FedTD3: An Accelerated Learning Approach for UAV Trajectory Planning

Uncrewed Aerial Vehicle (UAV) trajectory planning has been realized to have a significant impact on precision agriculture in enabling more efficient monitoring of crops and soil through optimized flight paths and data collection. While current learning-based algorithms may yield promising results, their training process and the system’s highly dynamic channel cause these algorithms to be extremely slow. To address this issue, we design a learning acceleration framework with an efficient algorithm, FedTD3. Our main contributions include a channel model that characterizes UAV-GS (Ground Sensors) communications features in rural areas, a quadrotor UAV energy consumption model for its movement in any direction, and FedTD3–an accelerated RL solver to deal with the problem of time efficiency and sustainability. We perform thorough evaluations to validate the algorithm’s performance. The results show that our design can achieve a significant speedup.