Multi-Objective Path Planning for Unmanned Sweepers Considering Traffic Signals: A Reinforcement Learning-Enhanced NSGA-II Approach

With the widespread popularization of unmanned sweepers, path planning has been recognized as a key component affecting their total work efficiency. Conventional path planning methods often only aim to improve work efficiency while ignoring energy optimization, a crucial factor for sustainable development. In this paper, an energy- and time-minimization unmanned sweeper arc path problem (ETM-ARP) is investigated, and the effects of road slope, dynamic changes in on-board mass, mode switching of vehicle work, and traffic lights are taken into consideration to meet the requirements of a realistic structured road scenario. A new multi-objective mixed-integer nonlinear planning model is proposed for this problem. To solve this model, we propose a deep Q-network (DQN) and Adaptive Large Neighborhood Search Algorithm (ALNS)-driven non-dominated sorting genetic algorithm II (QALNS-NSGA-II). The novelty of this algorithm lies in integrating DQN into ALNS, to guide high-quality adaptive operator selection during the search process based on additional information. The computational results of various examples confirm the effectiveness of the proposed method. The proposed method can be used to improve the efficiency and sustainability of unmanned sweepers for sweeping on structured roads.