Lifted model checking for relational MDPs

Probabilistic model checking has been developed for verifying systems that have stochastic and nondeterministic behavior. Given a probabilistic system, checker takes property checks whether or not the holds in system. For this reason, provide rigorous guarantees. So far, however, focused on propositional models where state is represented by symbol. On other hand, it commonly required to make relational abstractions planning reinforcement learning. Various frameworks handle domains, instance, STRIPS Markov Decision Processes. Using settings requires one ground model, which leads well known explosion problem intractability. We present pCTL-REBEL, lifted approach pCTL properties of MDPs. It extends REBEL, model-based learning technique, toward checking. PCTL-REBEL lifted, means rather than grounding, exploits symmetries reason about group objects as whole at level. Theoretically, we show decidable MDPs possibly infinite domain, provided states bounded size. Practically, contribute algorithms an implementation checking, improves scalability approach.