Robot Grasp Pose Planning for Cluttered Object Picking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current robotic systems face challenges in determining stable grasp poses for objects in cluttered environments, especially when complete 3D models are not available, leading to collisions and reduced grasp success rates.
Innovation Solution
The development of a learning-based approach that generates 6-DOF grasps using partial point cloud observations, incorporating instance segmentation and a cascaded grasp generation method to reason about object-level grasps and collisions, allowing for collision-free grasp planning in structured clutter scenarios without requiring complete object models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If geometry-inspired heuristics are used to select grasp points, then grasp planning efficiency is improved, but grasp accuracy deteriorates when objects are in clutter
Solution Approach 1:
The patent segments the cluttered scene into individual object instances using instance segmentation, allowing the system to process each object separately. This segmentation enables accurate grasp point selection for target objects while filtering out interfering objects, thus maintaining both efficiency and accuracy in cluttered environments.
Solution Approach 2:
The patent introduces an intermediary collision determination step between grasp point selection and final grasp execution. This intermediary checks whether grasp points on target objects are occluded by interfering objects, allowing the system to maintain high accuracy by filtering out false positive grasp points that would cause collisions.
2Measurement precision
If complete 3D models are required for accurate grasp determination, then grasp accuracy is improved, but system complexity and data requirements worsen
Solution Approach 1:
The patent applies partial action by using only the necessary portion of object data - specifically, point cloud observations of visible surfaces are sufficient for accurate grasp determination. The system does not require complete 3D models of all objects, but only partial observations of target objects combined with instance segmentation to identify and filter interfering objects.
Solution Approach 2:
The patent replaces traditional mechanical approaches that rely on complete physical models with a learning-based approach using neural networks. The grasp determination network processes partial point cloud observations directly, substituting complex mechanical modeling with data-driven patterns recognition, thereby reducing system complexity while maintaining accuracy.
3Device complexity
If traditional grasp methods are used in cluttered scenes, then computational simplicity is maintained, but collision with interfering objects increases
Solution Approach 1:
The patent performs preliminary instance segmentation and identification of interfering objects before grasp point selection. By pre-processing the scene to separate target objects from interfering objects, the system eliminates collisions before they occur, maintaining computational simplicity while preventing harmful effects through advance preparation.
Solution Approach 2:
The patent converts the presence of interfering objects (harmful factor) into beneficial information by using instance segmentation to identify and characterize these objects. The interfering objects become part of the scene understanding that enables more accurate grasp planning, as their presence provides contextual information about the cluttered environment that helps select appropriate grasp points.
Data Source
AI summary
Apparatuses, systems, and techniques determine a set of grasp poses that would allow a robot to successfully grasp an object that is proximate to at least one additional object. In at least one embodiment, the set of grasp poses is modified based on a determination that at least one of the grasp poses in the set of grasp poses would interfere with at least one additional object that is proximate to the object.