3D Grasp Point Clustering for Consistent Robot Grip Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies for generating grasp information for robot hands are inconsistent due to parametric-based analysis, requiring numerous parameter adjustments, leading to high computational costs and generation of redundant or unrealistic grasp information, and are limited to external gripping, failing to address internal gripping scenarios.
Innovation Solution
A grasp information generation device that acquires 3D target object information and gripping section data, performs clustering to extract representative gripping points, and generates relative positional information using pre-processing techniques like simplification, smoothing, and optimization, while considering stability and interference, to produce consistent grasp information efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If parametric-based analysis approach is adopted to generate grasp information, then grasp information can be generated, but the generated grasp information is inconsistent and depends on parameter magnitude values, requiring specialist knowledge for parameter adjustment
Solution Approach 1:
The patent replaces the parametric-based mechanical analysis approach with a data-driven machine learning approach. Instead of adjusting multiple parameters manually, the system uses a trained model that automatically generates consistent grasp information based on object geometry, eliminating the need for specialist knowledge in parameter tuning
Solution Approach 2:
The system performs self-service by automatically selecting optimal parameters through the trained machine learning model. The model independently determines appropriate grasp parameters based on input object data without requiring external expert intervention for parameter adjustment
2Reliability
If parametric-based analysis with mesh segmentation and multiple impact detections is performed, then grasp information can be generated, but computation time becomes excessively long
Solution Approach 1:
The patent applies preliminary action by pre-training the machine learning model offline with comprehensive grasp data. This pre-processing phase captures essential grasp patterns, allowing the deployed system to generate high-quality grasp information rapidly without performing computationally intensive mesh segmentation and multiple impact detections during runtime
Solution Approach 2:
The system creates a simplified copy of complex grasp scenarios through the trained model. Instead of repeatedly performing full mesh segmentation and impact detection simulations, the model uses learned patterns from training data to directly predict grasp parameters, significantly reducing computation time while maintaining quality
3Adaptability or versatility
If parametric-based analysis is used to generate grasp information, then grasp information covering various orientations can be produced, but redundant grasp information for unrealistic gripping orientations is generated
Solution Approach 1:
The patent changes the approach from manual parametric adjustment to data-driven parameter selection. The machine learning model learns realistic grasp parameter ranges from training data, automatically filtering out unrealistic orientations while maintaining coverage of valid grasp configurations, thus reducing redundancy without sacrificing versatility
Data Source
AI summary
An acquisition section 30 acquires target object information indicating a three dimensional shape of a target object for gripping by a gripping section, and gripping section information related to a shape of the gripping section. An extraction section (50) performs clustering on pairs of two points on the target object that are pairs satisfying a condition based on the target object information and the gripping section information acquired by the acquisition section (30) and are pairs remaining after similar pairs have been removed, and extracts as gripping points a pair representative of each cluster. Based on the gripping points extracted by the extraction section (50), the generation section (60) generates grasp information indicating a relative positional relationship between the target object and the gripping section for a case of the target object being gripped by the gripping section.