Unknown-Object Robotic Grasping and Suctioning With 3D Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for grasping and suctioning objects, particularly in messy or cluttered environments, face challenges with accuracy, speed, reliability, and complexity, especially when dealing with unknown objects, and lack effective collision checking during grasping.
Innovation Solution
A system comprising a robot arm with gripping and suctioning units, a 3D camera for panoramic imaging, and neural network models for object segmentation and grasping/suctioning pose prediction, which calculates feasible poses based on visible scores and collision avoidance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If multiple 3D images from different views are used to generate grasping poses, then the completeness of object information is improved, but the processing time and computational complexity increase
Solution Approach 1:
The patent segments the 3D point cloud into multiple object instances, processing each object separately rather than handling all objects from multiple views simultaneously. This reduces the computational burden while maintaining complete object information through systematic segmentation and individual processing.
Solution Approach 2:
The patent performs preliminary collision checking between the robot arm and surrounding objects before generating final grasping poses. By pre-identifying colliding objects and filtering invalid poses in advance, the system avoids redundant computations and reduces overall processing time while ensuring information completeness.
2Reliability
If collision checking is performed during grasping generation, then the reliability of grasping poses is improved, but the computational load increases
Solution Approach 1:
The patent performs collision checking as a preliminary filtering step before final pose generation. By identifying colliding objects in advance and using their bounding boxes for quick collision detection, the system ensures reliable grasping poses while minimizing computational complexity through early elimination of invalid poses.
Solution Approach 2:
The patent uses simplified bounding box representations of objects for collision checking instead of detailed 3D models. This copying approach with simplified geometries maintains the essential collision information while significantly reducing computational complexity during the pose generation process.
3Measurement precision
If a pre-built grasping database with labeled 3D models is used, then the accuracy of grasping predictions is improved, but the data collection difficulty and computational requirements increase
Solution Approach 1:
The patent employs a robot arm to automatically collect 3D point cloud data of objects by physically interacting with them in the product cart. This self-service data collection approach eliminates the need for manual labeling and pre-built databases, achieving accurate grasping predictions through autonomous data gathering while reducing data collection complexity.
Solution Approach 2:
The patent changes the approach from using pre-built labeled databases to generating grasping predictions directly from raw 3D point cloud data captured in real-time. By modifying the data source and processing parameters, the system achieves accurate predictions without the complexity of data collection and labeling associated with pre-built databases.
Data Source
AI summary
The present invention relates to a system for grasping/suctioning unknown objects, using a robot arm or an industrial robot. The system comprising a 3D camera for taking images of the objects contained in said containing space and creating a 3D panoramic image; an unknown object segmentation model unit for receiving said 3D panoramic image as an input, and a processing the 3D panoramic image for creating a 2D mask and a 3D point cloud of each individual object/product in the 3D panoramic image; an unknown object grasping/suctioning model unit for outputting grasping/suctioning poses related to the target object based on the 3D point cloud of the target object, and selecting one of the feasible grasping/suctioning poses for controlling the robot arm to grasp/suction the target object according to said feasible grasping/suctioning pose.