Robot Gripping Point Selection Using Quality and Preference Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI methods for robot object manipulation struggle to reliably detect grippable regions while avoiding damage-prone areas, as specifying general criteria for machine learning is challenging.
Innovation Solution
A method involving neural networks to generate manipulation-quality and manipulation-preference images, combining pixel-wise assessments to select optimal gripping points based on user input, using manipulation-quality and manipulation-preference values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If AI methods are used to detect grippable regions, then automation is improved, but reliability of detecting preferred manipulation points deteriorates
Solution Approach 1:
The patent introduces descriptor images as an intermediary representation between the raw object image and the manipulation quality assessment. Neural networks first generate descriptor images that capture geometric and physical properties, then compare these descriptors against recorded preferences to determine reliable manipulation points. This intermediary step enables the system to incorporate user preferences and object properties systematically, improving reliability while maintaining automation.
2Adaptability or versatility
If general criteria are specified for machine learning, then adaptability is improved, but difficulty of specifying criteria deteriorates
Solution Approach 1:
The patent performs preliminary actions by recording descriptors of preferred and avoided manipulation points before actual object manipulation. These recorded descriptors serve as pre-established criteria that can be reused across different objects. The system pre-processes object images into descriptor images and pre-stores manipulation preferences, so that during operation, the robot can quickly compare new objects against the stored criteria without requiring complex real-time decision-making about manipulation preferences.
3Measurement precision
If pixel-wise assessment is performed for each point, then measurement precision is improved, but computational complexity deteriorates
Solution Approach 1:
The patent segments the manipulation quality assessment into multiple independent components: first generating a manipulation quality image that assesses geometric and physical properties at each pixel, then separately generating a manipulation preference image that evaluates user preferences at each pixel. These segmented assessments are performed independently using neural networks, allowing the system to maintain high measurement precision through detailed pixel-wise analysis while managing computational complexity by dividing the overall task into manageable sub-tasks that can be processed separately and then combined.
Data Source
AI summary
A method for controlling a robot for manipulating, in particular picking up, an object. The method includes: creating an image which depicts the object; generating a manipulation-quality image from the image, in which, for each pixel which represents a point on the surface of the object, the pixel value of the pixel provides an assessment of how well the object may be manipulated at the point; recording descriptors of points of the object which should be used during the manipulation and/or of points which should be avoided during the manipulation; mapping the image onto a descriptor image; generating a manipulation-preference image by comparing the recorded descriptors of points to the descriptor image; selecting a point for manipulating the object taking into account the pixel values of the manipulation-quality image and the pixel values of the manipulation-preference image; and controlling the robot to manipulate the object at the selected point.