ETA — Efficiency through Thinking Ahead

Checkpoints for ETA: Efficiency through Thinking Ahead, A Dual Approach to Self-Driving with Large Models.

Code and evaluation instructions: https://github.com/OpenDriveLab/ETA

Contents

ETA-base/
  config.json       model configuration
  best_model.pt     weights (742 MB)
ETA-async/
  config.json       model configuration
  best_model.pt     weights (1005 MB)

directory	model	parameters
ETA-base	Base Model	371.1 M
ETA-async	Async Model	502.3 M

Usage

Follow the evaluation setup in docs/TRAIN_EVAL.md, then point the evaluator at a checkpoint directory:

python leaderboard/scripts/run_eval_leaderboard.py \
  user=YOURNAME port=30000 trafficManagerPort=20000 \
  experiments=eta viz=0 \
  experiments.ponderer_model_name=ETA-base \
  experiments.agent_root=/PATH/TO/ETA-base \
  experiments.root_path=/PATH/TO \
  experiments.runnickname=myrun \
  checkpoint_file=results.json resume=0

experiments.epoch_num=best is the default and resolves to best_model.pt.

To load the weights directly:

from carformer.ponderer import Ponderer
model = Ponderer.from_pretrained("/PATH/TO/ETA-base", epoch="best").eval().cuda()

Evaluation

Evaluated on the 220 routes of Bench2Drive.

Acknowledgements

This codebase builds on open sourced code from CARLA Garage and Bench2DriveZoo among others. We thank the authors for their contributions. This project is funded by the European Union (ERC, ENSURE, 101116486) with additional compute support from Leonardo Booster (EuroHPC Joint Undertaking, EHPC-AI-2024A01-060). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council. Neither the European Union nor the granting authority can be held responsible for them. This study is also supported by National Natural Science Foundation of China (62206172) and Shanghai Committee of Science and Technology (23YF1462000).

License

Released under the MIT License, matching the ETA repository.

Citation

@article{hamdan2025eta,
  title={ETA: Efficiency through Thinking Ahead, A Dual Approach to Self-Driving with Large Models},
  author={Hamdan, Shadi and Sima, Chonghao and Yang, Zetong and Li, Hongyang and G{\"u}ney, Fatma},
  journal={arXiv preprint arXiv:2506.07725},
  year={2025}
}