A Deep Reinforcement Learning-Based Method for Lawn Full Coverage Path Planning

Efficient lawn full coverage path planning is essential for improving coverage efficiency and path quality in automated lawn maintenance tasks. However, conventional coverage methods may produce redundant traversal and frequent turns in lawn environments with complex obstacle distributions. To address these limitations, this study proposes an improved Deep Q-Network (DQN)-based method for lawn full coverage path planning. A multi-component reward function is incorporated into the proposed algorithm to comprehensively consider coverage efficiency, motion smoothness, and boundary-aware exploration. In addition, Double DQN, prioritized experience replay, Noisy Networks, and Distributional DQN are integrated into a unified learning framework to improve value estimation stability, sample utilization efficiency, exploration capability, and action-value representation. The proposed method was validated using simulation experiments. The results show that the proposed method achieves stable convergence and generates more compact and smoother coverage paths. Ablation experiments verify the contribution of each improved module, while reward function validation confirms that the proposed reward design effectively reduces path length and unnecessary turns. Comparative experiments further demonstrate that the proposed method outperforms other coverage algorithms in terms of repetition rate, number of turns, and overall coverage quality.