Project
FireAIDSS
An AI-driven drone swarm that monitors and predicts the 4D progression of wildfires in the field, in real time.
- Role
- Lead, full stack
- Dates
- Sep 2024 – Sep 2026
- Stack
- PyTorch, PINNs, custom drones
- Recognition
- ISEF 3rd Grand Award
FireAIDSS is an end-to-end system for wildfire monitoring and prediction. A swarm of customized drones collects temperature and wind measurements over an active fire, and an AI model reconstructs the full thermofluidic field from those sparse samples, giving responders a live 4D picture of where the fire is and where it is heading.
The reconstruction model combines attention, convolutional layers, and physics-informed loss terms derived from the governing equations, reaching a temperature MAE of 2.97 K and a wind velocity MAE of 0.08 m/s. On the hardware side I customized the drones, built the multi-agent sensing platform and data pipeline, and designed a feedback-based swarm search strategy inspired by operator theory and particle swarm optimization that improved search efficiency by 78.5% over back-and-forth sweeps.
The system was validated against more than 240 logged simulation and field runs and has been deployed in fire stations in Shanghai. It earned a Third Place Grand Award at the Regeneron International Science and Engineering Fair.
