Foot type robot motion control method and system based on deep reinforcement learning

Abstract

The invention discloses a foot type robot motion control method and system based on deep reinforcement learning. The foot type robot motion control method based on the deep reinforcement learning comprises the steps that a foot type robot 3D model is established; reward and punishment functions are designed; a motion network and a target motion network are established, and network initialization is completed; controlled quantity is generated by using the motion network to obtain a state of a robot at a next moment, and a reward and punishment value is calculated; a certain number of samples are selected randomly, a state-motion value of a target evaluation network is calculated, and output of the evaluation network is updated according to a Bellman equation; a weight of the motion networkis updated by using the evaluation network; the target evaluation network and the target motion network are updated by using the evaluation network and the weight of the motion network; the steps arerepeated until network convergence; and a control instruction of the robot motion is obtained according to the motion network. According to the foot type robot motion control method and system based on the deep reinforcement learning, a foot type robot is enabled to realize the high-efficiency and stable movement under unknown environments.

Description

technical field [0001] The invention belongs to the technical field of motion control of legged robots, and in particular relates to a method and system for motion control of legged robots based on deep reinforcement learning. Background technique [0002] Footed robots have the characteristics of multi-joints, strong nonlinearity, and multiple motion modes. At the same time, in the unstructured environment of extraterrestrial bodies, information such as surface materials, stiffness, and terrain are missing or inaccurate, which gives the footed robots motion control. In particular, high-performance motion control that takes into account both fastness and stability poses a huge challenge. The classic model-based motion control method is highly dependent on model accuracy, the design process is complex, it is difficult to deal with high-dimensional objects, and it cannot give full play to the mobility of the robot, especially the adaptability to model uncertainty and environme...

AI Technical Summary

Problems solved by technology

[0002] Footed robots have the characteristics of multi-joints, strong nonlinearity, and multiple motion modes. At the same time, in the unstructured environment of extraterrestrial bodies, information such as surface materials, stiffness, and terrain are missing or inaccurate, which gives the footed robots motion control. In particular, high-performance motion control that takes into account both fastness and stability poses a huge challenge

The classic model-based motion control method is highly dependent on model accuracy, the design process is complex, it is difficult to deal with high-dimensional objects, and it cannot give full play to the mobility of the robot, especially the adaptability to model uncertainty and environmental unknowns is very limited. Difficulty coping with unknown unstructured environments on the surface of extraterrestrial objects

Images