Reinforcement Learning-Based Multi-Objective Route Planning for Collaborative Control of Group Robots

摘要

Traditional path planning algorithms are often designed for one single objective, making it difficult to effectively address multi-objective scenarios in collaborative control of group robots. As a result, we propose a reinforcement learning-based multi-object route planning approach for this purpose in complex environment. It first analyzes the motion trajectory and positioning method of group robots, and constructs a basic multi-objective path planning method for group robots. Then, it designs obstacle avoidance paths for group robots through reinforcement learning. In this method, we set reinforcement learning parameters based on the path of group robots, complete adaptability analysis using constraint functions, and finally optimize the path through evaluation. By comparing with several typical methods, some experiments were conducted to evaluate the proposed algorithm. The experimental results show that when the iteration number is 5, the constraint function value of the algorithm in this study is 832, while the values of other algorithms are 326 (AGV algorithm), 373 (Multi-UAV algorithm), and 493 (MRDWA-MADDPG algorithm), respectively. Taking into account other experimental results, it can be seen that the algorithm in this study performs the best at this iteration number. In summary, this study can provide theoretical and technical support for group robots to efficiently collaborate and complete multi-objective tasks in complex environments.