Neural Vision Picking of Flexible Containers in Overlapping Stacks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for removing compliant containers, such as sacks and bags, from stacks are error-prone and inefficient, especially when containers are partially overlapping or in unknown arrangements, due to their reliance on predefined geometries and rigid object handling techniques, which limits their applicability and increases the risk of container damage.
Innovation Solution
The use of trained neural networks to process image data from compliant containers, including two-dimensional and three-dimensional information, to determine precise position and orientation data, allowing for reliable and damage-free removal even in complex or overlapping arrangements, and enabling automatic processing without user-defined programming.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If predefined geometries and rigid object handling techniques are used for removing compliant containers, then the removal process is simple and fast, but the accuracy of position determination deteriorates and container damage increases
Solution Approach 1:
The patent replaces traditional mechanical vision systems and rigid object handling techniques with trained neural networks that process image data to determine container positions and orientations. This substitution enables accurate identification of compliant containers with varying geometries, filling levels, and arrangements without requiring predefined models, thereby maintaining high productivity while achieving precise position determination.
Solution Approach 2:
The patent changes the approach from using fixed geometric parameters to using trained neural networks that can adapt to varying container parameters including different geometries, filling levels, and stack arrangements. This parameter adaptation allows accurate position determination across diverse container types while maintaining efficient automated removal processes.
2Ease of operation
If predefined geometries and rigid object handling techniques are used, then the system is simple to operate, but the reliability of container removal deteriorates due to errors with overlapping or unknown arrangements
Solution Approach 1:
The patent replaces traditional mechanical vision systems and rigid object handling techniques with trained neural networks that process image data to determine container positions and orientations. This substitution enables accurate identification of compliant containers with varying geometries, filling levels, and arrangements without requiring predefined models, thereby maintaining high productivity while achieving precise position determination.
Solution Approach 2:
The patent changes the approach from using fixed geometric parameters to using trained neural networks that can adapt to varying container parameters including different geometries, filling levels, and stack arrangements. This parameter adaptation allows accurate position determination across diverse container types while maintaining efficient automated removal processes.
3Reliability
If trained neural networks are used to process image data and determine precise position data, then the reliability and accuracy of container removal improve, but the device complexity increases
Solution Approach 1:
The patent implements a universal neural network-based system that handles multiple container types, geometries, filling levels, and stack arrangements through a single trained model. This universal approach consolidates what would otherwise require multiple specialized systems, reducing overall device complexity while maintaining high reliability across diverse应用场景.
Solution Approach 2:
The patent uses image data copying and processing through neural networks to create virtual representations of container positions and orientations. This allows the system to analyze and determine precise positions without physical contact or complex mechanical sensing, reducing hardware complexity while improving measurement accuracy and reliability.
4Device complexity
If traditional image processing with limit value queries is used for compliant containers, then the device complexity is low, but the measurement precision deteriorates due to varying container shapes and arrangements
Solution Approach 1:
The patent replaces traditional mechanical vision systems and rigid object handling techniques with trained neural networks that process image data to determine container positions and orientations. This substitution enables accurate identification of compliant containers with varying geometries, filling levels, and arrangements without requiring predefined models, thereby maintaining high productivity while achieving precise position determination.
Solution Approach 2:
The patent changes the approach from using fixed geometric parameters to using trained neural networks that can adapt to varying container parameters including different geometries, filling levels, and stack arrangements. This parameter adaptation allows accurate position determination across diverse container types while maintaining efficient automated removal processes.
Data Source
Figure 1~2
Figure 3~4
Figure 5~9
AI summary
A method and a device for removing a flexible container from a plurality of containers are proposed. The method involves generating image data of each container and, by applying at least one trained neural network to the image data, determining at least one positional piece of information for the container to be removed. This positional information is transmitted to a programmed manipulator, which then removes the container. The device includes a control unit designed to carry out the aforementioned method. Furthermore, the invention relates to a computer program with program code elements configured to perform the steps of the aforementioned method.