Unsupervised Gesture Classification via Neuromuscular Signal Clustering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing systems for recognizing and modeling human gestures require large sets of labeled data, which are prone to human bias and errors, and fail to capture meaningful features such as force applied during gestures.

Innovation Solution

An unsupervised machine learning approach is used to train a classification model based on neuromuscular signals from wearable sensors, allowing the system to learn gestures from few training samples and capture features indicative of force applied during gestures.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If supervised learning models are trained with large sets of labeled data, then gesture recognition accuracy is improved, but data labeling time and human resource requirements increase

Engineering Contradiction:
Improvegesture recognition accuracyVSAvoiddata labeling time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs self-labeling by automatically identifying gesture patterns and force levels from raw neuromuscular signals without human annotation. The unsupervised learning algorithm clusters EMG signal patterns to automatically categorize gestures and their intensity levels, eliminating the need for manual data labeling while maintaining recognition accuracy.

Inventive Principle:
Principle #25Self-service

2Adaptability or versatility

If supervised learning with labeled data is used, then gesture classification is achieved, but force application information is lost

Engineering Contradiction:
Improvegesture classification capabilityVSAvoidforce application information
Core Design Contradiction:
Adaptability or versatilityVSLoss of information

Solution Approach 1:

The system extends traditional gesture classification by adding a force dimension to the analysis. Instead of only categorizing gesture types, the unsupervised learning model simultaneously identifies force application levels by analyzing the magnitude and pattern of neuromuscular signals, creating a two-dimensional classification space of gesture type and force intensity.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

3Reliability

If traditional machine learning approaches are used, then gesture recognition requires extensive training data, but learning speed from few samples is reduced

Engineering Contradiction:
Improvegesture recognition reliabilityVSAvoidlearning speed
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The system performs preliminary unsupervised clustering of neuromuscular signal patterns during the training phase to pre-identify gesture categories and force levels. This preliminary organization of data structures enables the model to rapidly generalize from very few training samples, achieving reliable recognition with minimal training data by leveraging the inherent structure in the signal patterns.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3897890B1Methods and apparatus for unsupervised machine learning for classification of gestures and estimation of applied forces
Publication Date: 2025.04.23 META PLATFORMS TECHNOLOGIES LLC
  • EP3897890B1 patent drawingFigure 1
  • EP3897890B1 patent drawingFigure 2A~2B
  • EP3897890B1 patent drawingFigure 3A~3B

AI summary

Methods and apparatus for training a classification model and using the trained classification model to recognize gestures performed by a user. An apparatus comprises a processor that is programmed to: receive, via a plurality of neuromuscular sensors, a first plurality of neuromuscular signals from a user as the user performs a first single act of a gesture; train a classification model based on the first plurality of neuromuscular signals, the training including: deriving value(s) from the first plurality of neuromuscular signals, the value(s) indicative of distinctive features of the gesture including at least one feature that linearly varies with a force applied during performance of the gesture; and generating a first categorical representation of the gesture in the classification model based on the value(s); and determine that the user performed a second single act of the gesture, based on the trained classification model and a second plurality of neuromuscular signals.