Nobody taught it to flip pancakes: Skild's S1 learns new tasks from one demo
lukas_m_ziegler · x · 2026-08-26
Skild AI's new general robot model S1 flipped a pancake nobody taught it. The team assumed pancake flipping must be in the training data and searched millions of hours of pre-training footage—nothing. S1 inferred the whole task from a single human demonstration.
- S1 is built as an in-context learner from the ground up: like an LLM, it never updates weights to learn a new task; the demo enters its context window and the policy acts on it.
- New robot tasks today normally require days of teleoperation plus fine-tuning a specialist policy—S1 skips all of it.
- It completes ten-minute novel tasks (a new coffee style, potting a plant, frying pancakes) by composing primitives learned in pre-training. Soil and pots arrived at 8:54 PM and the robot was soon running the task autonomously.
More from Embodied
- 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
- WEDGE paper generates training environments conditioned on navigation agent failures — _akhaliq · 2026-10-16