Tire Gripping Control for Unknown Overlapping Arrangements
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing tire sorting and gripping technologies struggle with efficiently handling tires in unknown arrangements, particularly when tires are partially overlapping, as they require precise knowledge of tire configuration and visibility, which is not feasible in loose or unpredictable loads.
Innovation Solution
The implementation of a computer-implemented control process using few-shot learning with an attention mechanism to control a gripping device. This process involves acquiring data from the tire arrangement, training extraction and attention neural networks, performing 3D reconstruction to identify the target tire's location and orientation, and guiding the gripping device to grasp and remove the tire.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional gripping technologies are used with laser scans and CAD models, then precise gripping control is achieved, but the system requires highly specialized hardware installations and cannot handle unknown tire arrangements
Solution Approach 1:
The patent replaces traditional mechanical laser scanning systems with a camera-based vision system. Instead of using complex laser scanners to capture tire geometry, the system uses standard cameras to capture images, which are then processed through neural networks to extract geometric information. This substitution reduces hardware complexity while maintaining the ability to achieve precise gripping control through software-based 3D reconstruction.
Solution Approach 2:
The patent creates a digital copy (3D model) of the tire from 2D images captured by a camera. The neural network processes the images to reconstruct the tire's three-dimensional geometry, creating a virtual representation that can be manipulated and analyzed without requiring physical measurement devices. This copying approach eliminates the need for specialized laser scanning hardware while preserving measurement precision.
2Measurement precision
If CAD model realignment is used to detect objects, then precise spatial configuration is identified, but the system requires objects to be mostly visible and identical to the model
Solution Approach 1:
The patent transitions from static CAD model matching to a dynamic learning approach. Instead of requiring the tire to match a pre-defined CAD model, the system uses neural networks that can adapt to various tire configurations and arrangements. The model is trained on diverse examples and can dynamically adjust to unknown arrangements, making the system versatile while maintaining precise spatial detection capabilities.
Solution Approach 2:
The patent performs preliminary training of the neural network on a dataset of tire images and configurations before actual operation. This preliminary action allows the system to learn various tire arrangements and geometries in advance, so that during actual gripping operations, it can handle unknown arrangements without requiring the tires to match specific pre-defined models. The pre-trained network adapts to new configurations more effectively.
3Ease of operation
If control systems attempt to manage the environment with specialized installations, then precise control is achieved, but the system becomes inflexible to significant variations in installation
Solution Approach 1:
The patent creates a universal control system that can handle multiple tire arrangements and configurations using the same hardware setup. The camera-based vision system with neural network processing serves as a multi-functional solution that adapts to various installation conditions and tire types without requiring specialized hardware for each scenario. This universal approach maintains control precision while providing flexibility to installation variations.
4Measurement precision
If deep learning models with large amounts of labelled data are used, then accurate classification is achieved, but data collection and annotation costs increase significantly
Solution Approach 1:
The patent uses a semi-supervised learning approach where the system is trained on a relatively small amount of labelled data and then applies the learned model to unlabeled data through techniques like pseudo-labeling. This partial supervision strategy achieves high classification accuracy without requiring extensive manual annotation of every training example, significantly reducing data collection and annotation costs while maintaining precision.
Data Source
AI summary
The invention relates to a computer-implemented control process (201) for controlling the movement of a gripping device that grips a target tire from an unknown arrangement of tires in order to optimize the gripping of a target tire for which a target location must be reached during a sorting cycle. The invention also relates to a tire gripping control system (100) that performs the process of the invention.