Code for the rfPG algorithm and the experiments in the paper: "Robust Finite-Memory Policy Gradients for Hidden-Model POMDPs" (IJCAI 2025 main track)

This repository contains code (in the .zip) and a virtual image (in the .tar) in the form of a Docker image to run the experiments in the paper.

For future reference and potential updates to rfPG, please refer to the GitHub repository used (https://github.com/marisgg/synthesis/tree/ijcai). An extended and standalone version may be added (or pointers to it). For example, you can now run rfPG for optimizing reachability probabilities (in addition to reachability rewards). Note that it is not well-tested.

EDIT:

Use the following command to execute the code after loading the Docker image from the .tar:

docker run  -v "$(pwd):/opt/payntdev" -v "/opt/payntdev/payntbind/" --name YOURCONTAINERNAMEHERE -it localhost/rfpg:ijcai python3 entrypoint.py