UAV assisted truck delivery route optimization via reinforcement learning

With the rapid growth of e-commerce, traditional logistics faces the "last mile" delivery problem, characterized by low efficiency and high costs. This paper models the truck-drone cooperative delivery problem as the TSPD problem, using a Markov decision process, and proposes a deep reinforcement learning-based encoder-decoder framework to solve it. The improved DFPG algorithm, based on A2C, is employed for training. During training, a multi-attention encoder is used to embed node information, and the xLSTM fusion attention mechanism decoder is employed for path optimization. The attention mechanism is redefined to embed unselected truck and drone nodes as context, reducing invalid path selections. Experimental results show that the proposed method performs well on TSPD and TSPDCSR datasets of varying customer sizes, particularly on large-scale datasets, where it outperforms two other models, confirming its effectiveness and superiority