Robotic Picking ML Retraining for Edge Case Object Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional machine learning systems in automated robotic picking systems struggle with edge cases (tail data) due to insufficient training data, leading to poor performance and frequent human intervention, which is costly and inefficient.
Innovation Solution
A system that allows for real-time or near-real-time generation of new machine learning models by incorporating human-in-the-loop feedback to address edge cases, augmenting training data, and deploying updated models to improve robotic picking efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If more training data is collected to improve edge case performance, then edge case handling is improved, but data collection time and cost increase
Solution Approach 1:
The patent performs preliminary action by proactively identifying edge cases and collecting targeted training data before they become performance problems. The system uses scenario classification to detect potential edge cases and pre-collects relevant training data, eliminating the need for extensive post-deployment data collection.
Solution Approach 2:
The system implements self-service by automatically identifying edge cases, collecting relevant data, and training specialized models without requiring extensive manual intervention. The ML system serves itself by detecting performance gaps and autonomously gathering the necessary training data to address them.
2Adaptability or versatility
If ML models are updated frequently to address new edge cases, then adaptability is improved, but system complexity increases
Solution Approach 1:
The patent segments the model update process into modular components: scenario classification, specialized model training, and model selection. This segmentation allows individual specialized models to be updated independently without affecting the entire system, reducing update complexity while maintaining high adaptability.
Solution Approach 2:
The patent applies local quality by making updates only where needed - specifically in specialized models for detected edge cases - rather than updating the entire ML system. This localized update approach improves adaptability to new edge cases while minimizing the complexity increase associated with system-wide updates.
3Measurement precision
If human intervention is used to label training data, then data quality is improved, but processing speed deteriorates
Solution Approach 1:
The patent applies partial action by having humans label only the critical portions of data - specifically edge case examples and ambiguous scenarios - while automated systems handle the majority of common cases. This selective human involvement maintains high data quality where it matters most while preserving processing speed for routine scenarios.
Data Source
AI summary
The present invention relates to systems and methods for accounting for edge cases (i.e. tail data) in automated decision making systems, for example automated robotic picking systems. The systems and methods provide for retraining machine learning (ML) models so that the edge cases can be handled in a manner that requires less (or no) human intervention. The disclosed systems and methods create updated ML models, replacement ML models, and/or supplementary ML models that can provide better performance (e.g. improved automated robotic picking) when edge cases are encountered. Furthermore, the present inventions disclose systems and methods for obtaining training data faster and in a more cost effective manner, which enables the systems and methods disclosed herein to update models at a faster rate, thereby enabling broader, system-wide handling of edge cases in a more effective and efficient manner.