Why robotics "capability jumps" are just aligning tasks to pretrained LLM skills
robrombach · x · 2026-11-02
elvisnavah argues robotics "shocking" capability jumps happen when robotic skills are aligned to quasi-identical skills frontier models already pretrain on: LLM-as-robot (astra) maps to visual computer use; real-time dexterity (VAM/WAM) maps to video pretraining. Adding computer-use-framed robotics tasks to training explains the visual servoing wins, but dexterity and fast control need baked system-1 weights — this approach defines system 2 for general deployment. Rob Rombach retweeted.
More from Embodied
- Amazon's robot fleet tops 1M as robotic arm count doubles in 2026 — ___Mufasaa · 2026-11-02
- Nous-Girl: an open-source StackChan robot powered by Nous Research's Hermes Agent — alexcovo_eth · 2026-11-02
- Neuralink in four years: from monkey cursor control to paralyzed users typing and speaking by thought — XFreeze · 2026-11-02
- Dev argues sub-vocal speech devices will erode your real voice over time — itsOmSarraf_ · 2026-11-02
- CUHK's CaPtain-4B Post-Trains Small VLMs as Code-as-Policy Robot Agents, Hitting 78.3 on LIBERO-PRO — CUOHK · 2026-11-02
- Stanford's RoboArchitect Auto-Designs Robot Harnesses, Lifting Unseen-Scene Success from 9.7% to 43.3% — Shengguang Wu · 2026-11-02