Research on multi-robot path planning based on deep reinforcement learning
Guicheng Shen, Yaxuan Cheng, Zhong Tang, Tianyun Qiu, Juntao Li · 2023
Multi-robot path planning (MRPP) enables the robot teams complete the tasks quickly and efficiently without collision. Based on the domestic and foreign research, the mainstream centralized algorithms for path planning have many shortages, so IL-A3C algorithm, which is based on the Dec-POMDP model, is proposed to solve these shortages. Then, the performance of IL-A3C is computed in terms of average path planning length, average path planning time, average collision probability and average planning success rate in different dimensions, and the simulation result shows that IL-A3C performs very well under low obstacle density and can be easily extended to a team of 128 robots. After that, IL-A3C is also compared with the centralized algorithm CBS and A3C algorithm, and the comparison proves that IL-A3C has higher success rate, stronger scalability and stability than CBS and A3C. The conclusion is that IL-A3C can be easily scaled to a large-scale robot team.