Intelligent agent reinforcement learning method and device for controlling robot dog to advance

• Main Authors: WANG YUANBIN, WANG ZHENJIE, LIU JUNTAO, REN WEILIN, GAO ZIWEN
• Format: Patent
• Language: Chinese; English
• Published: 18.07.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention discloses an intelligent agent reinforcement learning method for controlling a robot dog to advance, and the method comprises the steps: controlling an intelligent agent to receive the position and motion information of each component of the robot dog as an input state in an execution process, calculating and outputting a control parameter of each joint as an action, and corresponding reward values are fed back according to the movement behaviors and the advancing distance of the robot dog, and the training target is to realize rapid and smooth advancing of the robot dog and avoid falling down. According to the method, entropy information is introduced into updating of the intelligent agent Critic, so that the evaluation network considers the stability degree of state change in an optimization target, the prediction and control capability of the intelligent agent to the environment is improved, and the learning stability is improved. The invention further provides a corresponding agent reinforce