Research And Implementation Of Centralized Motion Coordination Strategy For Multiple Objects In Specific Road Network Environment Based On Reinforcement Learning Algorithm

In recent years,the application of agent technology has enabled humans to avoid many repetitive and dangerous tasks.Due to the increasing complexity of the task,a single agent cannot meet the demand.The multi-agent system emerges and is applied to many fields such as industry,military,aerospace and so on.In the multi-agent system,the mobile agent system is one of the most widely used branches.It accomplishes the task through collaborative work among the agents,while the multiple agents may collide during the movement.How to coordinate the movement of the multiple agents is one of the important research topics of the mobile agent system.In a two-dimensional scene including a road network,with the vehicle-based mobile robot as the research object,the research goal of the thesis is to use the reinforcement learning method to generate a motion coordination strategy which can ensure multiple robots without collision during the movement and the overall movement time is as short as possible.Aiming at the multi-robot motion coordination problem in specific road network environment,the thesis proposes a multi-robot centralized motion coordination reinforce-ment learning algorithm which combines with the Double Deep Q-Network(DDQN)method.Firstly,the thesis use scene division and collision detection algorithm based on the rectangular bounding box to obtain the collision region between robot paths.Then the path segmentation and transformation method designed in the thesis is used to convert collision area and path set into a specific path checkerboard graph model,which provides an environment for interacting with the agent in the reinforcement learning training.Finally,the state space of the environment,the action space of the agent and the reward model of the environment are designed.And iterative training is carried out in the environment of path checkerboard graph model by using DDQN method to obtain a feasible motion coordination strategy.In order to verify the correctness and robustness of the proposed algorithm,the thesis builds a related experimental platform using the PyQt application framework in the Pycharm development environment.A large number of simulation experiments show that the proposed algorithm in the thesis can effectively solve the collision problem in multi-robot motion process,so it has practical application value.