August 2023

This repository holds the code for our paper, "Reinforcement Learning with Graph Neural Networks Enables Zero-Shot Deceptive Path Planning Over Arbitrary Graphs", which we are submitting to AAMAS 2024.

Run pip install -r requirements.txt to install the required packages.

Run the file train_for_deceptiveness.py to train a model. This will output a model to the checkpoints folder, which you can use in experiment.

We have populated some models in the models/sage_ambiguity_2 and models/sage_exaggeration_4 folders. You can render animations of their performance or compare different levels of deceptiveness statically by running continuous_sim.py.

For example:

python3 continuous_sim.py --deception-type exaggeration --action animate --seed=51

Renders an animation of an exaggeration-tuned model on a random graph with seed 51 to the file animation.mp4.