Robotics is not solved: frontier models are ~6 orders of magnitude off real assembly lines
ZeeshanZiaML · x · 2026-11-11
Pushing back on "robotics is solved" claims, Zeeshan Zia compares real assembly-line footage with frontier-lab demos: current tasks are 1/10 the complexity of real lines, best models hit 75% completion, run 3 orders of magnitude slower than humans, and cost 2 orders of magnitude more (token costs alone in an agentic harness) — roughly 6 orders of magnitude from real-world autonomy. Distillation and scaled training data might each add an order of magnitude, still leaving a 4-order gap. He expects Astra and Fable to accelerate simulation-building and training signal, "but they're not taking us to the promised land."
More from Embodied
- Laser-as-a-sensor: monolithic VCSEL oxygen detector shrinks sampling volume to subnanoliter — jwt0625 · 2026-11-11
- Beni robot, #1 personal robot on Kickstarter, officially launches with self-recovery and 4K60 — Scobleizer · 2026-11-11
- Berkeley's Social Humanoid teaches Unitree G1 socially responsive interaction — berkeley_ai · 2026-11-11
- MacBooks do have a hinge angle sensor—verified with Codex, prompt included — lipeng0820 · 2026-11-11
- DynaRobotics robot demo at CoRL sparks 'I want one at home' reactions — JasonMa2020 · 2026-11-11
- Humanoid Factory Deploy Stalls: Teleop Intervention Every 6 Minutes After 15 Weeks — ZeeshanZiaML · 2026-11-11