Robot Workpiece Picking via Machine Learning Data Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional robot systems face challenges in accurately picking out workpieces from a container due to errors in measurement data, leading to potential collisions with the container or other workpieces, which decreases operation efficiency.
Innovation Solution
A robot system that employs a measurement data processing device with a model data storage unit, a measurement data correction unit using machine learning to create teacher data and adjust parameters for correcting measurement data, and a position and orientation calculation unit to improve the accuracy of workpiece positioning and orientation calculation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If measurement data is used directly for workpiece positioning and collision detection, then the system operation is simple, but the positioning accuracy and collision detection reliability are insufficient leading to potential collisions
Solution Approach 1:
The system performs preliminary correction of measurement data using machine learning models before the actual picking operation. Teacher data is created in advance by arranging model data at calculated positions and orientations, and the correction model is trained offline. During operation, the pre-trained model quickly corrects measurement data, achieving high positioning accuracy without adding significant real-time processing complexity.
Solution Approach 2:
A machine learning correction model serves as an intermediary between raw measurement data and the positioning system. The model takes measurement data as input and outputs corrected position and orientation information, acting as a mediator that transforms inaccurate data into reliable positioning information without requiring direct modification of the underlying measurement system.
2Reliability
If measurement data correction using machine learning is implemented, then positioning accuracy is improved, but the system complexity and processing time increase
Solution Approach 1:
The machine learning model is trained offline in advance using teacher data generated from model data and measurement data. This preliminary training phase allows the system to learn correction patterns without affecting real-time operation. During actual picking operations, the pre-trained model applies corrections rapidly, ensuring both high reliability and efficient processing time.
Solution Approach 2:
The system creates teacher data by comparing measurement data with model data at calculated positions and orientations, then uses this feedback to train and refine the correction model. This feedback mechanism continuously improves the model's ability to correct measurement errors, enhancing collision detection reliability while maintaining processing efficiency through learned patterns.
3Productivity
If conventional collision detection methods are used, then the system structure is simple, but the detection accuracy is insufficient leading to false collision avoidance and reduced productivity
Solution Approach 1:
The system replaces conventional mechanical or rule-based collision detection methods with a machine learning-based detection approach. The correction model learns from teacher data to accurately predict workpiece positions and orientations, enabling more precise collision detection than traditional methods. This substitution reduces false positives that would otherwise cause unnecessary operation interruptions and maintain high productivity.
Data Source
AI summary
A robot system includes a robot including a hand for holding a workpiece; a sensor for measuring a work area in which the workpiece exists to obtain a three-dimensional shape in the work area as measurement data; a measurement data processing device including a model data storage unit, a measurement data correction unit, a position and orientation calculation unit; and a robot control unit for control the robot based on an output from the position and orientation calculation unit, wherein the measurement data correction unit in a learning stage, creates teacher data by arranging the model data in the position and the orientation calculated by the position and orientation calculation unit and adjusts a parameter for correcting the measurement data based on the measurement data and the teacher data, and in a picking stage, outputs corrected measurement data obtained by correcting the measurement data using the adjusted parameter.