Wearable Multimodal Sensing for Problem Behavior Precursor Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to effectively detect precursors of problem behaviors in individuals with developmental disabilities and autism spectrum disorder, which can lead to physical harm, non-compliance, property destruction, and elopement, limiting the ability to intervene and prevent these behaviors.
Innovation Solution
A multimodal data capture system using wearable sensors and machine-learning models to identify precursors of problem behaviors by analyzing motion, audio, physiological signals, and facial expressions, with a self-monitoring application providing notifications and recommendations for mitigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing systems are used to monitor behavior, then they fail to detect precursors of problem behaviors, but implementing a new multimodal system increases device complexity
Solution Approach 1:
The system segments the monitoring task across multiple independent sensor modalities (accelerometers, gyroscopes, microphones, cameras, physiological sensors) that each capture specific types of data. This segmentation allows the system to detect precursors through multiple independent channels, improving measurement precision while managing complexity through modular architecture
Solution Approach 2:
The system merges data from multiple sensor modalities and sources (wearable sensors, environmental sensors, historical data) into a unified analysis framework. By combining these diverse data streams, the system achieves superior precursor detection capability that exceeds what any single sensor could provide alone
2Measurement precision
If a multimodal data capture system is implemented, then precursor detection capability is improved, but the quantity of data to be processed increases
Solution Approach 1:
The system extracts only the most relevant features and patterns from the multimodal data streams that are predictive of problem behavior precursors. By filtering and extracting only the critical signal components, the system maintains high detection accuracy while reducing the volume of data that requires extensive processing and storage
Solution Approach 2:
The system performs preliminary processing, feature extraction, and pattern recognition on the raw multimodal data before full analysis. This preliminary action prepares the data in advance, identifying potential precursor patterns early in the data pipeline, which reduces the computational burden on subsequent processing stages
3Loss of time
If real-time monitoring is implemented, then intervention timing is improved, but energy consumption increases
Solution Approach 1:
The system employs periodic monitoring and analysis cycles rather than continuous full-power processing. Sensors activate and data analysis occurs in periodic intervals, allowing the system to detect precursors in real-time while consuming energy only when needed, thus reducing overall energy consumption while maintaining timely intervention capability
Solution Approach 2:
The system includes self-monitoring capabilities where the individual can provide feedback about their own state and precursor experiences. This self-service aspect reduces the need for constant external monitoring and processing, allowing the system to operate more efficiently with lower energy consumption while still providing timely interventions
Data Source
AI summary
Systems and methods for predicting problem behavior in individuals with developmental and behavior disabilities. A plurality of sensors are configured to collect multimodal data signals of a subject individual including a wearable upper body motion sensing device with a plurality of inertial measurement units (IMUs). An electronic controller is configured to receive output signals from each of IMUs and to model an upper body position of the subject individual based on the output signals from the IMUs. A trained machine-learning model is then applied by providing an input data set that includes multimodal signal data (e.g., including signal data from at least one IMU) and/or features extracted from the multimodal signal data. The machine-learning model is trained to produce as output an indication of whether a precursor to the problem behavior is detected and, in response to detecting the precursor, a notification (or alarm) is generated.