File size: 2,800 Bytes
7802818 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 | ---
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}
}
```
|