Gait planning method for biped robot based on deep reinforcement learning

Abstract

The invention discloses a gait planning method for a biped robot based on deep reinforcement learning, and the method utilizes the stability and flexibility of human gait and combines deep reinforcement learning to effectively control the gait of the biped robot. The method includes the following steps: 1), establishing a passive biped robot model; 2), acquiring and processing human gait data andtarget gait data; 3), extracting the implicit features of the gait data of the biped robot and human gait data through using a noise reduction auto-encoder; 4), learning the gait characteristics of ahuman body through using deep reinforcement learning, and then planning the gait of the biped robot. The method combines deep reinforcement learning with human gait data to control the biped robot towalk as stable and compliant as a human.

Description

technical field [0001] The invention relates to the technical field of biped robots, in particular to a gait planning method for biped robots based on deep reinforcement learning. Background technique [0002] At present, the moving modes of mobile robots include crawler, wheel, biped and so on. Compared with tracked and wheeled robots, biped robots have stronger adaptability, and can move on both flat ground and irregular environments (up and down steps, walking on uneven ground, etc.). However, the biped robot itself is a highly nonlinear hybrid dynamic system, and its gait planning has always been a difficult problem. [0003] In addition to maintaining the stability of walking, the gait planning of biped robots must also consider issues such as energy efficiency, compliance, and environmental adaptability of walking movements. Gait planning methods based on simplified models are commonly used in biped robot gait planning. The method based on the simplified model start...

AI Technical Summary

Problems solved by technology

The invention utilizes the characteristics that the structure and walking process of the knee-legged robot model are similar to the human body model, and combines the deep reinforcement learning method driven by big data to solve the weak anti-interference ability and conventional intelligent gait of the model-based gait planning method. Improve the stability and compliance of the robot when walking, such as stiff gait in the planning method

Images