Multiline Dispatching-Based Path Optimization for Metro-Connected Customized Autonomous Shuttle Services

To address the challenges of last-mile connectivity between residential communities and metro stations, this study investigates the multiobjective routing and scheduling problem of metro-feeder customized autonomous shuttles under cross-line operations. We formulate a biobjective model that jointly minimizes passenger-side cost (waiting and in-vehicle time) and operator-side cost (energy consumption and cross-line penalties) under reservation-based time windows and vehicle-capacity constraints. A structure-enhanced Nondominated Sorting Genetic Algorithm II (NSGA-II) framework is developed, featuring a dual-segment chromosome encoding, a multitrip dispatch decoder, and a cross-line redispatch mechanism. Using a real reservation data set of 28 passengers from a morning-peak feeder scenario, the proposed method yields a representative dispatch plan in which all requests are served with an average waiting time of 2.43 min, average in-vehicle time of 9.54 min, and an operational cost of 47.8 CNY per planning horizon. Ablation experiments show that disabling the multitrip mechanism reduces the service rate from 100% to 71.43%, and removing cross-line penalties leads to excessive cross-line trips and longer routes. Comparative tests against two dispatch heuristics and a multiobjective evolutionary algorithm based on decomposition (MOEA/D) baseline on instances scaled up to 40, 60, and 80 passengers demonstrate that the proposed NSGA-II produces better-spread Pareto fronts and consistently lower operating costs; in the 80-passenger case, it reduces the operational cost from 81.49 CNY (MOEA/D) to 56.14 CNY while maintaining similar passenger costs and full coverage. These results confirm the scalability and practical potential of the proposed cross-line autonomous shuttle scheduling framework for smart urban transit systems.