Robot Training Data Screening for Faster Learning Convergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robot operation systems using machine learning learn both intended and unintended operation data, leading to prolonged training times and inefficiencies, requiring manual visual checks to ensure intended operations are achieved, and often necessitating redoing the data collection process if the robot fails to perform as intended.
Innovation Solution
A training data screening device that includes a data evaluation model, evaluator, and screener to filter training data based on machine learning, allowing operators to quickly evaluate and select appropriate data for constructing learning models, reducing the time and effort required for trial and error.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If the machine learning model learns from all collected data including unintended operations, then the training data quantity is maximized, but the training time becomes excessively long and the learning convergence is delayed
Solution Approach 1:
The patent extracts and removes unintended operation data from the collected training data using an evaluation model that identifies and separates harmful data points. This extraction process eliminates the negative impact of inappropriate operations while preserving the beneficial intended operation data, thereby reducing training time without sacrificing data quantity quality
Solution Approach 2:
The patent changes the parameter of data quality by introducing an evaluation mechanism that assigns quality scores to training data points. By filtering and selecting only high-quality data points for training, the system achieves faster convergence while maintaining an effective training dataset, resolving the contradiction between data quantity and training time
2Productivity
If the machine learning model learns from all collected data without screening, then the data collection process is simple and fast, but the learning model requires multiple trial-and-error cycles and manual visual checks
Solution Approach 1:
The patent applies preliminary action by evaluating and screening training data before the actual machine learning training process begins. The evaluation model pre-identifies and filters out unintended operations, so that when training commences, only high-quality data is used. This preliminary filtering eliminates the need for multiple trial-and-error cycles and manual visual checks during and after training
3Adaptability or versatility
If the robot performs operations based on learned data including unintended operations, then the learning model is trained on comprehensive data, but the robot fails to achieve intended operations and requires redoing data collection
Solution Approach 1:
The patent converts the harmful effect of unintended operation data into a benefit by using the evaluation model to identify patterns of inappropriate operations. The system learns from these harmful examples by explicitly filtering them out, thereby improving the reliability of the robot's operations while maintaining the adaptability gained from diverse training scenarios
Data Source
AI summary
A training data screening device includes a data evaluation model, a data evaluator, a memory, and a training data screener. The data evaluation model is constructed by machine learning on at least a part of the collected data, or by machine learning on data different from the collected data. The data evaluator evaluates the input collected data using the data evaluation model. The memory stores the evaluated data, which is the collected data evaluated by the data evaluator. The training data screener screens the training data tier constructing the learning model from the evaluated data stored by the memory by an instruction of an operator to whom an evaluation result of the data evaluator is presented, or automatically screens the training data based on the evaluation result.