Image Processing for Robotic Bin Picking Grip Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for calculating grip success probability in robotic grasping during bin picking require excessive computational load due to the need to evaluate multiple rotational positions of robot hands.
Innovation Solution
An image processing device and method that corrects the position of target ranges within patch images using a learned correction amount, generating a second patch image for precise grip success probability calculation, reducing the need to calculate probabilities for each rotational position.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If grip success probability is calculated for each rotational position of the robot hand, then the accuracy of grip selection is improved, but the computational load becomes excessive
Solution Approach 1:
The patent segments the grip evaluation process into two stages: first evaluating translational positions to identify candidate components, then evaluating rotational positions only for those candidates. This segmentation reduces the total number of grip probability calculations from N_components × N_rotational_positions to N_candidates × N_rotational_positions, where N_candidates << N_components, thereby reducing computational load while maintaining accuracy for critical evaluations
Solution Approach 2:
The patent performs preliminary evaluation of translational positions before evaluating rotational positions. By first determining which components are reachable and viable candidates based on translational grip probability, the system prepares a reduced set of candidates that require full rotational evaluation. This preliminary filtering action prevents unnecessary computational waste on impossible or low-probability grips
Data Source
AI summary
If the patch image, referred to as a first patch image, cut from an image within the cutting range, referred to as a target range, set for one component is input to the alignment network unit, the correction amount for correcting the position of the cutting range for one component included in the patch image is output from the alignment network unit. Then, the image within the corrected cutting range obtained by correcting the cutting range by this correction amount is cut from the composite image, referred to as a stored component image, to generate the corrected patch image, referred to as a second patch image, including the one component, and the grip success probability is calculated for this corrected patch image.