Research on Reinforcement Learning-Based Order Control Methods for Multi-Handling Robots

Autonomous mobile robots carrying people in large public environments not only have to solve the obstacle avoidance but also face the challenge of passage ordering. To this end, an augmented exploration MADDPG control algorithm for manned access is investigated on the basis of the proposed basic rules for manned access. Firstly, the 3D radar localization and 2D radar obstacle information collected by the robot are processed by data fusion to form the input spatial matrix for reinforcement learning, and the motion control speed and angular velocity of the manned robot are used as outputs to be trained in the Gazebo environment, and secondly, a model of decision-making ability training for multi-mobile robots facing complex environments is completed by staged training and randomly generating the positions of obstacles, and the model is generated. The model is invoked on the manned robot to ensure the motion control of the manned robot and at the same time to ensure the orderliness of multiple manned robots. Simulations and physical experiments verify the effectiveness of the proposed method.