Official Implementations of "Near-Optimal Policy Identification in Robust Constrained Markov Decision Processes via Epigraph Form"

This repository is the official implementation to reproduce the figures in Near-Optimal Policy Identification in Robust Constrained Markov Decision Processes via Epigraph Form.

Please see solve-rcmdp.ipynb , algorithm-concept.ipynb, and grad-conflict.ipynb.

Requirements

# make sure you are in RCMDP-Epigraph/
poetry install

Run Experiments and Plot the results

Run all the cells in

• algorithm-concept.ipynb.
• grad-conflict.ipynb.

For the algorithm comparison, run the following notebook

• solve-rcmdp.ipynb
  • and then plot the results by plot-all.ipynb