Skild AI unveils S1: robots learn 10-minute novel tasks from a single video
Haoyu_Xiong_ · x · 2026-08-26
Skild AI introduced S1, a new foundation model built on in-context learning: from just one video prompt, with no fine-tuning, it can be taught 10-minute-long tasks it has never seen before, operating in real time.
Engineer Jared Mejia recounts that the team believed from Skild's founding that ICL was exactly what a truly capable robotic intelligence needed, but repeated attempts failed — manipulation seemed too complex. Earlier this year they finally saw the spark: steerable robot systems completing new tasks beyond their training data. He's now convinced ICL is not only possible but the most promising path to scaling to thousands of tasks, scenarios, and environments.
More from Embodied
- Niantic Spatial's places library ships robot-ready Gaussian splat warehouse scenes for policy training — RexDouglass · 2026-10-16
- Volvo and NVIDIA expand partnership to bring safe automated driving to next-gen cars — nvidia · 2026-10-16
- Tevel's robotic system harvests fruit around the clock using computer vision — LexiLove · 2026-10-16
- SHOES: 10 minutes of real-robot RL lets quadrupeds walk in shoes, CoRL 2026 paper — kscottz · 2026-10-16
- Tesla Cybercab robotaxis logged zero crashes in first two weeks of public rides, NHTSA data shows — elonmusk · 2026-10-16
- Robotics LLMs tested on 16 tasks: GPT-6 Astra hits 17.2% success vs Fable's 9.1% — pabbeel · 2026-10-16