Wasserstein Auto-encoded Markov Decision Processes
Official implementation of the Wasserstein Auto-encoded MDP (WAE-MDP) framework. The latter enables the distilation of (any kind of) reinforcement learning policies into simpler controllers which are paired with a discrete, tractable model of the environment (a so called latent space model). The two are provided with bisimulation guarantees, which allow formally verifying the behaviors of the agent operating under the simplified policy. The source code provided allows for replicating the experiments of the paper Wasserstein Auto-encoded MDPs: Formal Verification of Efficiently Distilled RL Policies with Many-sided Guarantees.
- Wasserstein Auto-encoded MDPs: Formal Verification of Efficiently Distilled RL Policies with Many-sided Guarantees
- Wasserstein Auto-encoded MDPs @ ICLR 2023
- The Wasserstein Believer: Learning Belief Updates for Partially Observable Environments through Reliable Latent Space Models
- WAE-PCN: Wasserstein-autoencoded Pareto Conditioned Networks