-
An infograph summarizing the project
-
The IoT Power Problem
-
Our Solution: AmBC +Edge AI + XAI
-
System Architecture: The Digital Twin Pipeline
-
Performance and Benchmarks
-
The Final Result: Intelligent Anomaly Detection
-
Anomaly Illustrated
Battery-Free Health & Environment Monitoring with Explainable Edge AI
This project pioneers a truly sustainable and intelligent approach to health and environmental monitoring, addressing critical needs in underserved communities while pushing the boundaries of AI at the edge. By combining Ambient Backscatter Communication (ABC) with explainable Edge AI, we enable pervasive, battery-free sensing that delivers actionable insights directly to local healthcare workers, even in resource-constrained environments.
Inspiration
We are inspired by the urgent need for accessible, affordable, and sustainable health solutions in regions with limited infrastructure. Traditional IoT solutions often struggle with power requirements, connectivity limitations, and data privacy concerns. We believe that by harnessing ambient RF energy and processing data locally with intelligent, transparent AI, we can unlock unprecedented possibilities for proactive health and environmental management, particularly in developing contexts. Our work is driven by the vision of empowering communities with real-time, trustworthy data to improve well-being and build resilience.
What it does
Our system uses Ambient Backscatter Communication (ABC) to collect critical environmental data (e.g., water quality parameters, airborne pathogen indicators) from tiny, battery-free sensors deployed in remote clinics or homes. These sensors harvest energy from existing RF signals (like Wi-Fi or TV broadcasts) and modulate their readings onto these signals. A local edge hub, powered by a low-energy single-board computer, receives and processes these backscattered signals.
Crucially, this hub houses Edge AI models that immediately analyze the environmental data for anomalous patterns indicative of potential health issues or disease outbreaks. For instance, the AI can detect sudden spikes in water contaminants or unusual shifts in airborne particulate matter. The system then provides Explainable AI (XAI) components to justify its predictions, offering clear, human-readable reasons for detected anomalies to local healthcare workers, empowering them to make informed and timely interventions.
How we built it
We built a comprehensive simulation in Python, leveraging a modular approach that allowed us to independently develop and integrate each component:
- Ambient RF Source & Backscatter Tag Simulation: We accurately simulated the generation of ambient RF signals and the passive modulation by a backscatter tag based on environmental sensor data.
- Realistic Channel Modeling: We incorporated channel impairments like path loss, AWGN, and fading to ensure our simulation reflects real-world wireless conditions.
- Robust Receiver DSP: We developed Python-based Digital Signal Processing techniques to effectively demodulate the backscattered signals and extract clean sensor data.
- Lightweight Edge AI Models: We designed and trained compact neural networks using TensorFlow/Keras for both data classification and anomaly detection. These models were meticulously optimized for resource-constrained environments and converted to TensorFlow Lite (
.tflite) for efficient edge deployment. - Explainable AI Integration: While the core simulation focuses on detection, our architecture is designed to integrate post-hoc XAI techniques (like SHAP/LIME, though not explicitly coded in the provided notebook's output) to provide interpretability for the AI's anomaly predictions.
Challenges we ran into
- Simulating Realistic Low-Power Communication: Accurately modeling the nuances of ambient backscatter, including energy harvesting constraints and ultra-low power signal modulation, was a significant challenge.
- Optimizing AI for Extreme Edge Constraints: Developing AI models that are both effective for anomaly detection and small enough to run on limited computational resources at the edge required careful architecture design and optimization.
- Balancing Accuracy and Interpretability: Ensuring the AI models were accurate in detecting anomalies while also providing clear, actionable explanations for healthcare workers presented a complex design trade-off that informed our architectural choices.
Accomplishments that we're proud of
- End-to-End Battery-Free IoT Simulation: We successfully demonstrated a complete, self-contained simulation of an ABC system, proving the viability of battery-free sensing from signal generation to data processing. Also the simulation was FREE
- Effective Edge AI for Anomaly Detection: Our Edge AI models accurately identified simulated environmental anomalies, showcasing the power of localized intelligence for proactive health monitoring.
- Foundation for Explainable Edge AI: We laid the groundwork for integrating XAI, recognizing its critical importance for trust and adoption in sensitive domains like healthcare, especially when empowering non-technical users.
- Accessibility for Education and Hackathons: By providing a pure Python implementation and detailed documentation within a Jupyter Notebook, we've made this complex technology accessible for learning and rapid prototyping.
What we learned
We gained deep insights into the intricate interplay between low-power wireless communication and edge intelligence. We learned the paramount importance of designing AI with interpretability in mind from the outset, especially when decisions have real-world implications. Furthermore, we reinforced our understanding of the immense potential of such integrated systems to address grand challenges in global health and environmental sustainability.
What's next for Battery-Free Sensing with Edge AI and Explainable AI (XAI)
Our next steps involve:
- Hardware Prototyping: Moving from simulation to physical prototypes, integrating actual environmental sensors with ABC tags and deploying the Edge AI on low-power microcontrollers (e.g., ESP32, Raspberry Pi Zero).
- Real-World Data Collection & Model Refinement: Collecting extensive real-world environmental data from clinics and homes to train and validate our AI models, improving their robustness and generalization.
- Enhanced Explainable AI (XAI) Features: Implementing and evaluating advanced XAI techniques (like SHAP/LIME) on the edge device to provide richer and more nuanced explanations for detected anomalies.
- User Interface Development: Creating intuitive dashboards and alert systems for healthcare workers to easily understand and act upon the insights generated by the system.
- Scalability and Network Optimization: Exploring mesh networking for ABC tags and optimizing data routing to support larger deployments across communities.
- Ethical Deployment Framework: Developing a robust ethical framework for data privacy, consent, and responsible AI usage in sensitive healthcare contexts, particularly in diverse cultural settings.
Log in or sign up for Devpost to join the conversation.