Shielding in Resource-Constrained Goal POMDPs

We consider partially observable Markov decision processes (POMDPs) modeling an agent that needs a supply of certain resource (e.g., electricity stored in batteries) to operate correctly. The is consumed by the agent's actions and can be replenished only states. aims minimize expected cost reaching some goal while preventing exhaustion, problem we call resource-constrained optimization (RSGO). take two-step approach RSGO problem. First, using formal methods techniques, design algorithm computing shield for given scenario: procedure observes prevents it from might eventually lead exhaustion. Second, augment POMCP heuristic search POMDP planning with our shields obtain solving implement present experiments showing its applicability benchmarks literature.