Web Analytics

Origin - On-Device Intelligence for Activity Recognition

Origin - On-Device Intelligence for Activity Recognition

Overview

Origin enables on-device intelligence for human activity recognition using energy harvesting wireless sensor networks. The system combines efficient machine learning algorithms with energy harvesting techniques to provide continuous monitoring with minimal power consumption.

Key Features

  • Energy Harvesting Integration: Seamless integration with various energy harvesting sources
  • On-Device Machine Learning: Efficient algorithms for activity recognition without cloud connectivity
  • Power-Aware Processing: Adaptive computation based on available energy
  • Wireless Sensor Networks: Distributed sensing for comprehensive activity monitoring
  • Real-Time Recognition: Low-latency activity classification for immediate responses

Technical Details

Origin consists of several components:

  1. Energy Harvesting Interface: Support for solar, thermal, and kinetic energy harvesting
  2. Sensor Fusion: Multi-modal sensor data integration for robust activity recognition
  3. ML Inference Engine: Optimized neural networks for on-device classification
  4. Power Management: Intelligent power allocation and sleep scheduling
  5. Communication Protocol: Energy-efficient wireless communication between sensor nodes

Results

Our evaluations show that Origin achieves high activity recognition accuracy while operating continuously on harvested energy. The system has been tested on various activity recognition tasks and demonstrates significant improvements in energy efficiency compared to traditional battery-powered approaches.

Publications

  • Origin: Enabling On-Device Intelligence for Human Activity Recognition Using Energy Harvesting Wireless Sensor Networks
    Cyan S. Mishra, Jack Sampson, Vijaykrishnan Narayanan
    Design, Automation and Test in Europe Conference (DATE), 2021

Future Work

We are currently extending Origin to support:

  • Multi-agent reinforcement learning scenarios
  • Integration with federated learning frameworks
  • More sophisticated reward shaping techniques
  • Broader range of reinforcement learning algorithms