The express delivery industry in China is driven by the rapid expansion of internet-based e-commerce. However, significant resource waste and efficiency bottlenecks are observed in the last-mile delivery phase. To address the inefficiencies caused by suboptimal path planning in last-mile logistics, an improved grey wolf optimization algorithm(AAVHCGWO) was proposed. This algorithm integrates logistic mapping, ALNS(adaptive large neighborhood search), an adaptive Metropolis criterion, and a crisscross strategy. Ablation experiments and real-world case studies were conducted to validate the algorithm. Significant improvements in both solution effectiveness and computational efficiency are demonstrated by AAVHCGWO, compared with the original algorithm, across simulation experiments of varying dimensions. Practical delivery trials were performed using the IGA(improved genetic algorithm), GO(growth optimizer), and SOA(starling optimization algorithm). Reductions in delivery distances of 69.0%, 54.0%, and 19.2% were observed in practical trials compared to IGA, GO, and SOA. The effectiveness and superiority of AAVHCGWO are confirmed in both simulated and real-world logistics optimization. This study is considered a novel strategy for enhancing last-mile delivery efficiency.