Distillation of RL Policies with Formal Guarantees via Variational Abstraction of Markov Decision Processes (Technical Report)
This work addresses safety-critical applications by providing formal guarantees for advanced RL algorithms, which is an incremental improvement over existing methods like DeepMDP.
The paper tackles the problem of verifying and simplifying reinforcement learning policies in continuous environments by deriving bisimulation bounds between the unknown environment and a learned discrete latent model, enabling formal verification methods and policy distillation with provable guarantees.
We consider the challenge of policy simplification and verification in the context of policies learned through reinforcement learning (RL) in continuous environments. In well-behaved settings, RL algorithms have convergence guarantees in the limit. While these guarantees are valuable, they are insufficient for safety-critical applications. Furthermore, they are lost when applying advanced techniques such as deep-RL. To recover guarantees when applying advanced RL algorithms to more complex environments with (i) reachability, (ii) safety-constrained reachability, or (iii) discounted-reward objectives, we build upon the DeepMDP framework introduced by Gelada et al. to derive new bisimulation bounds between the unknown environment and a learned discrete latent model of it. Our bisimulation bounds enable the application of formal methods for Markov decision processes. Finally, we show how one can use a policy obtained via state-of-the-art RL to efficiently train a variational autoencoder that yields a discrete latent model with provably approximately correct bisimulation guarantees. Additionally, we obtain a distilled version of the policy for the latent model.