LoRA fine-tune of
pi0 on 60 real Franka teleop trajectories. This is the pi0 half of a
model comparison: the
pi0.5 counterpart
was trained on the same data with the same recipe, so the two differ only in the model.
The pi0.5 sibling also ships 10k–50k, so any rung compares like-for-like.
The best real-robot rung is not assumed to be the last one; sweep the ladder on the robot.
1@misc{peng2026maniguard,
2 title = {{MANIGUARD}: A Benchmark and Data Suite for Specification-Grounded
3 Safety Evaluation and Improvement of Robotic Manipulation},
4 author = {Peng, Yiyan and Wang, Philip and Zhan, Simon Sinong and Lyu, Yiqi
5 and Ni, Zhenyang and Yan, Jixin and Wong, Fiorelli and Jiao, Ruochen
6 and Yin, Hang and Cao, Xinyu and Shao, Huajie and Li, Manling
7 and Zhang, Ruohan and Zhu, Qi},
8 year = {2026},
9 eprint = {2608.17386},
10 archivePrefix = {arXiv},
11 primaryClass = {cs.RO},
12 url = {https://arxiv.org/abs/2608.17386},
13}
The fine-tuned weights derive from a Physical Intelligence openpi base model whose
VLM backbone is PaliGemma; use of these weights is therefore subject to the
Gemma Terms of Use (including the
Gemma Prohibited Use Policy),
which downstream users must pass on. The openpi training code and ManiGuard's own
contributions are Apache-2.0.
