Roboticist's field notes: LLMs handle grasping, but collision avoidance and dexterity remain hard
YiMaTweets · x · 2026-09-08
Roboticist @Winniechen02 shares three observations from recent agentic robotics experiments:
- Grasping works well: it's essentially static-scene processing, even for dexterous hands, so LLM-driven approaches suffice
- Collision and avoidance are hard: LLMs struggle to solve them without human help, since avoidance relies on implicit mobility experience humans accumulate
- Dexterous manipulation still belongs to imitation learning and RL — though he wonders how long that will last
He closes by asking: beyond data collection, what skills do you master that LLMs can't replace within five years?
More from AGI Musings
- AI2027 Forecast Updated with New Astra Scenario — Oriuke · 2026-09-08
- Researcher challenges scaling orthodoxy: benchmark trends aren't universal laws of perception and reasoning — GeorgiaChal · 2026-09-08
- If AI Helps Crack the Navier–Stokes Millennium Problem, It Would Signal a Scientific Revolution — kimmonismus · 2026-09-08
- "The last genuinely necessary thing you did at work was in 2026 — and that day is under a year away" — rand_longevity · 2026-09-08
- Noam Brown's Cryptic Tweets Spark Speculation About an Imminent Breakthrough — socoolandawesome · 2026-09-08
- Mathematician slams AI benchmark craze: 'a meaningless trophy hunt' — tak3sh8 · 2026-09-08