Skild AI Unveils S1: Robot Model Learning Tasks from One Video Demo

DeryaTR_ · x · 2026-08-26

Skild AI introduced S1, a robot foundation model designed to learn new manipulation tasks via in-context prompting. It can execute 10-minute tasks it has never seen before using just a single video prompt without any fine-tuning. In internal benchmarks, S1 achieved a 66% step-success rate on unseen tasks compared to 9% for a language-prompted model, with one video demonstration providing roughly the benefit of 380 post-training examples.

Related event: Skild Unveils S1: Robots Learn New Tasks From a Single Video(27 posts)→

Original post →

More from Embodied

Embodied channel →