Skild releases S1, a robot foundation model that learns tasks from video demos
deepakpathak · x · 2026-08-26
Skild has released S1, a robot foundation model that takes task specifications via video demonstrations rather than language instructions. S1 handles unseen, long-horizon tasks up to 10 minutes with dozens of steps (e.g., potting plants, pour-over coffee) and can recover from mistakes or imperfect demos. It translates visual demonstrations into its own body context, addressing the limitations of language for delicate or complex tasks.
More from Embodied
- Skild AI unveils S1: robots learn 10-minute novel tasks from a single video — Haoyu_Xiong_ · 2026-08-26
- Q-Planning boosts robot self-improvement from 25% to 80% success — chris_j_paxton · 2026-08-26
- Engineer Swaps SystemVerilog for LLM-Generated PALASM Code — satnam6502 · 2026-08-26
- Google Research unveils AgentHands: interactive hand gestures for grounded agents in XR — Google Research · 2026-08-26
- Peking Univ. & MSRA Introduce BCP: Boosting Robot Success Rates via Autonomous Replanning — jiqizhixin · 2026-08-26
- Pemba, the first robot to attempt Everest, launches this fall — 0xAllen_ · 2026-08-26