Charles River Analytics was awarded a ~$500K contract from the US Navy under the Office of Naval Research’s Minerva Innovative Naval Prototype (INP) program to develop Probabilistic COA Reasoner Enhanced by Meta Reinforcement Learning (PERCIVAL-MERLIN).
Under PERCIVAL-MERLIN, we are conducting research to formalize course-of-action (COA) modeling in a novel way that allows commanders and analysts to encode realistic and high performing friendly and enemy COAs in both a human-understandable and computationally defined format.
The PERCIVAL-MERLIN system will use this new method to rapidly generate and evaluate candidate friendly and adversary COAs, accurately predicting adversary COAs and delivering the results in a way that makes it easy for commanders to make decisions regarding mission planning, re-planning, and execution.
PERCIVAL-MERLIN uses Charles River’s Neural Program Policies, a novel approach to reinforcement learning, to construct artificially intelligent agent models.
*Above figure: Potential adversary actions will be predicted using a unique reinforcement learning approach within AI agents, and presented to commanders in a way that’s easy to understand
This material is based upon work supported by the Office of Naval Research under Contract No. N00014-20-C-2063. Any opinions, findings and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the Office of Naval Research.
Distribution Statement A: Approved for public release. Distribution is unlimited.