Implementation of the Interactive Nested Tree Monte-Carlo Planning (I-NTMCP) algorithm for online planning in I-POMDPs.

This repository contains all the code used for the "Online planning for Interactive POMDPs using Nested Tree Monte Carlo planning" paper. While we have done our best to make it as user friendly as possible it should be treated as a prototype only and for verifying results. We hope to release a more robust implementation in the future.

I-NTMCP is implemented using Python 3.9.

As with any python seperate python library, project, etc, we recommend installing I-NTMCP in a seperate virtual environment such as those provided by Conda.

1. Download or clone the repo.
2. Install I-NTMCP using PIP by navigating to the i-ntmcp root directory (the one containing the setup.py file), and running:

pip install -e .

# use the following to install all dependencies (including for search tree
visualization, pygraphviz)
pip install -e .[all]

I-NTMCP can be run using the run script provided: run_scripts/run.py. This script takes the following compolsory as arguments:

1. the environment name - rc for the Runner-Chaser domain, and pe for the Pursuit-Evasion domain.
2. the name of the policy/algorithm to use for agent 0
3. the name of the policy/algorithm to use for agent 1

The available policies are:

1. NST, NestedSearchTree - this is the I-NTMCP algorithm
2. random - a uniform random policy
3. PESPPolicy - the shortest path policy for the Pursuit-Evasion problem
4. RCNestedReasoningPolicy - the finite nested reasoning policy for the Runner-Chaser problem

The script also takes additional arguments depending on the problem and policy/algorithm being run. These additional arguments can be viewed using the --help flag.

For example, the different size grids for the Runner-Chaser problem can be run using the --grid_name flag with and either cap3, cap4, or cap7.

IROS paper results and other files can be found in the iros_paper directory. For more details for each environment, including instructions for re-running the experiments, see the README files in the iros_paper/rc and iros_paper/pe directories.

Jonathon Schwartz - Jonathon.schwartz@anu.edu.au (primary author) Ruijia Zhou - Ruijia.zhou@anu.edu.au Hanna Kurniawati - Hanna.kurniawati@anu.edu.au

MIT © 2022, Jonathon Schwartz