Submarine Tracking with Reinforcement Learning

Sponsored by: The US Navy, Maritime Patrol and Reconnaissance Weapons School

 

Abstract
The US Navy performs missions where they seek and track submarines using Sonobuoys on a regular basis. Deployed in the water, Sonobuoys use underwater sounds to identify the general location of a submarine. However, human error and maneuvers performed by the submarines make this mission more difficult. Our project utilizes a Reinforcement Learning agent using Deep Q-Learning to assist in controlling an existing simulator built by the Navy and provides suggestions about when and where to drop Sonobuoys. Community Benefit: Better assistance in locating and tracking submarines whose intentions are malicious.

 
 
Team Members

Jacob Perrone
Carlos Pinto