RoboICL: teaching robots new tasks via in-context learning instead of retraining
青稞AI · wechat · 2026-09-19
RoboICL explores letting robots learn new tasks on the fly from demonstrations in context, without retraining: show the robot examples → it infers task structure → executes directly. In a sequence-imitation task, the robot observes a series of operations, grasps spatial relations and ordering, then reproduces the process in a new scene.
Combined with LLMs, the large model handles vision → task understanding → action planning → execution → environment feedback, moving robotics from memorizing training data to context-adaptive behavior. The post suggests this could shift training from "one model per task" toward "one general model for many tasks." Blog and GitHub links in the original.
More from Embodied
- DISCO distilled tree search named CoRL 2026 best-paper finalist, top 6 of 687 — yining_hong · 2026-11-04
- Berkeley's Value Forcing: enforcing Bellman consistency at every noise level for accurate RL value functions — berkeley_ai · 2026-11-04
- 22 Creative SAM 3.1 Use Cases From Meta's AI Developer Hackathon, From Wildfire Detection to Document Reassembly — Vjeux · 2026-11-04
- Ex-OpenAI robotics intern breaks down viral Astra robot-painting demo on Cerebras podcast — AccBalanced · 2026-11-04
- Embodied Agent Arena: Closed-Weight VLMs Lead All Five Robot Task Domains, Open-Weight Models Lag — Haojian Huang · 2026-11-04
- PrimeBot's PrimeSim: a generalist world model simulator trained on cross-embodiment demos — Travis Davies · 2026-11-04