Skip to content

(ICCV 2021, Oral) RL and distillation in CARLA using a factorized world model

License

Notifications You must be signed in to change notification settings

dotchen/WorldOnRails

Repository files navigation

World on Rails

teaser

Learning to drive from a world on rails
Dian Chen, Vladlen Koltun, Philipp Krähenbühl,
arXiv techical report (arXiv 2105.00636)

PWC

This repo contains code for our paper Learning to drive from a world on rails.

ProcGen code coming soon.

Reference

If you find our repo or paper useful, please cite us as

@inproceedings{chen2021learning,
  title={Learning to drive from a world on rails},
  author={Chen, Dian and Koltun, Vladlen and Kr{\"a}henb{\"u}hl, Philipp},
  booktitle={ICCV},
  year={2021}
}

Updates

  • We have released the pre-computed Q values in our dataset! Check DATASET.md for details.
  • Checkout our website for demo videos!

Getting Started

  • To run CARLA and train the models, make sure you are using a machine with at least a mid-end GPU.
  • Please follow INSTALL.md to setup the environment.

Training

  • Please refer to RAILS.md on how to train our World-on-Rails agent.
  • Please refer to LBC.md on how to train the LBC agent.

Evaluation

If you evaluating the pretrained weights, make sure you are launching CARLA with -vulkan!

Leaderboard routes

python evaluate.py --agent-config=[PATH TO CONFIG]

NoCrash routes

python evaluate_nocrash.py --town={Town01,Town02} --weather={train, test} --agent-config=[PATH TO CONFIG] --resume
  • Use defaults for RAILS, and --agent=autoagents/lbc_agent for LBC.
  • To print a readable table, use
python -m scripts.view_nocrash_results [PATH TO CONFIG.YAML]

Pretrained weights

Dataset

We also release the data we trained for the leaderboard. Checkout DATASET.md for more details.

Acknowledgements

The leaderboard codes are built from the original leaderboard repo. The scenariorunner codes are from the original scenario_runner repo. The waypointer.py GPS coordinate conversion codes are build from Marin Toromanoff's leadeboard submission.

License

This repo is released under the MIT License (please refer to the LICENSE file for details). The leaderboard repo which our leaderboard folder builds upon is under the MIT License.