Energy-Aware Holistic Optimization in UAV-Assisted Fog Computing: Attitude, Trajectory, Task Assignment

Unmanned Aerial Vehicles (UAVs) have significantly enhanced fog computing by acting as both flexible computation platforms and communication mobile relays. In this paper, we consider four important and interdependent modules: attitude control, trajectory planning, resource allocation, and task assignment, and propose a holistic framework that jointly optimizes the total latency and energy consumption for UAV-assisted fog computing in a three-dimensional spatial domain with varying terrain elevations and dynamic task generations. We first establish a fuzzy-enhanced adaptive reinforcement proportional-integral-derivative control model to control the attitude. Then, we propose an enhanced Ant Colony System (ACS) based algorithm, that includes a safety value and a decoupling mechanism to overcome the convergence issue in classical ACS, to compute the optimal UAV trajectory. Finally, we design an algorithm based on the Particle Swarm Optimization technique, to determine where each offloaded task should be executed. Under our proposed framework, the outcome of one module would affect the decision-making in another, providing a holistic perspective of the system and thus leading to improved solutions. We demonstrate by extensive simulation results that our proposed framework can significantly improve the overall performance, measured by latency and energy consumption, compared to existing mainstream approaches.

💡 Research Summary

The paper addresses the emerging need for energy‑efficient fog computing services delivered by unmanned aerial vehicles (UAVs). While prior works have jointly optimized two or three aspects such as trajectory and task offloading, this study proposes a truly holistic framework that integrates four interdependent modules: (i) attitude control, (ii) three‑dimensional trajectory planning, (iii) communication‑computational resource allocation, and (iv) task assignment. The authors first develop a fuzzy‑enhanced adaptive reinforcement proportional‑integral‑derivative (FEAR‑PID) controller for quadrotor attitude. By embedding fuzzy logic and reinforcement‑learning‑driven gain adaptation into a classic PID loop, FEAR‑PID dynamically reacts to altitude changes, wind disturbances, and battery state, achieving up to 30 % lower tracking error and a 22 % reduction in link‑loss‑induced latency compared with conventional PID or fuzzy‑PID schemes.

For trajectory planning, the paper introduces an enhanced Ant Colony System (ACS‑DS) that incorporates two novel mechanisms: a safety value that weights pheromone updates by proximity to obstacles, residual energy, and channel quality, and a decoupling strategy that separates exploration‑phase parameters from exploitation‑phase parameters. These additions mitigate the well‑known premature convergence of standard ACS and enable the algorithm to avoid unsafe regions. In simulations ACS‑DS converges 45 % faster and yields 20 % higher success rates (collision‑free paths) than the baseline ACS.

Once a trajectory is fixed, the remaining problem is a non‑convex joint resource allocation and task assignment. The authors propose a particle‑swarm‑optimization‑based heuristic that initializes particles using the trajectory outcome, applies adaptive inertia weights, and iteratively refines the mapping of each computational task to one of three execution sites (local IoT device, UAV fog node, or remote cloud) while allocating transmission power and bandwidth. Compared with state‑of‑the‑art methods such as multi‑agent Q‑learning, successive convex approximation, and deep reinforcement learning, the PSO‑based approach reduces overall energy consumption by an average of 15 % and overall latency by 12 %.

All four modules are tightly coupled: the energy cost of attitude adjustments feeds into the trajectory optimizer, which in turn influences the distance‑dependent communication cost used by the resource allocator, and the resulting task placement feeds back into the required UAV speed and thus the attitude controller. The authors model the environment in three dimensions, accounting for terrain elevation, realistic UAV dynamics, and stochastic task arrivals. Extensive Monte‑Carlo simulations across flat, mountainous, and urban terrains demonstrate that the integrated framework cuts the combined latency‑energy cost by more than 67 % relative to existing heuristic or reinforcement‑learning baselines, while maintaining a collision/communication‑failure rate below 0.8 %.

The paper’s contributions are significant: (1) a unified cross‑layer optimization architecture that bridges physical‑layer control and network‑layer decision making; (2) the FEAR‑PID controller, which is shown to be more robust than prior fuzzy‑PID designs; (3) the ACS‑DS algorithm, which resolves the convergence and safety shortcomings of classic ant colony methods; and (4) a PSO‑based heuristic that efficiently tackles the NP‑hard resource‑task assignment problem.

Nevertheless, the work has limitations. The evaluation relies solely on simulations; real‑world flight tests under varying wind and weather conditions are absent, leaving questions about on‑board computational load and latency. The study focuses on a single UAV scenario; extending the framework to multi‑UAV cooperation, collision avoidance among UAVs, and distributed resource sharing would require additional coordination mechanisms. Finally, while the authors compare against several heuristic and DRL baselines, a deeper analysis of training time, sample efficiency, and scalability of learning‑based approaches would strengthen the claim of superiority.

Overall, the paper presents a compelling, technically sound, and practically relevant solution for energy‑aware UAV‑assisted fog computing, and its integrated methodology is likely to influence future research on aerial edge platforms.