Sensor Unit With On-Device Unsupervised Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems and devices that utilize inertial sensors struggle to efficiently detect, learn, and classify user and machine activities, particularly in recognizing new motions or anomalies without extensive user input or supervision.
Innovation Solution
An electronic device equipped with a sensor unit that includes one or more inertial sensors and a sensor processing unit, which uses unsupervised learning to recognize user activities. The device prompts the user to remain stationary before performing a motion, records sensor data, and generates a template for the motion, allowing it to automatically recognize future instances of the activity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If unsupervised learning is used to automatically recognize user activities, then the need for user input and supervision is reduced, but the complexity of the sensor processing unit increases
Solution Approach 1:
The sensor processing unit performs unsupervised learning automatically without requiring user input or supervision. The system self-trains by collecting sensor data during normal operation, generating templates autonomously, and continuously improving its activity recognition capabilities without external intervention.
Solution Approach 2:
The system collects and processes sensor data in advance to generate activity templates before actual activity recognition is needed. By pre-processing the data and creating reference templates during stationary periods, the system prepares recognition patterns ahead of time, reducing real-time processing complexity.
2Measurement precision
If sensor data is recorded continuously to improve activity detection accuracy, then measurement precision increases, but energy consumption increases
Solution Approach 1:
The sensor processing unit operates in periodic cycles, alternating between data collection mode and template generation mode. During stationary periods, the system collects sensor data and generates templates; during active periods, it uses these templates for efficient recognition. This periodic operation reduces continuous processing and lowers energy consumption while maintaining detection accuracy.
Solution Approach 2:
The system extracts only the essential features from sensor data to create compact activity templates. By identifying and extracting key motion patterns and characteristics rather than processing raw sensor data continuously, the system achieves high detection accuracy with reduced computational effort and lower energy consumption.
3Adaptability or versatility
If templates are generated for multiple motions and activities, then adaptability increases, but device complexity increases
Solution Approach 1:
The sensor processing unit uses a universal template structure that can represent multiple different activities and motions. By creating a flexible template format that captures essential motion characteristics applicable to various activities, the system achieves high adaptability without requiring separate complex processing pathways for each activity type.
Solution Approach 2:
The system segments activity recognition into distinct phases: data collection during stationary periods, template generation from collected data, and template-based recognition during active periods. This segmentation allows the system to manage multiple activity templates efficiently by processing them in discrete stages rather than simultaneously, reducing overall complexity.
Data Source
AI summary
An electronic device includes a sensor unit. The sensor unit includes a sensor and low power, low area sensor processing unit. The sensor processing unit performs an unsupervised machine learning processes to learn to recognize an activity or motion of the user or device. The user can request to learn the new activity. The sensor processing unit can request that the user remain stationary for a selected period of time before performing the activity. The sensor processing unit records sensor data while the user performs the activity and generates an activity template from the sensor data. The sensor processing can then infer when the user is performing the activity by comparing sensor signals to the activity template.