Exoskeleton Intent Labeling With Temporal Sensor Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional intent recognition systems for exoskeletons face challenges in accurately predicting user motions in real-time due to sensor noise and imperfections, leading to delays in classification and a need for expert supervision.
Innovation Solution
A data-driven intent recognition program that adapts over time, using unsupervised refinement methods to improve classification accuracy by considering sensor data before and after the time of interest, and incorporates user feedback to tune the classification program.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If conventional intent recognition systems use sensor data for real-time prediction, then responsiveness is improved, but accuracy deteriorates due to sensor noise and imperfections
Solution Approach 1:
The system performs preliminary classification using sensor data before and after the time of interest to establish candidate intent predictions. This preliminary action allows the system to prepare multiple potential outcomes in advance, which are then refined using additional data, thereby improving both responsiveness and accuracy simultaneously
Solution Approach 2:
The system introduces an intermediary refinement process that takes preliminary classifications and improves them using additional sensor data from before and after the time of interest. This intermediary step acts as a mediator between the raw sensor data and final intent recognition, filtering out noise while maintaining real-time performance
2Measurement precision
If expert supervision is used to improve classification accuracy, then accuracy is improved, but device complexity and operational burden increase
Solution Approach 1:
The system implements self-service through unsupervised refinement methods that automatically improve classification accuracy using sensor data from multiple time points. The system refines its own classifications without requiring external expert intervention, thereby maintaining high accuracy while reducing system complexity and operational burden
Solution Approach 2:
The system incorporates feedback mechanisms where classification results from preliminary processing are fed back into the refinement process. This internal feedback loop allows the system to continuously improve accuracy automatically, replacing the need for external expert supervision with an autonomous refinement cycle
3Measurement precision
If sensor data from multiple time points is processed, then classification accuracy is improved, but loss of time increases due to additional processing
Solution Approach 1:
The system performs preliminary classification using sensor data from multiple time points (before and after the time of interest) in parallel rather than sequentially. This preliminary action captures temporal context without adding significant processing delay, as the data from multiple time points is processed simultaneously to establish candidate predictions
Solution Approach 2:
The system processes sensor data from selected time points (before and after the time of interest) rather than continuously processing all available data. This partial action approach provides sufficient temporal context for accurate classification while minimizing processing delay by focusing only on the most relevant time windows
Data Source
AI summary
An exoskeleton network. The exoskeleton network includes an exoskeleton system that has one or more sensors and a memory; a user device that is local to the exoskeleton system and that operably communicates with the exoskeleton system; and a classification server that operably communicates with at least one of the exoskeleton system and the user device. The exoskeleton network performs feature extraction on sensor data obtained from the one or more sensors to generate feature-extracted sensor data and performs label derivation on the feature-extracted sensor data to generate labeled sensor data.