MAPF scenario modelling for human-robot collaboration

With the rapid development of artificial intelligence, the multi-agent pathfinding problem (MAPF) has emerged in recent years. It is regarded as an NP-Hard problem that involves coordinating the movements of multiple agents within the same environment to perform different tasks. In recent years, MAPF has found numerous applications in automated industrial scenarios, prompting the proposal of several learning-based methods to solve MAPF challenges. In fact, in industrial application scenarios, the completion of complex tasks can be hampered by the robots' limited observational abilities to perceive their environment. Therefore, it is necessary to combine the flexibility of humans with intelligent manufacturing to achieve high levels of flexibility, efficiency, and safety through human-machine collaboration. This study proposes a model for human-machine collaboration in MAPF scenarios, based on the characteristics of human-machine interactions. The effectiveness of a multi-agent reinforcement learning algorithm in addressing dynamic MAPF problems in human-machine collaboration is verified. Finally, building on the classical multi-agent reinforcement learning algorithm MAA2C, the A*-MAA2C algorithm is proposed. Ablation experiments have verified its superiority over the standard MAA2C algorithm.