Reinforcement Learning-Based Energy-Aware Coverage Path Planning for Precision Agriculture
This work proposes an energy-aware reinforcement learning framework for coverage path planning in agricultural robotics, addressing the frequent task failures caused by neglecting energy constraints. The approach uniquely integrates Soft Actor-Critic (SAC) with a CNN-LSTM architecture, where the CNN extracts spatial environmental features and the LSTM models temporal dynamics. A multi-objective reward function is designed to jointly optimize coverage rate, energy consumption, and return-to-charging constraints. Evaluated in grid environments with obstacles and charging stations, the method achieves over 90% coverage—outperforming baseline algorithms such as RRT, PSO, and ACO by 13.4–19.5%—while reducing constraint violation rates by 59.9–88.3%. These results demonstrate a significant improvement in both energy-safe coverage efficiency and environmental adaptability.