|
Dynamic programming with partial information to overcome navigational uncertainty in a nautical environment
C. Beeler, X. Li, C. Bellinger, M. Crowley, M. Fraser, I. Tamblyn Preprint (2023) Making safe navigation decisions under uncertainty is a fundamental challenge, whether for autonomous vessels or other agents operating in partially observable environments. In this work, we use a toy nautical navigation environment to show that dynamic programming can be applied effectively even when only incomplete information about a partially observed Markov decision process (POMDP) is available. By incorporating uncertainty directly into the planning model, we construct navigation policies that maintain safety and outperform traditional dynamic programming approaches that assume full state knowledge. We further demonstrate that adding controlled sensing methods allows these policies to simultaneously reduce measurement costs, showing that smarter observation strategies and better decision-making can go hand in hand. |
