Hand Grip Pose Estimation from Point Clouds for Unstable Sensing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in accurately and stably determining the position and attitude of a hand gripping a target object due to unstable sensing results from range sensors, leading to potential failure in gripping the object.
Innovation Solution
An information processing device and method that estimates candidates for the position and attitude of a hand based on point cloud data, using machine learning to stabilize these estimates and approximate the target object's shape to basic shapes like spheres, cylinders, or rectangular plates, thereby constraining the degree of freedom and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the sensing result from a range sensor is used to recognize the shape of the target object, then the position and attitude of the hand can be estimated, but the estimation becomes unstable when the sensing result is unstable
Solution Approach 1:
The system uses template matching feedback to validate and correct the hand position and attitude estimates. By comparing the estimated hand configuration against stored templates of hand-object interactions, the system can detect inconsistencies in the sensing results and adjust the estimation accordingly, thereby stabilizing the position and attitude determination even when the range sensor data is unstable.
Solution Approach 2:
The system performs preliminary template matching to pre-identify valid hand configurations before finalizing the position and attitude estimation. This preliminary action filters out unstable or incorrect sensing results by comparing them against pre-stored valid interaction patterns, ensuring that only reliable estimates are used for gripping operations.
2Reliability
If template matching is used to recognize the target object, then the gripping position can be determined, but unknown target objects not present in the template cannot be recognized
Solution Approach 1:
The system segments the object recognition task into two independent parts: first, it uses template matching to identify the target object category and obtain preliminary position information; second, it uses range sensor data to determine the precise hand position and attitude for gripping. This segmentation allows the system to handle both known and unknown objects effectively, as the template matching provides category-level guidance while the range sensor data enables precise positioning regardless of object specificity.
Data Source
AI summary
To more stably determine the position and attitude of a hand gripping a target object.An information processing device includes: a position/attitude estimation unit that, taking each of points included in point cloud data generated based on a sensing result for a target object as contact points, estimates, for each of the points, candidates for a position and an attitude of a hand that grips the target object; a target object shape estimation unit that estimates a shape of the target object based on a distribution of the candidates for the position and the attitude estimated for each of the points; and a position/attitude determination unit that determines the position and the attitude of the hand gripping the target object based on the shape of the target object estimated.