Geo-Informed Machine Learning Agents for Search and Rescue

Karagozian and Case, Inc. seeks to develop, integrate, and test the feasibility of utilizing machine learning for personnel recovery and isolated personnel evasion within the field of search and rescue through implementing a novel machine learning approach that draws from the large dataset called the International Search and Rescue Incident Database, also known as ISRID. Partnering with Dr. Robert Koester, from dbS Productions that manage ISRID, K&C will isolate verifiable, and generalizable data that is essential in any machine learning methodology. Instead of performing supervised learning on this dataset, K&C plans on reframing the problem of personnel recovery and evasion within the domain of reinforcement learning. Not only is reinforcement learning incredibly data efficient, but it also complements the problem space. Through using a carefully constructed reward metric based on the Extended Ensemble Kalman Filter, K&C will be able to accurately capture the underlying behavior of lost persons without introducing any heuristics. Additionally, the proposed methodology will be able to isolate specific behaviors for specific groups of personnel or underlying environment type. The final resulting trained agents will be integrated and deployed within the LandSAR framework.