In order for close proximity satellites to safely perform their missions, the relative states of all satellites and pieces of debris must be well understood. This presents a problem for ground based tracking and orbit determination since it may not be practical to achieve the required accuracy. Using space-based sensors allows for more accurate relative state estimates, especially if multiple satellites are allowed to communicate. Of interest to this work is the case where several communicating satellites each need to maintain a local catalog of communicating and non-communicating objects using angles-only limited field of view (FOV) measurements. However, this introduces the problem of efficiently scheduling and coordinating observations among the agents. This paper presents a decentralized task allocation algorithm to address this problem and quantifies its performance in terms of fuel usage and overall catalog uncertainty via numerical simulation. It was found that the new method significantly outperforms the uncertainty-fuel Pareto frontier formed by current approaches.
As the number of objects in LEO continues to grow, so too does the importance of maintaining accurate and up to date state information about neighboring objects. This information enables operators (or the satellites themselves) to make more informed decisions about collision avoidance maneuvers. Additionally, it allows for a wider variety of close proximity mission sets such as orbital construction, inspection, and formation flying [1,2,3].
One option for addressing the need for more accurate relative state estimates is by using a collection of communicating agents equipped with limited FOV angles-only sensors. Each agent is tasked with maintaining its own local catalog of the states and covariances of objects within the local operating environment. The goal is that the uncertainty (some function of the covariance) of all objects is below a predefined threshold in each catalog. This approach presents several subproblems, namely control of each agent, distributed state estimation, construction of the communication graph, and observation task allocation. These problems and potential solutions are studied in Hays et al. [4]. The specific contribution of this paper is to improve upon the task allocation algorithm from that work in terms of fuel expenditure and the ability to maintain the uncertainty of all objects below a desired threshold. Additionally, the new algorithm does not require a centralized supervisor meaning that it is more robust and reduces communication requirements.
With the increasing importance of autonomous multi-agent systems like satellites and UAVs, there has been a corresponding interest in efficient and effective task allocation algorithms for these systems. For instance, Choi et al. [5] presents a general-purpose algorithm for decentralized task allocation of multi-agent systems called the Consensus-Based Bundle Algorithm (CBBA). Several follow-up works have looked at modifications to this algorithm that make it better suited for dynamic environments. Johnson et al. [6] presents an asynchronous version of the algorithm (called ACBBA) that is much more practical for real multi-agent systems. Additionally, Buckman et al. [7] focused on allowing the agents to incorporate a new task into their plans without completely re-planning. There have also been several studies that demonstrate the effectiveness of this algorithm for multiagent systems via simulation. For instance, variants of the CBBA approach were tested for UAVs swarms [8,9]. Its performance has also been assessed for Earth observation missions with satellites [10,11,12]. All of these studies used dynamic or complex tasks of some form. These include timevarying task locations and values, heterogeneous agents and tasks, limited time availability of tasks, and dependencies between tasks. To the best of the authors’ knowledge, the feasibility of the CBBA algorithm for the catalog maintenance problem has not yet been assessed. However, given that the problem has some of the complexities looked at in these studies, the CBBA algorithm represents a promising research direction.
There are several factors that add to the complexity of designing a task allocation algorithm for the local catalog maintenance problem. First and foremost is that the states and covariances of all objects are time-varying. Any algorithm scheduling future tasks either needs to simulate the state of the entire catalog of objects forward in time or assume that changes occur slowly enough to be considered quasi-static for planning purposes. Additionally, it is not obvious when an observation task should be considered “complete.” The naive approach calls a task complete as soon as that object falls below the uncertainty threshold. However, because uncertainty monotonically increases with time, tasks that were just completed would quickly rise back above the threshold and need follow-up observations. Finally, because there is communication between the agents, when one agent observes an object, its uncertainty will decrease for all of the agents. This means the observation tasks assigned to each agent will affect the quality of observation tasks assigned to other agents. All of these issues are addressed with the algorithm developed in this work. Every time a target is observed, a modified version of CBBA plans a small number of observations into the future based on the current state of the system. This frequently revises the plans of the agents based on the current state of the system and the tasks chosen by other agents. Observation tasks are considered complete when the rate of decrease of a new observation score function falls below a certain value.
Let R be the set of real numbers and N be the set of natural numbers. Next, let R n be the set of n dimensional real-valued column vectors, and R n×m be the set of real-valued matrices in with n rows and m columns. The matrix I n ∈ R n×n is the n-dimensional identity matrix. Given a vector x H expressed in frame H, i
This content is AI-processed based on open access ArXiv data.