| --- |
| license: mit |
| library_name: pytorch |
| pipeline_tag: robotics |
| tags: |
| - autonomous-driving |
| - end-to-end-driving |
| - carla |
| - bench2drive |
| - self-driving |
| --- |
| |
| # ETA — **E**fficiency through **T**hinking **A**head |
|
|
| Checkpoints for **[ETA: Efficiency through Thinking Ahead, A Dual Approach to Self-Driving with |
| Large Models](https://arxiv.org/abs/2506.07725)**. |
|
|
| 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](https://github.com/OpenDriveLab/ETA/blob/main/docs/TRAIN_EVAL.md), then point the |
| evaluator at a checkpoint directory: |
|
|
| ```bash |
| 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: |
|
|
| ```python |
| 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](https://github.com/Thinklab-SJTU/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](https://github.com/OpenDriveLab/ETA). |
|
|
| ## Citation |
|
|
| ```bibtex |
| @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} |
| } |
| ``` |
|
|
