Learning Markov State Abstractions for Deep Reinforcement Learning

https://arxiv.org/abs/2106.04379

An older version of this paper was presented at the 2020 NeurIPS workshop on Deep Reinforcement Learning -- paper available here. Click here for the corresponding code release.

git clone --recurse-submodules https://github.com/camall3n/markov-state-abstractions.git

The visual gridworld experiments and the DeepMind Control experiments were originally developed in separate repositories, so it's better to use a separate virtual environment for each. The installation instructions can be found in the respective README files in each experiment folder:

• Visual Gridworld: markov_abstr/gridworld/README.md
• DeepMind Control: markov_abstr/dmcontrol/README.md