We propose "Efficiency through Thinking Ahead" (ETA), an asynchronous dual-system that pre-processes information from past frames using a large model in tandem with processing the current information with a small model to enable real-time decisions with strong performance.
[2025/06/10]ETA paper and code release!
To get started with ETA:
- Training
- Evaluation
- ETA Training code
- ETA Evaluation
- Inference Code
- Checkpoints
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).
This project is released under the MIT License. If you find our project useful for your research, please consider citing our paper with the following 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}
}