Vision-Guided Picking Tool Pose Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automated single box picking and decanting systems face challenges in adapting to varying box configurations and require extensive human intervention or training, with machine learning approaches being opaque, inefficient, and prone to overtraining or undertraining, limiting their flexibility and reliability.
Innovation Solution
A method and system that use three-dimensional convolution to optimize the pose of a picking tool by generating and ranking legal poses based on the configuration of items on a transport structure, allowing for autonomous manipulation of box-like objects without pre-defined parameters, utilizing a vision-guided robot and hybrid 2D/3D sensors to determine the optimal picking pose.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If machine learning approaches are used for automated box picking, then the system can perform automated manipulation, but the system becomes opaque and requires extensive training
Solution Approach 1:
The patent replaces machine learning-based automated manipulation with a vision-guided system that uses classical computer vision algorithms. The system captures images of box configurations, processes them through deterministic algorithms to determine pick sequences, and controls robotic manipulators accordingly. This substitution eliminates the opacity and training requirements of machine learning while maintaining automation capability.
2Extent of automation
If machine learning approaches are used for box picking, then automation is achieved, but the system becomes inefficient and prone to overtraining or undertraining
Solution Approach 1:
The system replaces machine learning with classical computer vision and deterministic algorithms. The vision system captures images of box configurations on conveyors, processes them through algorithms that determine optimal pick sequences, and controls robotic manipulators. This approach eliminates training time, avoids over/undertraining issues, and provides immediate, efficient operation.
3Extent of automation
If traditional automated systems are used, then automation is provided, but they require pre-defined parameters and extensive human intervention
Solution Approach 1:
The system enables the automated manipulation system to determine its own operating parameters in real-time. The vision system continuously captures images of box configurations, and the deterministic algorithms automatically calculate optimal pick sequences based on current conditions. This eliminates the need for pre-defined parameters and extensive human setup, allowing the system to adapt autonomously to varying box arrangements.
4Extent of automation
If machine learning approaches are used, then automated picking is achieved, but the system lacks flexibility and reliability
Solution Approach 1:
The patent replaces machine learning with classical computer vision and deterministic algorithms that provide transparent, reliable decision-making. The vision system captures images of box configurations, processes them through predictable algorithms to determine pick sequences, and controls robotic manipulators. This substitution eliminates the unreliability associated with machine learning model failures, providing consistent and trustworthy automated operation.
Data Source
AI summary
A method and system for optimizing the pose of a picking tool with respect to at least one previously selected item in a topmost layer of target items to be picked from a transport structure are provided. The method includes the step of generating all legal poses of the picking tool with respect to the configuration of items on the topmost layer in which the picking tool subtends the at least one previously selected item. The method also includes selecting the picking tool pose for picking the at least one previously selected item based on the generated legal poses.