Joint Optimization of Perception Offloading and Planning Performance of Autonomous Systems Over Wireless Networks

Perception and planning are critical components of autonomous systems, where the perception module supplies essential data for the planning module to generate effective actions. Achieving high perception accuracy, however, often incurs substantial resource consumption. Thus, optimizing the perception module's configuration is essential to balance accuracy, resource constraints, and the requirements of the planning module. To address the computational demands of perception, we propose a novel approach that dynamically decides whether to offload perception subtasks to an edge server or process them locally. This decision-making accounts for the dynamic conditions of the wireless network, ensuring efficient energy usage while meeting planning objectives. By jointly optimizing perception resource allocation and planning performance, our method strikes a trade-off between energy consumption and planning effectiveness. Simulation results demonstrate that the proposed joint optimization approach significantly outperforms traditional methods that decouple perception and planning configurations.