Training method of motion control strategy network and motion control method of humanoid robot

By constructing environmental and human geometry models in a humanoid robot, obtaining effective contact areas, and training a motion control strategy network, the problem of unstable movement of humanoid robots in complex terrain is solved, thereby improving motion stability and task success rate.

CN122323205APending Publication Date: 2026-07-03BEIJING HUMANOID ROBOTICS INNOVATION CENTER CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-05-28
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing technologies cannot simultaneously recover the three-dimensional geometric information of human movement and the surrounding environment, resulting in physical inconsistencies such as contact slippage, mesh penetration, and foot suspension in complex terrains for humanoid robots, affecting movement stability and mission success rate.

Method used

By acquiring human motion image sequences, performing multimodal perception, constructing an environmental geometric reconstruction model and a human geometric motion model, obtaining the effective contact area, constructing a local motion simulation scene, and training a motion control strategy network to obtain the target motion control strategy network.

Benefits of technology

It enables humanoid robots to perceive terrain in real-world environments, improving motion stability and task success rate, and eliminating problems such as contact slippage and mesh penetration.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122323205A_ABST
    Figure CN122323205A_ABST
Patent Text Reader

Abstract

This application provides a training method for a motion control strategy network and a motion control method for a humanoid robot, relating to the field of robot control technology. The method includes acquiring a sequence of human motion images of a preset scene; the sequence of human motion images includes multiple consecutive frames of human motion images; performing multimodal perception on the human motion image sequence to obtain an environmental geometric reconstruction model and a human geometric motion model; obtaining the effective contact area in the preset scene from the environmental geometric reconstruction model; constructing a local motion simulation scene with a simulated robot based on the human geometric motion model and the effective contact area; and training a motion control strategy network using the local motion simulation scene to obtain a target motion control strategy network. Through the method of this application, a motion control strategy with terrain perception capability can be trained, enabling the humanoid robot to adaptively adjust its motion according to terrain geometric features, improving motion stability and task success rate in real complex environments.
Need to check novelty before this filing date? Find Prior Art