A control method and system for a general-purpose legged robot

By unifying and tokenizing multi-source sensor data and introducing masking codes and cross-morphological fusion, the problem of reduced control performance caused by incomplete sensor data is solved, and universal adaptation and stable control for different robot morphologies are achieved.

CN122018332BActive Publication Date: 2026-06-30HARBIN INSTITUTE OF TECHNOLOGY (SHENZHEN) (INSTITUTE OF SCIENCE AND TECHNOLOGY INNOVATION HARBIN INSTITUTE OF TECHNOLOGY SHENZHEN)
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
HARBIN INSTITUTE OF TECHNOLOGY (SHENZHEN) (INSTITUTE OF SCIENCE AND TECHNOLOGY INNOVATION HARBIN INSTITUTE OF TECHNOLOGY SHENZHEN)
Filing Date
2026-04-10
Publication Date
2026-06-30

AI Technical Summary

Technical Problem

Existing control methods for legged robots exhibit a sharp decline in control performance when sensor data is incomplete, and multi-source sensor fusion methods struggle to handle asynchronous data issues, leading to robot instability in complex environments.

Method used

By unifying and tokenizing multi-source sensor data and adding modal, temporal, and sensor configuration embeddings, combined with occlusion masking and occlusion-time attention encoders, world state latent variables are generated. Robot morphological description information is then fused in the cross-morphological fusion module to output joint-level control commands.

Benefits of technology

It achieves robustness against sensor data occlusion and frame loss in complex environments, adapts to different robot forms, and forms a unified, end-to-end deployable general-purpose legged robot control scheme, improving control stability and generalization ability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122018332B_ABST
    Figure CN122018332B_ABST
Patent Text Reader

Abstract

This invention provides a control method and system for a universal legged robot, relating to the field of robot control technology. The control method constructs a unified input representation compatible with heterogeneous sensors and different sampling frequencies by unifying and tokenizing multi-source sensor temporal observation data and embedding modal, temporal, and sensor configuration information. It introduces occlusion masks to enable the occlusion-based temporal attention encoder to handle data occlusion and frame loss. The robot morphological description information and world state latent variables are fused in a cross-morphological fusion module, enabling a single policy network (composed of an occlusion-based temporal attention encoder, a cross-morphological fusion module, and a universal policy head) to adapt to legged robot morphologies with different kinematic topologies and dynamic parameters. Finally, it outputs joint-level control commands to drive robot movement, forming a complete universal legged robot control scheme. This effectively solves the core problems of existing technologies, such as reliance on specific sensors and morphologies and difficulty in handling missing perception data.
Need to check novelty before this filing date? Find Prior Art