FactorSim: Generative Simulation via Factorized Representation

NeurIPS 2024

Fan-Yun Sun, Harini S I, Alex Zook, Jonathan Tremblay, Logan Cross, Jiajun Wu, Nick Haber

Simulation Generation

Refer to the README under the directory factorsim.

$ cd factorsim
$ ./go.sh GAME_NAME

Zero-shot transfer experiments

Refer to rl_training/rl_train.sh.

To train RL policies on the PLE environments ("ground-truth" environments used in the paper), run

$ cd rl_training
$ ./rl_train.sh pong ppo gt --train_on_ple

Miscellaneous

To export video trajectory of a policy

$ cd rl_training
$ python -m utils.export_video pong

Acknowledgement

• PLE: A Reinforcement Learning Environment
• RLLib