Robot Vision Calibration Using Learned Pose Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for calibrating the coordinate system of a robot system using a vision sensor fixed on a hand-tip part are inefficient and lack precision in determining the position and posture of objects.
Innovation Solution
An arithmetic apparatus that includes a control part for controlling an imaging part and a learning part to generate a model for determining parameters of arithmetic processing, allowing precise calculation of the position and posture of objects based on imaging results, using a learning target object's predetermined positional relation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a vision sensor is fixed on a hand-tip part of a robot for coordinate system calibration, then the system can perform mobile observation from multiple positions, but the determination precision of object position and posture deteriorates
Solution Approach 1:
The system separates the calibration target into multiple markers distributed across its surface, and the vision sensor captures images from multiple robot positions. Each marker provides independent measurement data, and through segmented coordinate system calculations for each marker position, the system achieves high-precision object positioning despite mobile observation
Solution Approach 2:
The system uses feedback from multiple imaging results captured at different robot positions to iteratively calculate and refine the coordinate systems. By feeding back the positional relationships between markers and the robot base coordinate system, the system compensates for measurement errors and achieves precise determination of object position and posture
2Area of stationary object
If multiple imaging results from different robot positions are used for coordinate system calculation, then measurement coverage is improved, but the calculation complexity and time consumption increase
Solution Approach 1:
The system pre-establishes the robot base coordinate system and defines fixed marker positions on the calibration target before actual measurement. By preliminarily setting up the coordinate framework and marker configurations, the system simplifies subsequent calculations when processing multiple imaging results from different robot positions
Solution Approach 2:
The system creates a virtual model (copy) of the calibration target with known marker positions and dimensions. This digital copy is used to match against actual imaging results, allowing the system to efficiently calculate coordinate transformations without complex geometric computations from scratch for each image
Data Source
AI summary
An arithmetic apparatus includes: a control part that outputs a control signal for controlling an imaging part and a robot equipped with the imaging part; and a learning part that generates a model for determining a parameter of arithmetic processing in the control part, by learning using an imaging result of a learning target object by the imaging part. The control part outputs a first control signal. The learning part generates the model, by learning using learning image data generated by the imaging part imaging the learning target object in the predetermined positional relation, by control based on the first control signal. The control part performs the arithmetic processing, by using the parameter determined by the model and processing target image data generated by the imaging part imaging a processing target object, and calculates at least one of a position and posture of the processing target object.