Robot Action Learning via 3D Pose Matching and Motion Stitching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for controlling robot actions, such as using action catchers or robot trajectory planning, are inefficient and lack the ability to seamlessly learn and execute complex actions from human demonstrations without extensive calibration or planning.
Innovation Solution
An action learning method that acquires human body moving image data, determines three-dimensional pose action data, matches it with a robot atomic action library, performs continuity stitching on robot sub-actions, and optimizes the sequence to ensure smooth and safe robot action execution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If action catchers or trajectory planning methods are used to control robot actions, then robot actions can be controlled, but the control efficiency is low and extensive calibration or planning is required
Solution Approach 1:
The patent uses motion capture technology to record human body movements and creates a digital copy of human actions. This motion data is then mapped to robot joints, allowing the robot to replicate human actions directly without requiring complex trajectory planning or calibration procedures. The motion capture system captures three-dimensional coordinates of human body parts, which are transformed into robot joint space coordinates, enabling efficient action transfer from human to robot.
2Loss of time
If human body movements are directly mapped to robot actions, then action learning speed improves, but action continuity and smoothness may be compromised
Solution Approach 1:
The patent performs preliminary action by pre-defining atomic actions as basic units of motion. These atomic actions are stored in a library and can be selectively combined and stitched together to form complex robot actions. This preliminary structuring of actions into reusable atomic units allows for both rapid action learning (by selecting from pre-defined atoms) and maintains action continuity (by properly stitching the atomic actions together with smooth transitions).
Solution Approach 2:
The patent segments complex human movements into discrete atomic actions, each representing a basic motion unit. By dividing the continuous human motion into separable atomic components, the system can efficiently process and reproduce these actions on the robot. The atomic actions are then stitched together in sequence to reconstruct the full motion, ensuring both speed and continuity.
3Measurement precision
If multiple action catcher devices are worn on the human body to capture movements, then action capture accuracy improves, but device complexity and ease of operation deteriorate
Solution Approach 1:
The patent employs a unified motion capture system that can track multiple body parts simultaneously using a single integrated system. Instead of requiring separate action catcher devices for each body part, the system uses multiple sensors or cameras that work together to capture three-dimensional coordinates of various human body parts. This multi-functional approach maintains high measurement precision while simplifying the operation for the user.
Data Source
AI summary
An action learning method, including: acquiring human body moving image data; determining three-dimensional human body pose action data corresponding to the human body moving image data; matching the three-dimensional human body pose action data with atomic actions in a robot atomic action library to determine robot action sequence data corresponding to the human body moving image data; performing action continuity stitching on all robot sub-actions in the robot action sequence data sequentially; determining a continuous action learned by a robot from the robot action sequence data subjected to the action continuity stitching.