ZETA: A Controlled Study of Zero-Shot Cross-Embodiment VLA Transfer for Tabletop Manipulation

CoRL 2026

📄 Paper (arXiv) · 🌐 Project Page

Overview

ZETA studies how vision-language-action (VLA) models transfer to unseen robots for tabletop manipulation. Our controlled benchmark spans 14 held-out target embodiments across simulation and real-world evaluation, examining four factors: state-action representations, source-embodiment diversity, auxiliary co-training, and target-embodiment exposure.

We distinguish strict zero-shot transfer, where the target robot is absent from all training data, from pretrain-exposed zero-shot transfer, where it appears only during pretraining. See the paper for details and the project page for results and robot rollouts.

Latest Update

• 2026-09-06: 🎉 ZETA has been accepted to CoRL 2026! The paper and project page are now released.

Citation

@article{yan2026zeta,
  title={ZETA: A Controlled Study of Zero-Shot Cross-Embodiment VLA Transfer for Tabletop Manipulation},
  author={Yan, Mi and Zhang, Wenhao and Zhang, Zhiqi and Peng, Yu and Wang, Tangxinyu and Zhai, Lingfei and Su, Jiayi and Deng, Shengliang and Peng, Lin and Liu, Yaowei and Chen, Yuxing and Wei, Zhiyuan and Wang, Jilong and Chen, Jiayi and Lyu, Jiangran and Zhang, Zhizheng and Wang, He},
  journal={arXiv preprint arXiv:2609.02546},
  year={2026},
  url={https://arxiv.org/abs/2609.02546}
}