency response systems ar e based on wir eless sensor networks, routing pr otocols that are initially used for packet networks have been borr owed or adapted to direct evacuees and improve communication quality in hazar dous envir onments. For instance, the research in [ 92 ] presents an emergency support system built on top of a WSN to guide evacuees out of a conﬁned space in a r eal time fashion; the embedded emer gency navigation algorithm is inspir ed by the Cognitive Packet Network routing pr otocol [ 139 , 140 ], which was initially designed for large-scale packet networks; differ ent fr om the original CPN that contains thr ee types of packets: smart packets (SPs), dumb packets (DPs) and acknowledgements (ACKs), the variant only consists of SPs and ACKs; SPs are sent from each sensor nodes in the WSN to search egress paths and collect hazard information in a distribute manner with their predeﬁned goals; when a SP reaches an exit, which means an egress path has been discovered, an ACK will be generated and bring back the ID and the hazard information of each sensor node along the path to the source node that emits the SP; when the ACK reaches the source node, it will update the QoS level of the discovered path by using a rolling average mechanism, which sums the newly discovered QoS and previously stored value in a weighted manner; in attempting to efﬁciently ﬁnd the route with the best quality of service (QoS), when a SP arrives at a sensor node, it will decide its next hop by m-Sensible r outing algorithm [ 141 ]; a QoS metric is deﬁned as sensitive if its value is affected by the trafﬁc through the path, such as congestion level; on the other hand, a QoS metric is insensitive if its value is independent on the trafﬁc assigned to the path, such as path length or number of hops on a path; m-Sensible routing algorithm calculates the probabilistic choices of all the neighbour nodes based on the QoS information brought back by the previous SPs; hence, future SPs decide their next hop by yielding the probabilistic choices of the neighbour nodes obtained by the m-sensible policy; the probability of choosing a neighbour node is determined by E [ L ( n π ( u ) , n π ( j ) , n π ( e ) , π ) ] − m ∑ N n i = 1 E [ L ( n π ( u ) , n π ( i ) , n π ( e ) , π ) ] − m , where L ( n π ( u ) , n π ( j ) , n π ( e ) , π ) repr esents the effective length of a path π from source node n π ( u ) via a neighbour node n π ( j ) to the exit node n π ( e ) ; term E [ L ( n π ( u ) , n π ( j ) , n π ( e ) , π ) ] repr esents the expectation of L ( n π ( u ) , n π ( j ) , n π ( e ) , π ) ; term N n repr esents the number of neighbour nodes of the node n π ( u ) ; it is proved in [ 141 ] that the QoS increases V ersion July 10, 2019 submitted to Electronics 14 of 33 with the increase of term m , hence, an m + 1-sensible r outing policy provides better QoS on the average than the m-sensible policy; to enhance the stabilisation of network, a predeﬁned “measur ement discard threshold” is set to discard the reported QoS (ef fective length) that is smaller than the threshold. Since communications which ar e essential in an evacuation pr ocess can easily malfunction due to the hazard, the resear ch in [ 19 ] proposes a resilient emer gency support system (ESS) to disseminate emergency messages among evacuees and dir ect evacuees out of a conﬁned space with the aid of opportunistic communications (Oppcomms); the proposed system is composed of pre-deployed sensor nodes (SNs) to collect environmental information and mobile communication nodes (CNs) which are portable devices carried by evacuees; to locate evacuees, each SN contains a location tag and can periodically send a location message (LM) to CNs in proximity; when a SN detects a hazar d, an emergency message (EM) will be generated and br oadcast to CNs carried by evacuees in vicinity by using the epidemic routing [ 142 ]; the EM will be stored in these CNs and forwarded to other CNs in contact by the “store-carry-forwar d” paradigm [ 21 ] during the movement of the evacuees; to guide evacuees, each CN stores the building graph in its local storage and updates the edge costs when r eceiving an EM; the Dijkstra’s shortest path algorithm is trigger ed to calculate the shortest safest path when the graph is updated; experimental results indicate that the pr oposed system is robust to network failures during an emergency; since Oppcomms are susceptible to malicious attacks such as ﬂooding or denial of service, an extended study [ 143 ] pr oposes a defence mechanism that uses a combination of identity-based signatures (IBS) and content-based message veriﬁcation to detect malicious nodes. However , network routing protocol based algorithms normally make decisions based on the collected sensory information rather than the predicted situation of a path. Ther efore, when evacuees traverse that path, the situation could have changed owing to the highly dynamic nature of an evacuation pr ocess, which normally induces a delayed feedback loop between living sensory data and r outing decisions. Similar to [ 92 ], the resear ch in [ 93 ] also borrows the concept of the cognitive packet network to calculate evacuation paths for evacuees with the aid of an on-site WSN in a distributed manner; however , rather than using m-Sensible routing algorithm, the random neural network (RNN) algorithm [ 144 ] is utilised as the decision-making algorithm for the SPs to explore the environment; each sensor node in the WSN is considered as a CPN node, in which a RNN is deployed to direct the passing-by SPs and a mailbox is used to store the discovered paths and the associated QoS measurements; the RNN consists of neurons that ar e associated with each potential forwarding dir ection of SPs; each neuron possesses an excitation probability to indicate the quality of the forwar ding dir ection, and the neuron with the highest excitation probability corr esponds to the optimal forwar ding direction; when an evacuation process begins, CPN nodes continuously sends out SPs or r elays SPs fr om other CPN nodes; when a SP reaches a CPN node, it can either select the forwarding direction corresponding to the most excited neuron or drift randomly to search new routes; as a SP arrives at an exit, an ACK will be generated to backtrack the discovered route in a loop-free manner; when an ACK reaches a CPN node, the training process of the local RNN will be triggered and the excitation probability of each neur on will be updated based on the learning mechanism of the RNN; the discovered routes will be stor ed in the local mailbox and sorted by quality; evacuees in the vicinity of a senor node always will be transferred the top-ranked r oute as their evacuation r oute; since each SP can gain “experience” fr om previous SPs, the CPN can rapidly discover the optimal or near-optimal evacuation routes by emitting very few packets [ 145 ]. The studies in [ 94 , 146 ] extends the work in [ 93 ] and makes use of the feature of CPN to develop a multi-path routing algorithm for differ ent categories of evacuees (prime-aged people, aged people, children or ill people, and disabled people in electric powered wheelchairs) with r espect to their speciﬁc r equirements; since each SP can search a distinct path based on its pr e-deﬁned goal function, during the evacuation process, various types of SPs are sent out to search distance-orient paths, time-orient paths, safety-oriented paths and energy-efﬁciency oriented paths for the associated evacuees. On top of the work in [ 146 ], The work in [ 147 ] designs a cooperative strategy that divides evacuees into health-oriented evacuees and evacuation-time-oriented evacuees, and can adjust the routing strategy of evacuees when their “virtual health value” fulﬁlls a certain condition; the use of V ersion July 10, 2019 submitted to Electronics 15 of 33 the strategy is proven to be more sensitive and adaptive to sudden changes in the hazard environment such as abrupt congestion or injury of civilians. By inferring the spreading rate and direction of the hazar ds, prediction-based algorithms predict the future status in the hazardous ar eas and reduce the fatality rate by avoid evacuees from traversing paths with high potential risk level. For instance, the resear ch in [ 95 ] proposes a Monte-Carlo stochastic model to predict the spreading of the ﬁr e hazard and the movement of evacuees during an evacuation process; the tar geted building is represented by two graph models composed of nodes (compartments) and edges (passageways), one for ﬁre spr ead modelling and the other for occupant egr ess modelling; a discr ete hazard function based on Bernoulli trials is used to mimic the propagation of the ﬁre; a Bernoulli trial, which is a random experiment with two possible results: “success” and “failure”, and the probability of success or failure is constant whenever the experiment is conducted, is performed at each time step to mimic the transmission of ﬁr e from a compartment to another; each edge of the two graph models is associated with a “defective” random variable to represent the time cost for evacuees or the ﬁre hazard to traverse this edge; these defective random variables, which take the value “inﬁnity” with non-zero probability , are used to simulate the phenomena such as evacuees cannot reach the next node owing to the capacity limitation or ﬁr e cannot r each the next node due to ﬁre ﬁghting activities. The study in [ 96 ] proposes a WSN based distributed navigation algorithm to search safest r outes for evacuees by maximising the time an evacuee will r emain ahead of the hazard while traversing the route; each sensor will maintain two weighted graphs of the built environment, a “hazard graph” and a “navigation graph”; in the hazard graph, nodes repr esent the locations of sensor while edges repr esent the possible movement dir ections of the hazar d (for example, hazard may spread thr ough walls or along corridors); the weight of an edge is the shortest time for the hazar d to propagate along the edge, these information can be obtained from of f-line hazard simulations [ 148 ] or estimated by emer gency engineers; in the navigation graph, nodes r epresent the sensor locations while edges repr esent the possible movement directions of evacuees; the weight of an edge is the longest time for an evacuee to traverse the edge; when a ﬁre breaks out, sensors that detect the hazard will broadcast the ﬁr e source location over the senor network, then each sensor calculates the safest path to exit by maximising the overall dif fer ence in time between an evacuee arriving each node on a path and the hazar d r eaching these nodes. Since it is dif ﬁcult for ﬁre-ﬁghters to be aware of the actual conditions in a built envir onment during a ﬁre disaster , the resear ch in [ 97 ] presents a novel e-infrastructur e to infer the spreading of hazar d based on predictive models and living sensory data in a faster-than-real time manner; the system consists of on-site sensors including smoke detectors and temperature sensors, and off-site computational models that are deployed on High Performance Computing (HPC) resour ces; gathered sensory data is used as inputs into a Monte-Carlo fashion ﬁre spread model called K-CRISP [ 149 ] to pr ediction the movement of ﬁre and smoke; the results ar e interpreted by using a knowledge-based r easoning scheme within an agent-based command-and-contr ol layer; the outputs are transmitted to ﬁr e-ﬁghters for r eference. T o deal with the highly uncertainties during an evacuation process in an unfamiliar built environment, the work in [ 98 ] proposes a Dynamic Bayesian network (DBN) [ 150 ] based spatio-temporal probabilistic model to captur e the uncertain nature of the hazard and crowd dynamics, and forecast the movement of evacuees; the integrated hazar d and crowd evacuation DBN model is composed of a hazar d model, a risk model, a behaviour model, a ﬂow model and a crowd model; each model embeds a DBN and is subject to the Markov condition; by using the integrated hazard and cr owd evacuation DBN model, the r elations between the location of evacuees and the hazard status of each location (dormant, growing, developed, decaying and burnt-out) are tracked and predicted over adjacent time steps; hence, the pr obabilistic risk level of each location can be derived from the model; the egr ess paths are calculated by Dijkstra’s algorithm with r espect to the estimated risk level of each location with the purpose of minimising the overall fatality rate. These algorithms are developed in r ecent years and ar e quite promising owing to the incr easing popularity and tremendous computing power of the cloud computing paradigm. Since the performance of many emergency navigation algorithms is sensitive to various initial conditions (e.g. the initial distribution V ersion July 10, 2019 submitted to Electronics 16 of 33 of evacuees, congestion level, type of disaster and initial disaster location) and the choice of certain parameters, the resear ch in [ 99 ] presents a faster-than-r eal-time simulation based r outing algorithm to predict the futur e situation befor e guiding evacuees to exits when a disaster breaks out; instead of guiding evacuees in a real time manner , this algorithm borrows the tremendous computational power of the cloud based simulator to rapidly identify the potential death victims by predicting the futur e movements of evacuees and spreading of the hazard; then, an iterative based algorithm is employed to gradually search appropriate paths for these potential death victims; ﬁnally , all the calculated paths are sent to evacuees for instruction and evacuees could follow the paths in sour ce routed manner (do not need to switch paths during the evacuation process). 3. Emergency Search and Rescue Planning Originating from maritime sear ch and r escue operations, sear ch and r escue planning in emergency situations has motivated considerable research over the last several decades owing to the unfortunate increasing threat of both manmade and natural disasters. During a disaster-related emergency evacuation, evacuees may become immobilised and incapacitated due to injuries or obstacle contact. Therefor e, to reduce the fatalities, various emergency management systems have been proposed to detect the location of incapacitated evacuees and dispatch rescuers to perform rescue operations. The main challenges of a rescue operation are threefold. The ﬁrst challenge is how to efﬁciently search and locate injured evacuees or other objects in unknown envir onments, especially on how to coordinate the activities among various rescuers. The second is to design an appropriate r escuer assignment algorithm to allocate rescuers to injur ed evacuees in a real time and computationally ef ﬁcient fashion under the highly dynamic hazar dous environment. This is actually a NP-har d assignment pr oblem [ 151 ], which aims to minimise the overall potential cost for the rescue operation. The third challenge is to search desir ed paths for rescuers and victims to fulﬁl their speciﬁc requir ements, which is difﬁcult since the “quality” of a path is affected by the spreading of the hazard, the dynamic congestion level, the movements and behaviours of evacuees and other rescuers on the path. In the following subsections, we will summarise various systems and algorithms that have been proposed to meet the aforementioned challenges. 3.1. Emergency Sear ch and Rescue Planning Systems Based on the facilities used, emergency sear ch and rescue planning systems can be divided into two categories, wearable computing-assisted rescue systems and search & r escue robotic systems. W earable computing-assisted rescue systems concentrate on providing various enhancements for emergency personnel (e.g. ﬁreﬁghters, rescuers) with the aid of wearable devices to increase the efﬁciency of r escue operations and impr ove the safety of emer gency responders. On the other hand, disasters that create harsh environments with extreme temperature, toxic substances or various obstacles have exposed the unsafety and inefﬁciency of the human-centered search and rescue planning systems. These limitations have therefor e inspired the development of the robot-center ed sear ch and rescue planning systems, which employ various mobile r obots to conduct rescue operations. 3.1.1. W earable Computing-assisted Rescue Systems W ith the rising interest in body-centric wireless communications, which has also been standardised as a part of the fourth generation mobile communication systems (4G) [ 152 ], considerable resear ch has been dedicated to develop low cost, light-weight wearable antennas to maintain and improve emer gency communications. The work in [ 153 ] implements an electrically-small wearable antenna, which is integrated into the sleeve of a jacket, to monitor the positions of emergency rescuers; the designed frequency is at around 860 M H z since higher frequencies could suffer fr om shadowing caused by obstacles; the antenna consists of an electromagnetically coupled square patch and a central shorting pin; to reduce the cost and weight of the antenna, various conductive textiles such as mixed cotton-steel threads are employed and tested as the material for both the patches and V ersion July 10, 2019 submitted to Electronics 17 of 33 EmergencySear ch andRescuePlanning EmergencySear ch andRescuePlanning Systems W earable Computing‐assisted Res cu eS yste ms W earableAnt ennas W earableSensor‐ basedSearchand Res cu eS yste ms AnimalAgent Assisted Searchand Res cu esy ste ms SearchandR escue RoboticSy stems Human‐robot Inter action Homogeneous SwarmRob ots Heterog eneous rob ots EmergencySear ch andRescuePlanning Algorithms Searching Algorithms Resource Allocation Algorithms Navig ation Algorithms Figure 2. A tree diagram explaining the structure of Section 3 . the ground plane. The research in [ 154 ] designs a wearable antenna integrated with the existing Cospas-Sarsat, a satellite-based search and rescue system that provides distress alert detection and information distribution services by locating and communicating with activated personal locator beacons, to provide emergency alert and location information for maritime rescue teams. The antenna is embedded into a life vest, and the moisture-absorption characteristics of the textile can af fect the anntenna’s performance. The textile material used must be evaluated for “Moisture regain”, i.e. their moisture-absorption rate, measured by the relative weight increase when kept in a high-humidity environment, and the antenna is placed on a foam substrate with low moisture-absorption rate, a thin carrier foil with an inkjet-printed antenna pattern, and a cover fabric to protect the antenna from wear and tear , and from water inﬁltration. The study in [ 155 ] presents an emergency rescue navigation system to direct ﬁremen to “key corridors” to eliminate ﬁre and congestion caused by evacuees or obstacles generated from hazard. The rescue navigation system is composed of a remote control center to generate instructions for ﬁreﬁghters, an on-site WSN to pr ovide hazar d & location information, as well as 802.15.4 compatible personal digital assistants (PDAs) carried by ﬁreﬁghters to communicate with the contr ol center and the WSN via naive ﬂooding or the opportunistic ﬂooding strategy . T o determine the “key corridors” that affects the efﬁciency of the evacuation most, this problem is ﬁrstly converted to a “maximum ﬂow problem” by connecting each normal node (represent the location of a sensor) with a virtual source and each exit with a virtual sink; the edges that can maximise the amount of ﬂow passing from the source to the sink are then determined by the Max-ﬂow Min-cut theorem ; ﬁnally , a breadth-ﬁrst-sear ch strategy is used to search the key corridors with existing hazard or obstacles fr om the afore-determined edges. The resear ch in [ 156 ] presents a wearable sensor -based search and r escue system, namely “CenW its”, to locate lost or injur ed hikers in wilderness areas with an opportunistic relaying scheme; each hiker carries a wearable sensor with a built-in GPS receiver that can provide position information and an integrated RF transceiver to exchange “witness information“ (movement and location information) with other hikers in vicinity; to maintain the communication between hikers and the rescue center , V ersion July 10, 2019 submitted to Electronics 18 of 33 access points (APs) are deployed at locations of interest that hikers are likely to pass through, and convey the witness information stored in the wearable sensors to the rescue center; due to the limited battery power and memory of wearable sensors, CenW its also provides an adaptive data storage strategy to optimise the trade-off between battery power and memory utilisation with the aid of a dynamic grouping and partitioning mechanism; if the remaining battery power of wearable sensors is low , hikers are partitioned into groups where only the group leaders stor e all witness information and communicate with the APs while the rest of group members are set to sleep mode; if the r emaining battery power of wearable sensors is suf ﬁcient while the remaining memory is running low , hiker groups ar e further divided into sub-groups wher e each gr oup member stores a subset of the witness information. As an effort to decrease the likelihood of emergency events from the “prevention” aspect, the resear ch in [ 157 ] presents a wearable personal healthcare and emergency aid system, namely “W AITER”, to monitor the health status of users with wearable vital signal sensors and alert the remote healthcare center when an emergency occurs; aiming at substituting the labour-intensive caregiver aid, this system comprises of a body-worn vital signal sensor to collect health status data, a mobile phone to perform on-site computation and storage operations for the raw data and a r emote healthcare center to provide timely medical aid; the vital signal sensor is integrat ed into a Bluetooth ear-set which consists of a heart beat sensor , a motion sensor , a body temperatur e sensor and a Bluetooth wir eless communication device; since the high energy consumption of the wireless data transmission, raw data gathered by the vital signal sensor is ﬁrst transmitted to the mobile phone for r eﬁning and validation via the Bluetooth transceiver; once an emergency is detected, the mobile phone generates an alert with the ﬁltered data and then sends to the healthcar e center via its GSM module. Owing to the superior performance of animal agents during sear ch and rescue operations in terms of mobility , energy utilisation ef ﬁciency , sensory acuity and intrinsic cognitive capacity , much ef fort has been dedicated to animal agent assisted search and rescue systems. The resear ch in [ 158 ] employs cyber-enhanced working dogs to locate and r each survivors trapped under rubble in the aftermath of large-scale disasters; this system contains thr ee components: a smart harness worn by a working dog to monitor the surr ounding environment and the canine, a r emote computer carried by the handler to analyse and control the behaviours of the canine, and mobile base stations such as unmanned vehicles to maintain the communication between the working dog and the handler; the smart harness is equipped with various sensors and actuators, including DC vibration motors to receive haptic commands, a mini-speaker to issue aural commands, a treat dispenser to rewar d the canine for desir ed actions, accelerometers and gyroscopes to monitor the posture of the working dog, as well as a GPS receiver , micr ophones and cameras to gather envir onmental information; on the other hand, the remote computer is employed to identify the posture and behaviours of the canine remotely , the activities of a canine ar e classiﬁed into 5 static postures (sitting, standing, lying down, standing on two legs, and eating off the ground) and 3 dynamic behaviours (walking, climbing stairs, and walking down ramps), three hidden Markov models, which are each associated with one of the dynamic behaviours, are utilised to classify the dynamic behaviours of a working dog; the starting state pr obabilities and transition probabilities of the hidden Markov models are estimated with the iterative Baum-W elch algorithm; the input sensory data that cannot be categorised into the 3 dynamic behaviours will be reﬁned with a moving average ﬁlter and then identiﬁed as a static posture. In attempting to penetrate the voids and narrow gaps under the collapsed ruins after an major earthquake, the study in [ 159 ] employs cockroaches to sear ch trapped survivors and map the under-rubble envir onment; to instruct the locomotory behaviours of a cockr oach, ﬁne wire electrodes are implanted into the antennae of the cockroach to perform neurostimulation-based locomotion control; the sear ching strategy of this system is to keep each cockroach moving and exploring naturally within a deﬁned area; when the cockroach r eaches the boundary of the area, left-turn or right-turn commands will be sent fr om in the form of neurostimulation pulses via the ﬁne wire electrodes; each cockroach is also mounted with three dir ectional microphones to locate the survivors by using the “r eceived signal strength indicator ” V ersion July 10, 2019 submitted to Electronics 19 of 33 and the ultrasonic ranging technologies; to construct a r obustness map of the environment, a novel topological mapping approach is used to ﬁrst extract topological features from encounters among the cockroaches, and then r eﬁne the persistent features fr om the generated encounter map. 3.1.2. Search and Rescue Robotic Systems Since the human access to the victims in the aftermath of a disaster such as an earthquake or radioactive leakage can be time consuming and may induce further casualties [ 160 ], the use of robots that can be r eleased rapidly to locate and rescue victims has drawn considerable attention in r ecent decades, and has gradually become a research domain after the Great Hanshin earthquake and the Oklahoma City bombing in 1995 [ 161 ]. Nowadays, resear ch on “sear ch and rescue robotic systems” has evolved as a major branch of “sear ch and rescue”, especially for urban scale structural collapse environments [ 162 ]. According to the different operating environments, search and rescue robots can be divided into autonomous underwater vehicles (AUVs), and urban search and r escue (USAR) vehicles [ 163 ], which contain various types of unmanned ground vehicles (UGVs) and unmanned aerial vehicles (UA Vs). This literature r eview focuses on the r esearch of USAR r obotic systems, and the resear ch on AUVs is excluded due to its distinctive emphases caused by the turbid underwater environment. Due to the limitations on the mobility (tracks, wheels or combination of both) and intelligence, most of the current sear ch and r escue robot systems ar e not advanced enough for fully autonomous operations. Hence, a research dir ection is to investigate the human-robot interaction during search and rescue operations. The ﬁrst known use of USAR robots for an actual unstaged rescue mission occurs during the W orld T rade Center collapse [ 164 ], when various models of tele-operated robots from industry , military and academia were invited to assist the rescue operation; these r obots carried various sensors such as cameras, microphones and speakers to pr ovide r eal time video and audio for remote human operators for victim identiﬁcation, and various actuators such as a r obot arm with a gripper attached topside for victim extrication, a medical tube for providing fresh air or water; moreover , this work conducts a post-hoc analysis on the performance of the human-robot interaction during the r escue mission based on the video tape and ﬁeld notes; the analysis r eveals the pr essing needs of improving the mobility , intelligence and assistive interfaces of rescue robots, as well as reducing the number of people needed to operate a robot since the performance of operators are signiﬁcantly affected by cognitive fatigue caused by lack of sleep. The resear ch in [ 165 ] presents a semi-autonomous robot contr oller to share and co-execute tasks with remote human operators during the exploration of unknown disaster scenes; instead of using fully human-supervised robotic control which may induce cognitive and physical fatigue, disorientation and incorr ect judgements, a MAXQ hierarchical r einforcement learning (HRL) based controller is developed in this study to determine whether the robot or the human operator should conduct a certain rescue task with respect to the prior knowledge and experiences; the MAXQ algorithm decomposes the overall SAR tasks into four types of subtasks that can be learned simultaneously: local navigation (obstacle avoidance), navigate to unvisited regions, victim identiﬁcation, and human control; each subtask can be further decomposed into child tasks recursively until each child task is a primitive action of the robot; then a task hierarchy can be built by iteratively mapping all the possible states of a subtask to its child tasks; each state-action pair is associated with a action-value function, which r epresents the expected cumulative rewar d for performing the action in the given state when following the task hierarchy; the MAXQ HRL-based learning algorithm updates the values of each state-action pair based on the received positive or negative rewar ds (e.g. a positive rewar d is given to the local navigation subtask when a collision is avoided) and the action with the highest value is selected as the optimal policy . One of the major challenges that teleoperated or semi-autonomous rescue robots face today is the high human-to-robot ratio (The human-to-robot ratio denotes the number of people needed to operate a robot.), which r eaches 2 : 1 or even 3 : 1 in the current state of practice [ 166 ]. The high human-to-robot ratio restricts the scale of rescue robots to be deployed in the devastated region since too many V ersion July 10, 2019 submitted to Electronics 20 of 33 operators can signiﬁcantly increase the logistic bur den and the training cost. One feasible solution to this problem is the use of homogeneous swarm robots, which are easier to control and understand. For instance, the r esearch in [ 167 ] employs a team of homogeneous r obots to sear ch and explor e several disaster sites; because of the possible environmental changes and robot failures in disaster sites, the robots ar e r equired to continuously r edistribute to fulﬁll the desir ed population fraction of r obots at each site; furthermore, due to the harsh physical conditions and r esource limitations, the inter-r obot communications can be unreliable and costly , and are not easily applicable during an search and rescue operation; hence, this study proposes a task allocation policy to redistribute the robots fr om an initial distribution to the desir ed distribution without inter-robot communications; the relation between population fraction of robots at each site and the elapsed time from the start of the process is expressed by a bunch of ordinary dif ferential equations, under the assumption that the graph model of the devastated region and all the transition probabilities for a certain robot to traverse from a certain site to another are known to all the r obots; by solving the ordinary dif ferential equations, each robot can compute the time point when the system r eaches the desired distribution in a decentralised manner . Inspired by the self-or ganised behavior of social insects, the study in [ 168 ] extends the work in [ 167 ] and proposes a stochastic task allocation strategy to assign a swarm of homogeneous robots to various tasks during an search and rescue operation without inter -robot communications; the latency (transition time) for a robot to switch from one task to another is modeled by an Erlang distribution to capture the fact that the transition times have positive, minimum possible values and a small likelihood of being large because of low battery power , accidents or breakdowns; hence, the behaviours of the swarm are modelled by a bunch of delay differ ential equations with population fractions as variables; these delay differ ential equations are then converted into equivalent ordinary differ ential equations by ﬁtting the transition times with empirical values; by solving these ordinary differ ential equations, each robot can calculate the time point when the swarm reaches the desired population fractions at tasks in a distributed manner; to ensur e the fast conver gence of the system, the transition pr obabilities for a certain robot to switch fr om a certain task to another are optimised with various optimisation tools. Owing to the diverse and complex tasks in r escue operations, most of the recently proposed USAR robot systems consist of heterogeneous r obots with different capabilities instead of homogeneous robots. For instance, the resear ch in [ 169 ] proposes a robot gr oup which consists of thr ee types of robots to rescue victims during nuclear plant accidents; information-collecting robots are used to gather the location and postur e information of victims; traction r obots are employed to manipulate the victims’ posture to an easy-to-carry state; transportation robots ar e coupled to form a str etcher to carry victims out of the hazardous area; to manipulate the posture of a victim to a desired attitude, the body of the victim is simpliﬁed as a link model where limbs are considered as links and joints are consider ed as vertices; the angle deviation between the targeted posture and the initial posture is calculated by the dot pr oduct of the vectors associated with these two postur es; a r obotic hand is attached on the traction r obots to grasp and move the limbs of the victim; to avoid the human body from colliding with the walls during the transport process, a Generalised V oronoi Graph based motion path planning algorithm is utilised to compute a path that is equidistant from the surr ounding walls and the obstacles. The study in [ 170 ] presents a USAR r obot system which contains aerial robots and a land robot to detect the potential survivors in high-destruction locations after an earthquake; the aerial robots ar e responsible for mapping the inter ested area, sear ching desired paths with less debris and more victims for the land robot, and maintaining the communications between the land robot and remote human operators; on the other hand, the land r obot is used to detect survivors with speciﬁc sensors such as onboar d cameras; to climb over debris or partially destroyed stairs, the land robot, which is composed of a main body and a frontal body with two side tracked wheels, is designed with the shape-shifting ability; the relative angular orientation between the main body and the frontal body is adaptable to adjust the mass center of the robot and provide extra traction power; moreover , to free the human operators from tedious and difﬁcult tasks, this system also provide an immersive 3D head-mounted display to impr ove the situation awareness of the human operators, and a visual V ersion July 10, 2019 submitted to Electronics 21 of 33 servoing based automatic docking subsystem to dock and undock the r obot to the tether , which carries an electrical power supply and communication devices. In addition, to verify the ef fectiveness of the proposed r obotic systems, various real and simulated scenarios have been designed in the last two decades. The resear ch in [ 171 ] designs a test arena with collapsed structures to evaluate the agility , planning or mapping algorithms and sensing ability of rescue robots. T o mimic large scale disasters, a proving ground namely “Disaster City” with a size of 210, 000 m 2 has been built by the US Federal Emergency Agency (FEMA) to conduct rescue drills for robots [ 172 ]. The European counterparts ar e the Rescue Robot Fieldtest which simulates a trafﬁc accident involving a hazar dous material truck and European Land Robot T rials (ELROB) that can imitate a large scale incident such as a terrorist attack at a public event. The RoboCupRescue simulator in [ 173 ] integrates various disaster models and intelligent agents to build a virtual environment, and a real-world interface is designed to connect the virtual world with facilities in the real disaster ﬁeld; evacuees’ decisions and rescue plans are transmitted back to the simulator via a human-machine interface. 3.2. Emergency Sear ch and Rescue Planning Algorithms The state-of-the-art emergency search and rescue planning studies are originated and derived from research in maritime search and rescue activities, which has been a signiﬁcant research topic since 1970s. Related algorithms in emergency search and r escue planning are commonly under the assumption that multiple agents (r escuers, r obots, etc.) are involved. The reason behind this is twofold. First, time efﬁciency is the dominant factor for the success of a search and rescue operation, and multi-agent based search and r escue operations ar e obviously more ef ﬁcient than single-agent based search and rescue operations. Second, multi-agent systems are generally more cost-efﬁcient and feasible than a single agent with all the capabilities [ 174 ]. In recent decades, r elated algorithms have aroused a new inter est in not only searching victims in hazar dous envir onments, but also de-mining [ 175 ] and planetary exploration [ 176 ]. The philosophy behind team-based search and rescue is to convert a complex problem into simpler sub-problems that ar e more efﬁcient to solve. These algorithms are be classiﬁed into thr ee types: ﬁrst, searching algorithms, which ar e dedicated to search and locate injured evacuees in unknown environments, speciﬁcally concentrating on coordinating the search hehaviours of rescuers to pr oduce “swarm intelligence”; second, resour ce allocation algorithms, which aim to assign rescuers to injured evacuees in a desired manner when the locations of evacuees are revealed; third, navigation algorithms, which focus on discovering appropriate paths for r escuers when the locations of evacuees ar e known. In this section, we r eview the r elated algorithms of ﬁrst two categories. The third category is excluded here since all the on-line emergency navigation algorithms are applicable to this case, and the detailed r eview can be found in Section 2.2.2 . 3.2.1. Searching Algorithms in Emer gency Search and Rescue Planning The resear ch in [ 177 ] proposes a particle swarm optimisation (PSO) based multi-r obot system to ﬁnd targets by gradually optimising the pre-deﬁned goal function; each r obot is modeled as a particle whose velocity and position is determined by the neighbouring particles and the previous best position of this particle. Similarly , the study in [ 178 ] designs a PSO based multi-robot system to sear ch targets and avoid obstacles in an unknown environment; specially , a relative coordinate system is used to avoid the dependence of the precise global location of r obots. By employing a bio-inspir ed random search behaviour called lévy ﬂight, the research in [ 179 ] presents an efﬁcient multi-robot searching algorithm to search targets in an unknown environment; an artiﬁcial potential ﬁeld is used to generate repulsive for ces to disperse r obots amongst the area of interest. The research in [ 180 ] deploys robots in a triangular grid pattern, which can minimise the number of r obot r equired for the coverage of an area, to search an unknown zone; the tar geted ar ea is ﬁrstly partitioned into equilateral triangles and then the robots will sear ch along the edges of the triangular grid. T o minimise the total travel distance during the search process, the study in [ 181 ] utilises an auction based algorithm to assign V ersion July 10, 2019 submitted to Electronics 22 of 33 desired tours to robots. The resear ch in [ 182 ] presents a multi-robot coor dination algorithm to identify ﬁre sour ces in an indoor environment; exploration frontiers are extracted and assigned to robots to minimise the overall exploration time. Frontier -based searching algorithms such as [ 165 , 183 ] normally concentrate on assigning robots to the boundary line between the visited and unvisited ar eas; the search direction is commonly affected by the cost for a robot to explore a boundary area (e.g. the probability of an obstacle or another r obot being present) and the cost for the robot to r each the target boundary area (e.g. the distance from the target boundary area). The study in [ 184 ] investigates an interesting r esear ch issue of sear ching a ﬁxed object in an unknown environment with a number of searchers performing Br ownian motions; the sear chers are sent out consecutively from the sour ce to locate the object at a ﬁnite distance D , the search space is unbounded and the distance D is unknown; each searcher is subject to a ﬁnite life span and could be destroyed or disabled during the search; a time-out is set by the source to eliminate searchers that sear ch for a long time without discovering the object; when the time-out elapses, a new searcher will be sent out to replace the “lost” sear chers; a closed form expression for the average search time is derived as a function of the distance between the source and the object (“ D ”) by modelling the search process as a multidimensional Br ownian process; the experimental r esults show that the average sear ch time is af fected by number , life span, routing uncertainty , destruction rate of the sear chers. The research in [ 185 ] presents a robotic search and rescue system to search for victims in rubbles with lost cost homogeneous robots carrying thermal array sensors; a general suppression control framework, which is inspired by the suppression mechanism of the human immune system, is proposed to regulate the searching and exception handling behaviours of the robots. 3.2.2. Resource Allocation Algorithms in Emer gency Search and Rescue Planning The study in [ 151 ] presents a r esource allocation algorithm on the basis of the Random Neural Network (RNN) with synchronised interactions [ 89 ] to assign rescuers to trapped victims in a ﬁre-af fected building; to optimise this task assignment process, several effect factors such as the cost of assigning a rescuer to a victim (e.g. the distance between the rescuer and the victim), the probability that a rescuer fails to save a victim, and the associated penalty that the rescuer fails to rescue the victim ar e taken into consideration, and are transformed into a goal function which reﬂects the trade-off between the total successful r escue costs and the failur e penalties; the RNN is utilised as a fast optimisation algorithm to minimise the goal function with a gradient descent learning procedure; in the RNN, each possible rescuer-victim pair is considered as a neuron, and the neuron with the highest excitation pr obability after the training pr ocess is selected as the decision. The r esearch in [ 186 ] proposes a task allocation algorithm for heter ogeneous agents in hazar d environments with respect to time, space and communication restrictions; to reduce the communication load and links among agents, each agent elects the agent with the most direct neighbours within its communication range as the communication network leader; the elected network leader will be responsible for maintaining the communication among intra-network agents and other network leaders; to ensure the appropriate distance among task locations and the agents, each network leader further divides its coverage ar ea into sub-areas by using the mean shift clustering algorithm; the window radius parameter h in the mean shift clustering algorithm is set to the communication range of the network leader to guarantee that the agents in the same sub-area can always communicate with each other; heterogeneous agents are then allocated into these sub-ar eas in terms of their abilities and the r equirements of tasks within the sub-areas: ﬁrstly , the “similarity” value of each agent sub-area pair is calculated by the dot product [ 187 ] of the vector of the requirements of tasks and the normalised vector of ability of the agent; secondly , the agent will be assigned to the sub-area with the highest similarity value. As aforementioned in subsection 3.1.2 , the studies in [ 167 , 168 ] employ an ordinary dif fer ential equation model to redistribute rescue r obots among various tasks in a decentralised manner without inter-robot communications; the relation between population fractions at tasks and the elapsed time since the start of the process is expr essed by a set of ordinary differ ential equations; by substituting the desired V ersion July 10, 2019 submitted to Electronics 23 of 33 population fraction in each task into the proposed model, each robot can independently calculate the time point when the system reaches the desir ed distribution. 4. Emerging Challenges and Opportunities After decades of study and exploration, emergency management has become a mature r esear ch ﬁeld. However , due to its open and inclusive nature, new technologies always tend to inﬂuence, change or even revolutionise this research area. In this section we discuss open issues and provide possible directions for future work. W e also visualise these research dir ections by using a sunburst chart as shown in Figure 3 .  Figure 3. A sunburst chart illustrating the potential research dir ections. The emergence of heuristic based algorithms has of fered near-optimal solutions for emergency management in a computationally and time efﬁcient manner at the trade-off of optimality and completeness. However , the setting of key parameters in heuristics could play a vital role in system performance and the perfect parameter settings for one evacuation scenario may not suit others. Hence, currently , signiﬁcant parameters involved in heuristics are mostly pr e-conﬁgured in a supervised-learning fashion rather than using a unsupervised learning paradigm and therefor e may not adapt to uncertain or fast-changing envir onments, and may even need to be calibrated manually in differ ent scenarios. T aking WSN-based algorithms for example, the work in [ 145 ] has shown that these parameters could contribute signiﬁcantly to the efﬁciency of route discovery and information collection. For instance, if the time-to-live of packets is too lar ge, the system will be overburdened with packets that are in effect lost. On the other hand, if the live-time constraint is too small, some distant exits may not be discovered. Likewise, the improper packet rate could induce unnecessary energy consumption when appropriate paths have been discovered and the network situation stays unchanged. Hence, future r esearch can be directed to optimising the setting of parameters fr om a systemic point of view by utilising proper queueing network models or dif fusion models. Potential differ entiated services (DiffServ) mechanisms could be also developed to optimise the packet behaviours and satisfy the QoS requir ements of differ ent categories of cognitive packets which are associated with diverse classes of evacuees. Furthermore, parameter optimisation algorithms or even parameter free algorithms can V ersion July 10, 2019 submitted to Electronics 24 of 33 be developed by gaining experience from iterations of various simulated scenarios with the aid of newly-proposed machine learning algorithms such as deep learning. Compared with emergency drills, computer simulations are commonly utilised to investigate the effectiveness of emer gency management algorithms due to their time-efﬁciency , repeatability and cost-efﬁciency . However , many human behaviours during emergency evacuations have been ignored in the simulations. Hence, futur e resear ch can also be dir ected to develop various human-computer interfaces for simulators to generate an artiﬁcial hazard environment with virtual reality and augmented r eality technologies [ 188 ]. By evacuating volunteers fr om the artiﬁcial hazar d envir onment, their emer gent behaviours can be collected and empirical collective behaviour models of human beings can be developed from the actual cr owd measurements. One trending in emer gency management is to develop artiﬁcial intelligence aided algorithms to improve path-ﬁnding and resource allocation during standard or emergent evacuations. However , as a typical category of cyber-physical systems, the intelligence of evacuees, as well as the possible pro-social behaviours such as helpfulness and sense of duty , have been excluded in the previous algorithms. As a result, the robustness of these algorithms cannot be ensured due to the high likelihood that evacuees do not follow the instructions. Moreover , unnecessary efforts have been dedicated to the use of AI, while in fact many tasks can be easily accomplished by evacuees in the system via using human intelligence. For instance, the identiﬁcation of a ﬁr e source location could take signiﬁcant efforts when using AI-based algorithms, which involves in the installation of cameras and the use of computer version related technologies. However , it would take no time or cost for evacuees to identify a ﬁre and report to authorities or the emer gency response system by using smart phones. In attempting to incentivise evacuees to conduct cooperative behaviours and avoid destructive behaviours, one future resear ch direction could be dedicated to the investigation of rewar d mechanisms to integrate the human intelligence of evacuees into the emergency r esponse systems to improve the efﬁciency , adaptiveness and robustness of these systems. On the other hand, since most of the previous multi-robot rescue systems only follow simple coordination rules and lack explicit teamwork models or goals [ 189 ], multi-agent technologies can also be utilised to model and develop various cooperative strategies with the aid of queueing network models such as G-networks [ 190 ] or genetic algorithms. The rapid development of Cloud technologies has facilitated the advancement of cloud-enabled emergency management systems that consist of front-end portable devices and back-end cloud severs. Future resear ch can be directed to further improve the ﬂexibility of the kind of system by leveraging the mobile agent technology , because current cloud-based emer gency response systems, which are based on the client-sever paradigm, demand pre-installed services in participating devices [ 191 ]. Moreover , since mobile agents can migrate seamlessly through multiple clouds and dif ferent portable devices, a mobile agent-based emergency r esponse system has the potential to reduce communication costs and ease network congestion in lar ge-scale emergency evacuations by dynamically optimising the locations of mobile agents. Moreover , we believe that future emergency management systems should not only make use of the computing power of portable devices and the Cloud, but also other individual devices in the vicinity to offer services with the aid of edge computing and fog computing. Although the ﬁre alarm system has become a norm for modern buildings, more complicated emergency management systems mostly remain in pr ototype form and are only veriﬁed in simulations or speciﬁc test ﬁelds. The raising of the concept “smart city” has provided a golden opportunity for emergency management systems to be integrated into the smart city “eco-system” and take advantage of this tide to achieve leapfrogging. The potential research directions are threefold. First, with the recent incr ease of manmade disasters, physical attacks are highly likely to be accompanied with cyber attacks. Hence, it is of critical importance to develop self-aware networks to self-defense and maintain communications during an evacuation process. On suggested framework is CPN, which is naturally efﬁcient on defending various forms of malicious attacks such as denial of service attacks [ 192 ]. This is because, unlike the IP pr otocol, smart packets in the CPN carry the “full path” fr om the sour ce node to the destination node. Hence, CPN can defend denial of service attacks by adaptively dropping V ersion July 10, 2019 submitted to Electronics 25 of 33 attacking packets upstream from the node being attacked via backtracking the full path of attacking ﬂows. Second, due to the ener gy hungry communication and computation pr ocesses inside emer gency management systems as well as the fact that wir eless sensors ar e difﬁcult to be replenished, ener gy efﬁcient and ener gy harvesting algorithms, such as dynamic programming and G-network models [ 193 – 195 ], can be applied to future emergency management systems to improve the performance and efﬁciency of these systems. Thir d, the massive deployment of sensors with the prosperity of the Internet of Things(IoT) and smart city technologies impr ove the ef ﬁcient collection of individual data on a vast scale. This suggests a crucial new opportunity to use big data technologies to analyse, model and reﬁne the personal preferences, human collective choices and behaviours, and general rules with r espect to the routinely collected data, and use them as guidance for the futur e emergency management algorithm and system design. 5. Conclusion In this paper , we provide a systemic review of the emergency management resear ch. In the ﬁrst section, we review the history and evolution of this ﬁeld, trace its transformation from a reactive manner to a proactive manner . W e also explor e the impact of the development of computer technologies, which has shaped the curr ent resear ch. In the second section, we r eview the emer gency search and rescue planning from both system design aspect and algorithm engineering aspect. In the system review part, various wearable computing-assisted r escue systems, and sear ch & r escue robotic systems ar e introduced. In the algorithm review part, r epresentative sear ching, resour ce allocation and navigation algorithms are discussed. Similarly , in the third section, we summarise classical emergency navigation systems and algorithms. The systems are classiﬁed into human experience driven systems, static WSN based systems, mixed WSN based systems, cloud based systems with WSN, and cloud based systems with mobile phones. The algorithms are divided into off-line navigation algorithms and on-line emergency navigation algorithms. The off-line navigation algorithms are further divided into cellular automata model based algorithms, social force model based algorithms, ﬂuid-dynamics model based algorithms, lattice gas model based algorithms, game theoretic model based algorithms, computer agents based algorithms and animal agent based algorithms. On the other hand, on-line emer gency navigation algorithms are classiﬁed into network ﬂow based algorithms, geometric algorithms, queueing model based algorithms, potential-maintenance algorithms, biologically inspired algorithms, routing pr otocol based algorithms and prediction-based algorithms. In the next section, we discuss the emerging challenges and opportunities under the backgr ound of the fast development and prosperity of new technologies such as smart city , artiﬁcial intelligence, virtual reality , big data, cyber security and energy ef ﬁciency & harvesting. Conﬂicts of Interest: The authors declare no conﬂict of interest. 1. Drabek, T .E.; Hoetmer , G.J. Emergency management: Principles and practice for local government. Washington, DC: International City Management Association 1991 . 2. Cronstedt, M.; others. Prevention, prepar edness, response, recovery-an outdated concept? Australian Journal of Emergency Management, The 2002 , 17 , 10. 3. Haddow , G.; Bullock, J.; Coppola, D.P . Introduction to emergency management ; Butterworth-Heinemann, 2017. 4. Gelenbe, E.; Cao, Y . Autonomous search for mines. European Journal of Operational Research 1998 , 108 , 319–333. 5. Gelenbe, E.; Hussain, K.; Kaptan, V . Autonomous search for mines. Journal of Systems and Softwar e 2005 , 74 , 255–268. 6. Belardo, S.; Karwan, K.R.; W allace, W . An investigation of system design considerations for emer gency management decision support. Systems, Man and Cybernetics, IEEE T ransactions on 1984 , pp. 795–804. V ersion July 10, 2019 submitted to Electronics 26 of 33 7. Chalmet, L.; Francis, R.; Saunders, P . Network models for building evacuation. Fire T echnology 1982 , 18 , 90–113. 8. Hughes, T . Graph processing by which to evacuate a mine. Applied Computing, 1990., Proceedings of the 1990 Symposium on, 1990, pp. 137–141. 9. Southworth, F .; Chin, S.M.; Cheng, P . A telemetric monitoring and analysis system for use during large scale population evacuations. Road T rafﬁc Monitoring, 1989., Second International Conference on, 1989, pp. 99–103. 10. Grifﬁth, D. The critical pr oblems of hurricane evacuation and alternative solutions. OCEANS 82, 1982, pp. 990–994. 11. Zorpette, G. Evacuation planning for Lilco’s Shoreham plant: Lack of an approved emergency evacuation strategy may prevent full operation of a $5 billion nuclear power plant on long Island’s north shore. Spectrum, IEEE 1987 , 24 , 22–24. 12. Serpa, D.P .; W alker , D.M.; Jenckes, T .A. Emergency Monitoring, Assessment and Response System for Diablo Canyon Nuclear Power Plant. Nuclear Science, IEEE T ransactions on 1981 , 28 , 236–241. 13. Filippoupolitis, A.; Gelenbe, E. A distributed decision support system for Building Evacuation. Human System Interactions, 2009. HSI ’09. 2nd Conference on, 2009, pp. 323–330. 14. T seng, Y .C.; Pan, M.S.; T sai, Y .Y . W ireless sensor networks for emergency navigation. Computer 2006 , 39 , 55–62. 15. Park, V .D.; Corson, M.S. A highly adaptive distributed r outing algorithm for mobile wir eless networks. INFOCOM’97. Sixteenth Annual Joint Conference of the IEEE Computer and Communications Societies. Driving the Information Revolution., Proceedings IEEE. IEEE, 1997, V ol. 3, pp. 1405–1413. 16. Li, Q.; De Rosa, M.; Rus, D. Distributed algorithms for guiding navigation across a sensor network. Proceedings of the 9th annual international conference on Mobile computing and networking. ACM, 2003, pp. 313–325. 17. Koditschek, D.E. Robot planning and control via potential functions. The robotics r eview 1989 , p. 349. 18. Gorbil, G.; Gelenbe, E. Opportunistic communications for emergency support systems. Procedia Computer Science 2011 , 5 , 39–47. 19. Gelenbe, E.; Gorbil, G. W ireless networks in emergency management. Pr oceedings of the ﬁrst ACM international workshop on Practical issues and applications in next generation wireless networks. ACM, 2012, pp. 1–6. 20. Gorbil, G.; Gelenbe, E. Resilient emergency evacuation using opportunistic communications. In Computer and Information Sciences III ; Springer , 2013; pp. 249–257. 21. Pelusi, L.; Passarella, A.; Conti, M. Opportunistic networking: data forwarding in disconnected mobile ad hoc networks. Communications Magazine, IEEE 2006 , 44 , 134–141. 22. Filippoupolitis, A.; Gorbil, G.; Gelenbe, E. Autonomous navigation systems for emergency management in buildings. GLOBECOM W orkshops (GC Wkshps), 2011 IEEE, 2011, pp. 1056–1061. 23. Zubair , S.; bnt Fisal, N.; Y erima, S.; Salihu, B.; Salihu, Y . CoW iSMoN: A framework for cognitive wir eless sensor mobile network system for emergency r escue management. Adaptive Science and T echnology (ICAST), 2011 3rd IEEE International Confer ence, 2011, pp. 237–241. 24. Inoue, Y .; Sashima, A.; Ikeda, T .; Kurumatani, K. Indoor Emergency Evacuation Service on Autonomous Navigation System using Mobile Phone. Universal Communication, 2008. ISUC ’08. Second International Symposium on, 2008, pp. 79–85. 25. T seng, Y .C.; W ang, Y .C.; Cheng, K.Y .; Hsieh, Y .Y . iMouse: An Integrated Mobile Surveillance and W ireless Sensor System. Computer 2007 , 40 , 60–66. 26. Dong, M.; Li, H.; Ota, K.; Y ang, L.; Zhu, H. Multicloud-Based Evacuation Services for Emergency Management. Cloud Computing, IEEE 2014 , 1 , 50–59. 27. Chu, L.; W u, S.J. An Integrated Building Fire Evacuation System with RFID and Cloud Computing. Intelligent Information Hiding and Multimedia Signal Pr ocessing (IIH-MSP), 2011 Seventh International Conference on, 2011, pp. 17–20. 28. Qiu, M.; Ming, Z.; W ang, J.; Y ang, L.T .; Xiang, Y . Enabling Cloud Computing in Emergency Management Systems. IEEE Cloud Computing 2014 , 1 , 60–67. 29. Gelenbe, E.; Bi, H. Emergency Navigation without an Infrastructure. Sensors 2014 , 14 , 15142–15162. 30. Helbing, D.; Farkas, I.; V icsek, T . Simulating dynamical features of escape panic. Nature 2000 , 407 , 487–490. V ersion July 10, 2019 submitted to Electronics 27 of 33 31. Y ang, L.; Zhao, D.; Li, J.; Fang, T . Simulation of the kin behavior in building occupant evacuation based on cellular automaton. Building and Environment 2005 , 40 , 411–415. 32. Gwynne, S.; Galea, E.; Owen, M.; Lawrence, P .J.; Filippidis, L.; others. A review of the methodologies used in evacuation modelling. Fire and Materials 1999 , 23 , 383–388. 33. Zheng, X.; Zhong, T .; Liu, M. Modeling crowd evacuation of a building based on seven methodological approaches. Building and Environment 2009 , 44 , 437–445. 34. W olfram, S. Statistical mechanics of cellular automata. Reviews of modern physics 1983 , 55 , 601. 35. Bandini, S.; Manzoni, S.; V izzari, G. Situated Cellular Agents: a model to simulate crowding dynamics. IEICE T ransactions on Information and Systems: Special Issue on Cellular Automata 2004 , E87-D(3) , 669–676. 36. Y u, Y .; Song, W . Cellular automaton simulation of pedestrian counter ﬂow considering the surr ounding environment. Physical Review E 2007 , 75 , 046112. 37. Spartalis, E.; Georgoudas, I.G.; Sirakoulis, G.C. CA Crowd Modeling for a Retir ement House Evacuation with Guidance. In Cellular Automata ; Springer , 2014; pp. 481–491. 38. Müller , F .; W ohak, O.; Schadschneider , A. Study of Inﬂuence of Groups on Evacuation Dynamics Using a Cellular Automaton Model. T ransportation Research Pr ocedia 2014 , 2 , 168–176. 39. Helbing, D.; Molnar , P . Social force model for pedestrian dynamics. Physical review E 1995 , 51 , 4282. 40. Parisi, D.; Dorso, C. Microscopic dynamics of pedestrian evacuation. Physica A: Statistical Mechanics and its Applications 2005 , 354 , 606–618. 41. Seyfried, A.; Steffen, B.; Lippert, T . Basics of modelling the pedestrian ﬂow . Physica A: Statistical Mechanics and its Applications 2006 , 368 , 232–238. 42. Henderson, L. The statistics of crowd ﬂuids. Nature 1971 , 229 , 381–383. 43. Helbing, D.; Farkas, I.J.; Molnar , P .; V icsek, T . Simulation of pedestrian crowds in normal and evacuation situations. Pedestrian and evacuation dynamics 2002 , 21 , 21–58. 44. Fredkin, E.; T offoli, T . Conservative logic ; Springer , 2002. 45. T ajima, Y .; T akimoto, K.; Nagatani, T . Scaling of pedestrian channel ﬂow with a bottleneck. Physica A: Statistical Mechanics and its Applications 2001 , 294 , 257–268. 46. T akimoto, K.; Nagatani, T . Spatio-temporal distribution of escape time in evacuation process. Physica A: Statistical Mechanics and its Applications 2003 , 320 , 611–621. 47. Hoogendoorn, S.; HL Bovy , P . Simulation of pedestrian ﬂows by optimal control and dif ferential games. Optimal Control Applications and Methods 2003 , 24 , 153–172. 48. Lo, S.M.; Huang, H.C.; W ang, P .; Y uen, K. A game theory based exit selection model for evacuation. Fire Safety Journal 2006 , 41 , 364–369. 49. Ehtamo, H.; Heliövaara, S.; Hostikka, S.; Korhonen, T . Modeling evacuees’ exit selection with best response dynamics. In Pedestrian and Evacuation Dynamics 2008 ; Springer , 2010; pp. 309–319. 50. Zheng, X.; Cheng, Y . Modeling cooperative and competitive behaviors in emergency evacuation: A game-theoretical appr oach. Computers & Mathematics with Applications 2011 , 62 , 4627–4634. 51. Bonabeau, E. Agent-based modeling: Methods and techniques for simulating human systems. Proceedings of the National Academy of Sciences of the United States of America 2002 , 99 , 7280–7287. 52. Zarboutis, N.; Marmaras, N. Searching efﬁcient plans for emer gency rescue thr ough simulation: the case of a metro ﬁr e. Cognition, T echnology & Work 2004 , 6 , 117–126. 53. Goldstone, R.L.; Janssen, M.A. Computational models of collective behavior . T rends in cognitive sciences 2005 , 9 , 424–430. 54. Pan, X.; Han, C.S.; Dauber , K.; Law , K.H. A multi-agent based framework for the simulation of human and social behaviors during emergency evacuations. Ai & Society 2007 , 22 , 113–132. 55. Saloma, C.; Perez, G.J.; T apang, G.; Lim, M.; Palmes-Saloma, C. Self-organized queuing and scale-free behavior in real escape panic. Proceedings of the National Academy of Sciences 2003 , 100 , 11947–11952. 56. Altshuler , E.; Ramos, O.; Núñez, Y .; Fernández, J.; Batista-Leyva, A.; Noda, C. Symmetry br eaking in escaping ants. The American Naturalist 2005 , 166 , 643–649. 57. Pelechano, N.; Malkawi, A. Evacuation simulation models: Challenges in modeling high rise building evacuation with cellular automata approaches. Automation in construction 2008 , 17 , 377–385. 58. Perez, G.J.; T apang, G.; Lim, M.; Saloma, C. Streaming, disruptive interfer ence and power -law behavior in the exit dynamics of conﬁned pedestrians. Physica A: Statistical Mechanics and its Applications 2002 , 312 , 609–618. V ersion July 10, 2019 submitted to Electronics 28 of 33 59. Burstedde, C.; Klauck, K.; Schadschneider , A.; Zittartz, J. Simulation of pedestrian dynamics using a two-dimensional cellular automaton. Physica A 2004 , 295 , 507–525. 60. Zheng, M.; Kashimori, Y .; Kambara, T . A model describing collective behaviors of pedestrians with various personalities in danger situations. Neural Information Processing, 2002. ICONIP ’02. Proceedings of the 9th International Conference on, 2002, V ol. 4, pp. 2083–2087 vol.4. 61. Hecht-Nielsen, R. Applications of counterpropagation networks. Neural networks 1988 , 1 , 131–139. 62. Hughes, R.L. A continuum theory for the ﬂow of pedestrians. T ransportation Resear ch Part B: Methodological 2002 , 36 , 507–535. 63. Colombo, R.M.; Rosini, M.D. Pedestrian ﬂows and non-classical shocks. Mathematical Methods in the Applied Sciences 2005 , 28 , 1553–1567. 64. Lighthill, M.J.; Whitham, G.B. On kinematic waves. II. A theory of trafﬁc ﬂow on long crowded roads. Proceedings of the Royal Society of London A: Mathematical, Physical and Engineering Sciences. The Royal Society , 1955, V ol. 229, pp. 317–345. 65. Liu, T .P . The Riemann problem for general systems of conservation laws. Journal of Differential Equations 1975 , 18 , 218–234. 66. Thouless, D.J. The quantum mechanics of many-body systems ; Courier Corporation, 2013. 67. Kirchner , A.; Klüpfel, H.; Nishinari, K.; Schadschneider , A.; Schreckenberg, M. Simulation of competitive egress behavior: comparison with aircraft evacuation data. Physica A: Statistical Mechanics and its Applications 2003 , 324 , 689–697. 68. Fudenberg, D.; T irole, J. Game theory , 1991. Cambridge, Massachusetts 1991 , 393 . 69. Nash, J. Non-cooperative games. Annals of mathematics 1951 , pp. 286–295. 70. Nakamura, T .; Asada, M. Motion sketch: Acquisition of visual motion guided behaviors. IJCAI, 1995, V ol. 95, pp. 126–132. 71. March, J.G. Primer on decision making: How decisions happen ; Simon and Schuster , 1994. 72. Kuligowski, E.D.; Peacock, R.D.; Hoskins, B. A review of building evacuation models ; US Department of Commerce, National Institute of Standar ds and T echnology , 2005. 73. Gelenbe, E.; W u, F .J. Large scale simulation for human evacuation and r escue. Computers & Mathematics with Applications 2012 , 64 , 3869–3880. 74. Gelenbe, E.; W u, F .J. Future r esearch on cyber -physical emergency management systems. Future Internet 2013 , 5 , 336–354. 75. Francis, R. A negative exponential solution to an evacuation problem ; National Bur eau of Standards, Center for Fire Resear ch, 1984. 76. Kisko, T .M.; Francis, R.L. EV ACNET+: a computer program to determine optimal building evacuation plans. Fire Safety Journal 1985 , 9 , 211–220. 77. Lu, Q.; Huang, Y .; Shekhar , S. Evacuation planning: a capacity constrained r outing appr oach. In Intelligence and Security Informatics ; Springer , 2003; pp. 111–125. 78. Lu, Q.; George, B.; Shekhar , S. Capacity constrained routing algorithms for evacuation planning: A summary of results. In Advances in spatial and temporal databases ; Springer , 2005; pp. 291–307. 79. Chen, P .Y .; Chen, W .T .; Shen, Y .T . A distributed area-based guiding navigation protocol for wireless sensor networks. Parallel and Distributed Systems, 2008. ICP ADS’08. 14th IEEE International Conference on. IEEE, 2008, pp. 647–654. 80. W ang, J.; Li, Z.; Li, M.; Liu, Y .; Y ang, Z. Sensor network navigation without locations. Parallel and Distributed Systems, IEEE T ransactions on 2013 , 24 , 1436–1446. 81. MacGregor Smith, J. State-dependent queueing models in emergency evacuation networks. T ransportation Research Part B: Methodological 1991 , 25 , 373–389. 82. Cruz, F .R.; Smith, J.M.; Queiroz, D. Service and capacity allocation in M/G/c/c state-dependent queueing networks. Computers & operations research 2005 , 32 , 1545–1563. 83. Stepanov , A.; Smith, J.M. Multi-objective evacuation routing in transportation networks. European Journal of Operational Research 2009 , 198 , 435–446. 84. Lino, P .; Maione, G.; Maione, B. Modeling and simulation of cr owd egress dynamics in a discrete event environment. Control Applications,(CCA) & Intelligent Control,(ISIC), 2009 IEEE. IEEE, 2009, pp. 843–848. 85. Desmet, A.; Gelenbe, E. Graph and analytical models for emergency evacuation. Future Internet 2013 , 5 , 46–55. V ersion July 10, 2019 submitted to Electronics 29 of 33 86. Bi, H. Evacuee Flow Optimisation Using G-Network with Multiple Classes of Positive Customers. 2016 IEEE 24th International Symposium on Modeling, Analysis and Simulation of Computer and T elecommunication Systems (MASCOTS), 2016, pp. 135–143. 87. Bi, H.; Abdelrahman, O.H. ENERGY -A W ARE NA VIGA TION IN LARGE-SCALE EV ACUA TION USING G-NETWORKS. Probability in the Engineering and Informational Sciences 2016 , pp. 1–13. 88. Chen, W .T .; Chen, P .Y .; W u, C.H.; Huang, C.F . A load-balanced guiding navigation protocol in wir eless sensor networks. Global T elecommunications Conference, 2008. IEEE GLOBECOM 2008. IEEE. IEEE, 2008, pp. 1–6. 89. Gelenbe, E.; T imotheou, S. Random neural networks with synchronized interactions. Neural Computation 2008 , 20 , 2308–2324. 90. Jankowska, A.; Schut, M.; Ferreira-Schut, N. A wireless actuator-sensor neural network for evacuation routing. Sensor T echnologies and Applications, 2009. SENSORCOMM’09. Third International Conference on. IEEE, 2009, pp. 139–144. 91. Li, Q.; Fang, Z.; Li, Q.; Zong, X. Multiobjective evacuation route assignment model based on genetic algorithm. Geoinformatics, 2010 18th International Conference on. IEEE, 2010, pp. 1–5. 92. Filippoupolitis, A. Emergency Simulation and Decision Support Algorithms. PhD thesis, Imperial College London (University of London), 2010. 93. Bi, H.; Desmet, A.; Gelenbe, E. Routing Emergency Evacuees with Cognitive Packet Networks. Proceedings of the 28th International Symposium on Computer and Information Sciences (ISCIS’13). Springer , London, 2013. 94. Bi, H.; Gelenbe, E. Routing diverse evacuees with Cognitive Packets. Pervasive Computing and Communications W orkshops (PERCOM W orkshops), 2014 IEEE International Conference on, 2014, pp. 291–296. 95. Hasofer , A.; Odigie, D. Stochastic modelling for occupant safety in a building ﬁre. Fire Safety Journal 2001 , 36 , 269–289. 96. Barnes, M.; Leather , H.; Arvind, D. Emergency evacuation using wireless sensor networks. Local Computer Networks, 2007. LCN 2007. 32nd IEEE Conference on. IEEE, 2007, pp. 851–857. 97. Han, L.; Potter , S.; Beckett, G.; Pringle, G.; W elch, S.; Koo, S.H.; W ickler , G.; Usmani, A.; T orer o, J.L.; T ate, A. FireGrid: an e-infrastructure for next-generation emergency r esponse support. Journal of Parallel and Distributed Computing 2010 , 70 , 1128–1141. 98. Radianti, J.; Granmo, O.C.; Sarshar , P .; Goodwin, M.; Dugdale, J.; Gonzalez, J.J. A spatio-temporal probabilistic model of hazar d-and crowd dynamics for evacuation planning in disasters. Applied Intelligence 2015 , 42 , 3–23. 99. Bi, H.; Gelenbe, E. Cloud enabled emergency navigation using faster -than-real-time simulation. 2015 IEEE International Conference on Pervasive Computing and Communication W orkshops (PerCom W orkshops), 2015, pp. 475–480. 100. Ford, D.; Fulkerson, D.R. Flows in networks ; Princeton university press, 2010. 101. Ahuja, R.K.; Magnanti, T .L.; Orlin, J.B. Network ﬂows 1988 . 102. Graves, G.W .; Br own, G.G.; Bradley , G.H. Design and Implementation of Lar ge-Scale Primal T ransshipment Algorithms. Management Science 1977 , 24 , 1–34. 103. Hoppe, B.; T ardos, É. Polynomial time algorithms for some evacuation pr oblems. Proceedings of the ﬁfth annual ACM-SIAM symposium on Discrete algorithms. Society for Industrial and Applied Mathematics, 1994, pp. 433–441. 104. Hoppe, B.; T ardos, É. The quickest transshipment problem. Mathematics of Operations Research 2000 , 25 , 36–62. 105. Megiddo, N. Optimal ﬂows in networks with multiple sources and sinks. Mathematical Programming 1974 , 7 , 97–107. 106. Minieka, E. Maximal, lexicographic, and dynamic network ﬂows. Operations Research 1973 , 21 , 517–527. 107. Dijkstra, E.W . A note on two problems in connexion with graphs. Numerische mathematik 1959 , 1 , 269–271. 108. Li, X.Y .; Calinescu, G.; W an, P .J. Distributed construction of a planar spanner and routing for ad hoc wireless networks. INFOCOM 2002. T wenty-First Annual Joint Conference of the IEEE Computer and Communications Societies. Proceedings. IEEE. IEEE, 2002, V ol. 3, pp. 1268–1277. V ersion July 10, 2019 submitted to Electronics 30 of 33 109. Li, X.Y .; Calinescu, G.; W an, P .J.; W ang, Y . Localized delaunay triangulation with application in ad hoc wireless networks. Parallel and Distributed Systems, IEEE T ransactions on 2003 , 14 , 1035–1047. 110. Bruck, J.; Gao, J.; Jiang, A. MAP: Medial axis based geometric routing in sensor networks. Wir eless Networks 2007 , 13 , 835–853. 111. Gelenbe, E.; Muntz, R.R. Probabilistic models of computer systems, Part I (Exact Results). Acta Informatica 1976 , 7 , 35–60. 112. Gelenbe, E. A uniﬁed approach to a class of page replacement algorithms. IEEE T ransactions on Computers 1973 , 22 , 611–618. 113. Fruin, J.J. Pedestrian planning and design ; Metr opolitan Association of Urban Designers and Environmental Planners, Inc., 1971. 114. T regenza, P . The design of interior circulation ; V an Nostrand Reinhold, 1976. 115. Francis, R.; Chalmet, L. Network models for building evacuation: A prototype primer ; 1980. 116. Y uhaski Jr , S.J.; Smith, J.M. Modeling circulation systems in buildings using state dependent queueing models. Queueing Systems 1989 , 4 , 319–338. 117. Cheah, J.Y .; Smith, J.M. Generalized M/G/c/c state dependent queueing models and pedestrian trafﬁc ﬂows. Queueing Systems 1994 , 15 , 365–386. 118. Kerbachea, L.; Smith, J.M. The generalized expansion method for open ﬁnite queueing networks. European Journal of Operational Research 1987 , 32 , 448–461. 119. Kerbache, L.; Smith, J.M. Asymptotic behavior of the expansion method for open ﬁnite queueing networks. Computers & Operations Research 1988 , 15 , 157–169. 120. Smith, J.M.; Gershwin, S.B.; Papadopoulos, C.T . Performance evaluation and optimization of production lines ; V ol. 93, Baltzer Science Publishers, 2000. 121. Lino, P .; Pizzileo, B.; Maione, G.; Maione, B. T uning and V alidation of a Discrete-Event Model of the Egress Dynamics from Buildings*. W orld Congress, 2011, V ol. 18, pp. 8743–8748. 122. Helbing, D.; Farkas, I.; V icsek, T . Simulating dynamical features of escape panic. Nature 2000 , 407 , 487–490. 123. W ang, P .; Luh, P .B.; Chang, S.C.; Sun, J. Modeling and optimization of cr owd guidance for building emergency evacuation. Automation Science and Engineering, 2008. CASE 2008. IEEE International Conference on. IEEE, 2008, pp. 328–334. 124. W atts, J.M. Computer models for evacuation analysis. Fire Safety Journal 1987 , 12 , 237–245. 125. Cruz, F .R.; Smith, J.M.; Medeiros, R. An M/G/C/C state-dependent network simulation model. Computers & Operations Research 2005 , 32 , 919–941. 126. Gelenbe, E.; Labed, A. G-networks with multiple classes of signals and positive customers. European journal of operational resear ch 1998 , 108 , 293–305. 127. Gelenbe, E. G-networks with triggered customer movement. Journal of Applied Pr obability 1993 , pp. 742–748. 128. Hill, J.; Szewczyk, R.; W oo, A.; Hollar , S.; Culler , D.; Pister , K. System architectur e directions for networked sensors. ACM SIGOPS operating systems review . ACM, 2000, V ol. 34, pp. 93–104. 129. Pan, M.S.; T sai, C.H.; T seng, Y .C. Emergency guiding and monitoring applications in indoor 3D environments by wir eless sensor networks. International Journal of Sensor Networks 2006 , 1 , 2–10. 130. Gelenbe, E.; Schmajuk, N.; Staddon, J.; Reif, J. Autonomous search by robots and animals: A survey . Robotics and Autonomous Systems 1997 , 22 , 23–34. 131. Mitchell, T .M. Machine learning. 1997. Burr Ridge, IL: McGraw Hill 1997 , 45 . 132. John, H. Holland, Adaptation in natural and artiﬁcial systems, 1992. 133. Deb, K.; Agrawal, S.; Pratap, A.; Meyarivan, T . A fast elitist non-dominated sorting genetic algorithm for multi-objective optimization: NSGA-II. Lecture notes in computer science 2000 , 1917 , 849–858. 134. Eppstein, D. Finding the k shortest paths. SIAM Journal on computing 1998 , 28 , 652–673. 135. Saadatseresht, M.; Mansourian, A.; T aleai, M. Evacuation planning using multiobjective evolutionary optimization approach. European Journal of Operational Research 2009 , 198 , 305–314. 136. Pan, X.; Han, C.S.; Law , K.H. A multi-agent based simulation framework for the study of human and social behavior in egr ess analysis. Proceedings of the ASCE International Confer ence on Computing in Civil Engineering, 2005, V ol. 92. 137. Samadzadegan, F .; Y adegari, M. A BIOLOGICALL Y -INSPIRED OPTIMIZA TION ALGORITHM FOR URBAN EV ACUA TION PLANNING IN DISASTER MANAGEMENT . V ersion July 10, 2019 submitted to Electronics 31 of 33 138. Karaboga, D. An idea based on honey bee swarm for numerical optimization. T echn. Rep. TR06, Erciyes Univ . Press, Er ciyes 2005 . 139. Gelenbe, E.; Lent, R.; Xu, Z. T owards networks with cognitive packets. In Performance and QoS of next generation networking ; Springer London, 2001; pp. 3–17. 140. Gelenbe, E.; Lent, R.; Xu, Z. Design and performance of cognitive packet networks. Performance Evaluation 2001 , 46 , 155–176. 141. Gelenbe, E. Sensible decisions based on QoS. Computational management science 2003 , 1 , 1–14. 142. V ahdat, A.; Becker , D.; others. Epidemic routing for partially connected ad hoc networks. T echnical report, T echnical Report CS-200006, Duke University , 2000. 143. Gorbil, G.; Gelenbe, E. Resilience and security of opportunistic communications for emergency evacuation. Proceedings of the 7th ACM workshop on Performance monitoring and measurement of heterogeneous wireless and wir ed networks. ACM, 2012, pp. 115–124. 144. Gelenbe, E. Learning in the recurrent random neural network. Neural Computation 1993 , 5 , 154–164. 145. Desmet, A.; Gelenbe, E. A Parametric Study of CPN’s Convergence Process. Information Sciences and Systems 2014 - Proceedings of the 29th International Symposium on Computer and Information Sciences, ISCIS 2014, Krakow , Poland, October 27-28, 2014, 2014, pp. 13–20. 146. Bi, H. Routing Diverse Evacuees with the Cognitive Packet Network Algorithm. Future Internet 2014 , 6 , 203. 147. Akinwande, O.; Bi, H.; Gelenbe, E. Managing Crowds in Hazards W ith Dynamic Grouping. Access, IEEE 2015 , 3 , 1060–1070. 148. Olenick, S.M.; Carpenter , D.J. An updated international survey of computer models for ﬁre and smoke. Journal of Fire Pr otection Engineering 2003 , 13 , 87–110. 149. Koo, S.H.; Fraser-Mitchell, J.; Upadhyay , R.; W elch, S. Sensor-linked ﬁre simulation using a Monte-Carlo approach 2008 . 150. Murphy , K.P . Dynamic bayesian networks: repr esentation, inference and learning. PhD thesis, University of California, Berkeley , 2002. 151. Gelenbe, E.; Han, Q. Near-Optimal Emergency Evacuation with Rescuer Allocation. In Proceedings of the 4th International W orkshop on Pervasive Networks for Emergency Management (PerNEM’14). IEEE, 2014, pp. 1–6. 152. Rais, N.; Soh, P .J.; Malek, F .; Ahmad, S.; Hashim, N.; Hall, P . A review of wearable antenna. Antennas & Propagation Confer ence, 2009. LAPC 2009. Loughborough. IEEE, 2009, pp. 225–228. 153. Oreﬁce, M.; Pirinoli, P .; Dassano, G. Electrically-small wearable antennas for emergency services applications. Antenna T echnology (iW A T), 2016 International W orkshop on. IEEE, 2016, pp. 131–134. 154. Lilja, J.; Pynttäri, V .; Kaija, T .; Mäkinen, R.; Halonen, E.; Sillanpää, H.; Heikkinen, J.; Mäntysalo, M.; Salonen, P .; de Maagt, P . Body-worn antennas making a splash: Lifejacket-integrated antennas for global search and rescue satellite system. IEEE Antennas and Propagation Magazine 2013 , 55 , 324–341. 155. Li, S.; Zhan, A.; Wu, X.; Chen, G. ERN: Emer gence r escue navigation with wir eless sensor networks. Parallel and Distributed Systems (ICP ADS), 2009 15th International Confer ence on. IEEE, 2009, pp. 361–368. 156. Huang, J.H.; Amjad, S.; Mishra, S. CenW its: A Sensor-based Loosely Coupled Sear ch and Rescue System Using W itnesses. Proceedings of the 3rd International Conference on Embedded Networked Sensor Systems; ACM: New Y ork, NY , USA, 2005; SenSys ’05, pp. 180–191. 157. W u, W .; Cao, J.; Zheng, Y .; Zheng, Y .P . W AITER: A W earable Personal Healthcare and Emergency Aid System. 2008 Sixth Annual IEEE International Conference on Pervasive Computing and Communications (PerCom), 2008, pp. 680–685. 158. Bozkurt, A.; Roberts, D.L.; Sherman, B.L.; Brugar olas, R.; Mealin, S.; Majikes, J.; Y ang, P .; Loftin, R. T oward cyber-enhanced working dogs for sear ch and rescue. IEEE Intelligent Systems 2014 , 29 , 32–39. 159. Bozkurt, A.; Lobaton, E.; Sichitiu, M. A biobotic distributed sensor network for under-rubble sear ch and rescue. Computer 2016 , 49 , 38–46. 160. T adokoro, S. Earthquake Disaster and Expectation for Robotics. In Rescue Robotics ; T adokoro, S., Ed.; Springer London, 2009; pp. 1–16. 161. Davids, A. Urban search and rescue robots: from tragedy to technology . IEEE Intelligent systems 2002 , 17 , 81–83. V ersion July 10, 2019 submitted to Electronics 32 of 33 162. Barbera, J.A.; Macintyre, A. Urban search and rescue. Emergency medicine clinics of North America 1996 , 14 , 399–412. 163. Stopforth, R.; Holtzhausen, S.; Bright, G.; Tlale, N.; Kumile, C. Robots for Search and Rescue Purposes in Urban and Underwater Envir onments - a survey and comparison. Mechatronics and Machine V ision in Practice, 2008. M2VIP 2008. 15th International Conference on, 2008, pp. 476–480. 164. Casper , J.; Murphy , R.R. Human-robot interactions during the robot-assisted urban search and rescue response at the W orld T rade Center . Systems, Man, and Cybernetics, Part B: Cybernetics, IEEE T ransactions on 2003 , 33 , 367–385. 165. Doroodgar , B.; Liu, Y .; Nejat, G. A Learning-Based Semi-Autonomous Controller for Robotic Exploration of Unknown Disaster Scenes While Sear ching for V ictims. IEEE T ransactions on Cybernetics 2014 , 44 , 2719–2732. 166. Murphy , R.R. The 100 : 100 challenge for computing in rescue robotics. Safety , Security , and Rescue Robotics (SSRR), 2011 IEEE International Symposium on. IEEE, 2011, pp. 72–75. 167. Halász, A.; Hsieh, M.A.; Berman, S.; Kumar , V . Dynamic redistribution of a swarm of robots among multiple sites. Intelligent Robots and Systems, 2007. IROS 2007. IEEE/RSJ International Conference on. IEEE, 2007, pp. 2320–2325. 168. Berman, S.; Halász, Á.; Hsieh, M.A.; Kumar , V . Optimized stochastic policies for task allocation in swarms of robots. IEEE T ransactions on Robotics 2009 , 25 , 927–937. 169. Iwano, Y .; Osuka, K.; Amano, H. Proposal of a Rescue Robot System in Nuclear-Power Plants -Rescue Activity via Small V ehicle Robots -. Robotics and Biomimetics, 2004. ROBIO 2004. IEEE International Conference on, 2004, pp. 227–232. 170. V entura, R.; Lima, P .U. Search and rescue robots: The civil protection teams of the future. Emerging Security T echnologies (EST), 2012 Third International Conference on. IEEE, 2012, pp. 12–19. 171. Jacoff, A.; Messina, E.; W eiss, B.; T adokoro, S.; Nakagawa, Y .; others. T est arenas and performance metrics for urban search and r escue robots. Intelligent Robots and Systems, 2003.(IROS 2003). Proceedings. 2003 IEEE/RSJ International Conference on. IEEE, 2003, V ol. 4, pp. 3396–3403. 172. V entura, R.; Lima, P . Search and Rescue Robots: The Civil Protection T eams of the Future. Emerging Security T echnologies (EST), 2012 Third International Conference on, 2012, pp. 12–19. 173. T adokoro, S.; Kitano, H.; T akahashi, T .; Noda, I.; Matsubara, H.; Joh, A.S.; Koto, T .; T akeuchi, I.; T akahashi, H.; Matsuno, F .; others. The robocup-r escue project: A robotic approach to the disaster mitigation problem. Robotics and Automation, 2000. Proceedings. ICRA ’00. IEEE International Conference on. IEEE, 2000, V ol. 4, pp. 4089–4094. 174. Gautam, A.; Mohan, S. A review of resear ch in multi-r obot systems. Industrial and Information Systems (ICIIS), 2012 7th IEEE International Conference on. IEEE, 2012, pp. 1–5. 175. Gelenbe, E.; Cao, Y . Autonomous search for mines. European Journal of Operational Research 1998 , 108 , 319–333. 176. Landis, G.A. Robots and humans: synergy in planetary exploration. Acta astronautica 2004 , 55 , 985–990. 177. Pugh, J.; Martinoli, A. Inspiring and Modeling Multi-Robot Sear ch with Particle Swarm Optimization. Swarm Intelligence Symposium, 2007. SIS 2007. IEEE, 2007, pp. 332–339. 178. Zhu, Q.; Liang, A.; Guan, H. A PSO-inspired multi-robot search algorithm independent of global information. Swarm Intelligence (SIS), 2011 IEEE Symposium on. IEEE, 2011, pp. 1–7. 179. Sutantyo, D.K.; Kernbach, S.; Levi, P .; Nepomnyashchikh, V .; others. Multi-robot sear ching algorithm using lévy ﬂight and artiﬁcial potential ﬁeld. Safety Security and Rescue Robotics (SSRR), 2010 IEEE International W orkshop on. IEEE, 2010, pp. 1–6. 180. Baranzadeh, A.; Savkin, A.V . A distributed algorithm for grid-based search by a multi-robot system. Control Confer ence (ASCC), 2015 10th Asian. IEEE, 2015, pp. 1–6. 181. Cavalcante, R.C.; Noronha, T .F .; Chaimowicz, L. A Local Search Approach for Improving Multi-Robot Routing in Exploration Missions. Robotics Symposium and Latin American Robotics Symposium (SBR-LARS), 2012 Brazilian. IEEE, 2012, pp. 85–90. 182. Marjovi, A.; Nunes, J.G.; Marques, L.; De Almeida, A. Multi-robot exploration and ﬁr e searching. Intelligent Robots and Systems, 2009. IROS 2009. IEEE/RSJ International Conference on. IEEE, 2009, pp. 1929–1934. 183. Burgar d, W .; Moors, M.; Stachniss, C.; Schneider , F .E. Coordinated multi-robot exploration. IEEE T ransactions on robotics 2005 , 21 , 376–386. 184. Gelenbe, E. Search in unknown random environments. Physical Review E 2010 , 82 , 061112. V ersion July 10, 2019 submitted to Electronics 33 of 33 185. Lau, H.Y .; Ko, A. An immuno robotic system for humanitarian sear ch and rescue (application stream). In Artiﬁcial Immune Systems ; Springer , 2007; pp. 191–203. 186. Su, X.; Zhang, M.; Bai, Q. Dynamic task allocation for heterogeneous agents in disaster environments under time, space and communication constraints. The Computer Journal 2015 , 58 , 1776–1791. 187. Kang, R.J.; Müller , T . Sphere and dot product repr esentations of graphs. Discrete & Computational Geometry 2012 , 47 , 548–568. 188. Gelenbe, E.; Hussain, K.; Kaptan, V . Simulating autonomous agents in augmented reality . Journal of Systems and Software 2005 , 74 , 255–268. 189. Nourbakhsh, I.R.; Sycara, K.; Koes, M.; Y ong, M.; Lewis, M.; Burion, S. Human-robot teaming for sear ch and rescue. Pervasive Computing, IEEE 2005 , 4 , 72–79. 190. Gelenbe, E. The ﬁrst decade of G-networks. European Journal of Operational Research 2000 , 126 , 231–232. 191. A versa, R.; Di Martino, B.; Rak, M.; V enticinque, S. Cloud agency: A mobile agent based cloud system. Complex, Intelligent and Softwar e Intensive Systems (CISIS), 2010 International Conference on. IEEE, 2010, pp. 132–137. 192. Gelenbe, E.; Loukas, G. A self-aware approach to denial of service defence. Computer Networks 2007 , 51 , 1299–1314. 193. Gelenbe, E. Energy packet networks: adaptive energy management for the cloud. Proceedings of the 2nd International W orkshop on Cloud Computing Platforms. ACM, 2012, p. 1. 194. Gelenbe, E. Adaptive management of energy packets. Computer Software and Applications Conference W orkshops (COMPSACW), 2014 IEEE 38th International. IEEE, 2014, pp. 1–6. 195. Gelenbe, E. A sensor node with energy harvesting. ACM SIGMETRICS Performance Evaluation Review 2014 , 42 , 37–39. c  2019 by the authors. Submitted to Electronics for possible open access publication under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.or g/licenses/by/4.0/).

Emergency Management Systems and Algorithms: a Comprehensive Survey

Original Paper

Comments & Academic Discussion

Leave a Comment

Original Paper

Related Papers

Comments & Academic Discussion

Leave a Comment