NVIDIA GR00T N1.7 fine-tuned with just 50 demos beats pi0.5 by ~10% on VLA-Replica

YuXiang_IRVL · x · 2026-09-20

The VLA-Replica team evaluated NVIDIA's GR00T N1.7 VLA model after fine-tuning it with only 50 demos. It performed roughly on par with the top out-of-distribution model MolmoAct2 and about 10% better than pi0.5 on their benchmark. The leaderboard is live, with details in the linked post.

Original post →

More from Embodied

Embodied channel →