Method of Multi-agent Path Planning for Lunar Robots Swarm Based on Improved CBS Algorithm
Haolong Feng, Songtai Wu, Shengyang Liu, Ting Song, Fei Han · 2024
As lunar missions become increasingly complex, the mission terrain will vary depending on different operating areas. In the future, there may be multiple lunar rovers working together, a more flexible and efficient algorithm to generate mission paths for lunar rover clusters is required. To address the path planning problem of lunar rover groups in large-scale and complex lunar terrain, this paper constructs a gridded map based on lunar image information and the actual lunar surface image is woven into a ring map. Path planning algorithms are modeled on this map, and the actual paths are projected onto the original lunar surface and into missions for implementing multi-lunar rover path planning tasks based on the conflict based search(CBS) algorithm. As the key to CBS algorithm is to resolve conflicts among agents, this paper combines the Deep Q Network with the CBS algorithm to propose a CBS algorithm based on a DQN conflict choice strategy to solve the multi-agent path planning problem for lunar rover clusters. The DQN can learn the policy of an agent, enabling it to better select paths. By integrating DQN with the CBS algorithm, path planning can become more intelligent and efficient. Finally, the simulation verifies that the CBS algorithm can effectively solve the multi-agent path planning problem and the improved CBS algorithm proposed has a certain degree of improvement in terms of solution efficiency.