Physics-informed deep learning models for wildfire detection

Wildfire tracking and mitigation is a critical challenge in this world. The Machine Learning & Autonomy for Diverse Domains (MADD) Lab is advancing a hybrid approach that fuses classical physics-based models with modern neural networks to equip drone swarms with early wildfire detection and suppression capabilities. Our prior work includes a low-cost, single-drone system using commercial off-the-shelf (COTS) sensors such as thermal and RGB cameras for high-altitude fire detection. On the software side, we are leveraging foundation models to generate realistic synthetic datasets that address the scarcity of controlled wildfire data and enable rigorous evaluation of perception algorithms.

In Fall, the student will conduct a systematic literature review of emerging neural network methods that integrate strong physics priors for modeling synthetic wildfires in controlled environments. In Spring, the student(s) will prototype a single neural-network design and evaluate it extensively using established lab metrics and benchmarks.

Students will build core research skills for robotics and autonomy in challenging domains. Example tasks include collaborating in small teams; gaining hands-on familiarity with lab hardware and sensors (UAV platforms, thermal and RGB cameras, IMUs, flight controllers, radio links); surveying state-of-the-art deep learning approaches (e.g., diffusion models, transformer architectures, text-to-image and text-to-video models); implementing models in PyTorch and using OpenCV and related computer-vision libraries; curating datasets and writing data loaders; performing code reviews and using Git for reproducibility; and contributing to experiment design, metrics,  and documentation.

Please submit the following as a single PDF to my email:

• Why do you want to be a part of the MADD lab and in particular, what are your interests in this project.
• Relevant experience (courses, projects, labs) and your technical contributions.
• Weekly availability (hrs/week) and any conflicts.
• Technical writing sample (s) from a prior class project.
• Unofficial transcript.

Interviews will be held after the initial review of applications.