Robot Feature Point Detection Using CAD Composite Image Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing learning models for robot operation systems face accuracy issues due to variations in lighting conditions, workpiece mounting positions, and photographing conditions, which affect the precision of feature point detection.
Innovation Solution
A system that generates and learns composite images of objects and workpieces from CAD data under random conditions, using a trained machine learner to accurately identify feature points, enabling teachless robot operation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If training is performed using model data of workpiece, then learning model can be constructed, but detection accuracy deteriorates due to changes in lighting conditions, mounting position, and photographing conditions
Solution Approach 1:
The system performs preliminary actions by generating diverse training images from CAD data that pre-encode various lighting conditions, mounting positions, and photographing conditions before actual detection occurs. This prepares the learning model in advance to handle real-world variations, resolving the contradiction between detection accuracy and adaptability.
Solution Approach 2:
The system changes parameters by systematically varying lighting conditions, mounting positions, and photographing conditions when generating training images from CAD data. This creates a comprehensive dataset that teaches the learning model to maintain detection accuracy across diverse real-world conditions, simultaneously improving both accuracy and adaptability.
2Loss of time
If teachless operation is implemented, then manual teaching time is reduced, but feature point detection accuracy deteriorates under varying conditions
Solution Approach 1:
The system uses copying by generating synthetic training images from CAD data that replicate real workpiece appearances under various conditions. This allows the learning model to learn from synthesized copies without manual teaching, achieving both time efficiency and detection accuracy simultaneously.
Data Source
AI summary
A robot operation system with improved accuracy of a learning model accuracy is provided. Provided is a robot operation system including a composite image generation unit to which CAD data of an object and a workpiece and a feature point of the workpiece are input configured to generate a plurality of composite images under random conditions from the CAD data of the object and the workpiece, an information processing apparatus configured to search for a route using a position of an end effector of a robot and a position of the feature point and move the end effector to the feature point along the searched route, and an imaging apparatus configured to photograph the object and the workpiece.