Point Cloud Grasp Planning for Unseen Objects in Dense Clutter
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional bin picking methods in dense clutter environments are domain-dependent and fail to perform well when applied to different target domains, requiring extensive training data and being sensitive to sensor and gripper changes, and struggle with occlusions and object diversity.
Innovation Solution
A point cloud-based grasp planning framework using RGB-D sensors for unsupervised clustering, grasp pose validation, refinement, and quality ranking to obtain optimal grasp poses, independent of the domain, by generating sampled grasp poses, computing depth differences, and refining feasible poses using a Grasp Quality Score (GQS).
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional bin picking methods are used in dense clutter environments, then grasp planning can be performed, but the methods are domain-dependent and fail to perform well on different target domains
Solution Approach 1:
The patent replaces conventional domain-dependent grasp planning methods with a point cloud-based approach using RGB-D sensors and unsupervised clustering algorithms. This substitution enables domain independence by using geometric features from point cloud data rather than domain-specific training data, while maintaining reliability through validation and refinement steps that ensure accurate grasp pose selection for diverse objects in dense clutter
Solution Approach 2:
The patent changes the parameters used for grasp planning from domain-specific features to point cloud geometric features such as depth differences, surface normals, and curvature. By computing these parameters directly from sensor data and using unsupervised clustering to identify grasp poses, the system achieves domain independence while maintaining reliability through the Grasp Quality Score (GQS) evaluation framework
2Productivity
If deep learning-based methods are used for grasp planning, then grasp quality can be predicted, but extensive training data is required which is time consuming to collect
Solution Approach 1:
The patent enables the system to generate its own training data through unsupervised clustering of point cloud data. By using the RGB-D sensor to capture depth information and automatically clustering points to identify object surfaces and grasp poses, the system eliminates the need for extensive manual data collection and labeling, thereby achieving fast grasp planning without time-consuming training data preparation
Solution Approach 2:
The patent replaces deep learning-based grasp quality prediction with a geometric approach using point cloud analysis and unsupervised clustering. This substitution eliminates the need for training phases entirely, as the system directly computes grasp poses and quality scores from sensor data, achieving immediate productivity without training data collection time
3Loss of time
If CNN models are trained on simulated dataset, then training time is reduced, but the grasp quality becomes sensitive to parameters used during dataset generation
Solution Approach 1:
The patent replaces CNN-based grasp quality evaluation with a geometric approach using point cloud depth analysis and surface normal computation. By directly calculating depth differences and using unsupervised clustering to identify valid grasp poses, the system eliminates parameter sensitivity associated with simulated dataset generation while maintaining fast execution without training time
Solution Approach 2:
The patent creates a universal grasp planning framework that works across different domains and parameter configurations by using point cloud geometric features rather than domain-specific training data. The unsupervised clustering approach and Grasp Quality Score evaluation are parameter-independent, making the system universally applicable to diverse objects in dense clutter without requiring retraining or parameter adjustment
4Productivity
If conventional grasp planning methods are used, then grasp poses can be sampled, but they struggle with occlusions and object diversity in dense clutter
Solution Approach 1:
The patent replaces conventional grasp pose sampling methods with a point cloud-based unsupervised clustering approach. By using depth information from RGB-D sensors and clustering points to identify object surfaces, the system can effectively detect and measure objects in dense clutter even with occlusions, while maintaining efficient grasp pose sampling through geometric feature analysis
Solution Approach 2:
The patent transitions from 2D image-based grasp planning to 3D point cloud analysis by utilizing depth information from RGB-D sensors. This dimensional change enables the system to detect and measure objects in dense clutter more effectively by analyzing spatial relationships and depth differences, allowing accurate grasp pose sampling even when objects are occluded or diverse in shape
Data Source
AI summary
A fully automated and reliable picking of a diverse range of unseen objects in clutter is a challenging problem. The present disclosure provides an optimum grasp pose selection to pick an object from a bin. Initially, the system receives an input image pertaining to a surface. Further, a plurality of sampled grasp poses are generated in a random configuration. Further, a depth difference value is computed for each of a plurality of pixels corresponding to each of the plurality of sampled grasp poses. Further, a binary map is generated for each of the plurality of sampled grasp poses and a plurality of subregions are obtained. Further, a plurality of feasible grasp poses are selected based on the plurality of subregions and a plurality of conditions. Further, the plurality of feasible grasp poses are refined and an optimum grasp pose is obtained based on a Grasp Quality Score (GQS).