Embodied Model Leads on RoboTwin 2.0

omarsar0 · x · 2026-07-11

A post shares a set of embodied learning results: an average score of 93.6 on RoboTwin 2.0, leading π0.5 and previous VA models, with only a 0.6-point difference between clean and randomized scenarios. The model can also adapt using just 10–15 demonstrations and transfer across different embodiments.

Related event: LingBot-VA/VLA 2.0 Released: Native Embodied Foundation Model(24 posts)→

Original post →

More from Embodied

Embodied channel →