Humanoid robot gait planning deep reinforcement learning new method

Abstract

The invention relates to an improvement on the robot technology, in particular to a humanoid robot gait planning deep reinforcement learning new method. The problem of walking control of a biped robotis solved through an improvement DQN algorithm. On the basis that there is no need to establish a complex biped robot dynamic model, the deep reinforcement learning method is used for realizing long-distance stable gait control of the biped robot under the condition of walking fast. The method comprises the steps of establishing of a humanoid biped robot model, pre-training control parameters andwalking movement of the biped robot, wherein a simplified six-degree-of-freedom connecting rod model is adopted as the humanoid biped robot model, the feet adopt planar sole structures; through the pre-training control parameters, training of a controller is carried out through the improved deep reinforcement learning method adopting a DQN network structure; and the stability of walking movementof the biped robot is realized through the trained controller.

Description

technical field [0001] The invention relates to an improvement of robot technology, in particular to a new deep reinforcement learning method for gait planning of a humanoid robot. Background technique [0002] The humanoid biped robot has a humanoid structure that can adapt to complex terrain environments such as stairs, streets, and uneven ground, and has the characteristics of flexible movement forms. Therefore, robots with this type of motion can be used in many industries such as medical treatment, rescue, service, and exploration. Human locomotion looks simple, but it is a very complex action involving multiple degrees of freedom. It can be seen as the complex nonlinear dynamics arising from the interrelationship of these degrees of freedom through the extensor and flexor groups of the lower body, which also became the motivation to properly understand the physiology involved in the study of locomotion and replicate it to bipedal robots . Biped walking robots have t...

AI Technical Summary

Problems solved by technology

[0004] The technical problem to be solved in the present invention overcomes the existing defects, provides a new method of deep reinforcement learning for gait planning of humanoid robots, uses the improved DQN algorithm to solve the walking control problem of biped robots, and does not need to establish complex bipeds Based on the robot dynamics model, the reinforcement deep learning method is applied to the biped robot to achieve long-distance stable gait control under fast walking conditions

Images