实时模仿人类动作的机器人及其实时模仿动作方法
By decomposing human movements into bone segment vector matrices and projecting them into a low-dimensional space, the problems of shaking and safety hazards in robot motion imitation are solved, achieving efficient and accurate motion imitation and improving the user interaction experience.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ARTIFICIAL PRODUCTIVITY (BEIJING) TECHNOLOGY CO LTD
- Filing Date
- 2025-12-24
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, robots that imitate human actions suffer from poor motion accuracy, poor coordination, and safety hazards. Furthermore, methods based on convolutional neural networks and recurrent neural networks have high computational complexity and poor real-time performance.
By dividing the actions of the imitated subject into bone segment vector matrices for high-dimensional representation and projecting them into a low-dimensional classification space, the robot can control the corresponding actions of the parts, reducing computational costs and achieving accuracy and generalization.
It achieves accuracy and generalization of robot movements, reduces computational costs, prevents jitter, improves interactive safety, and enhances the user's interactive experience.
Smart Images
Figure CN121733598B_ABST