Multi-pass planning for multi-vehicle cooperative urban demining: A knowledge-driven evolutionary approach with RL-enhanced neighborhood search

It has become increasingly urgent and necessary to coordinate multiple unmanned systems to efficiently execute a variety of complex tasks in place of humans. This paper focus on the problem of multi-vehicle demining in urban road networks (MVDMP). First, a mixed-integer programming model is established, taking into account both the topological connectivity of the road network and the demining width of the vehicles. Second, an evolutionary learning algorithm incorporating Q-learning (QEA) is proposed to effectively solve this problem. In the initialization phase, a hybrid initialization strategy, which includes two heuristic rules, is introduced to generate high-quality initial solutions. During the local search phase, six neighborhood search operators are proposed based on problem characteristics, and Q-learning is used to adaptively customize perturbation schemes for individuals. Additionally, the Metropolis acceptance criterion is employed to balance exploration and exploitation. Finally, extensive experiments on instances of varying sizes derived from urban road networks (Sioux Falls, Sydney, etc.) demonstrate the efficiency and superiority of the proposed method compared to other four state-of-the-art approaches.