Robot autonomous control method based on graph neural network reinforcement learning

Abstract

The invention discloses a robot autonomous control method based on graph neural network reinforcement learning. The robot autonomous control method comprises the following steps: 1, establishing a feature map of a robot according to information fed back to a robot sensor by an environment and a physical structure of the robot; 2, inputting the obtained feature map into a graph neural network, andtraining the graph neural network by using supervision information sensed by the robot in a training stage; 3, processing the perception state information by using the trained graph neural network toobtain updated robot state information, and predicting the state information of the robot at the next moment; and 4, using the established environment model and a model prediction control algorithm tomake a decision on the robot action in the next step. According to the method, the reinforcement learning model based on the graph network is used, actions which cannot be well completed by previousreinforcement learning are successfully completed, the stability and reliability of robot autonomous control are improved, and the robot autonomous control is more convenient for practical application.

Description

technical field [0001] The invention relates to the field of robot simulation and intelligent control, in particular to a robot autonomous control method based on graph neural network reinforcement learning. Background technique [0002] Robot automatic control refers to a technology in which the robot can perform forward, turning, obstacle avoidance and other actions without human intervention, so that the robot can realize intelligent autonomous control to a certain extent. The robot perceives its own state through the sensors carried by itself in the environment, and through the intelligent program compiled in advance, it makes autonomous action decisions according to the current state. The traditional autonomous control of robots generally performs preset feedback operations on the values ​​of specific sensors, and cannot achieve true autonomous control for more complex tasks. [0003] In order to solve the autonomous control of the robot under complex tasks, researcher...

AI Technical Summary

Problems solved by technology

This method uses a reinforcement learning method, combined with the feature extraction of the relationship between elements native to the graph neural network, to solve the problem of robot autonomous control

Images