Glen Berseth: foundation models made 1960s robot tasks cheap, but performance is still the bottleneck
GlenBerseth · x · 2026-09-16
Robotics researcher Glen Berseth published a piece tempering enthusiasm around foundation models for robots.
- Historical framing: Minsky's lab in the 1960s already had a robot arm using vision to understand object geometry, plan paths, and assemble parts — today's foundation models solve the same class of problem, just far cheaper and faster to set up.
- Core argument: the goal shouldn't only be moving the cost curve left; performance must advance — object variety, robustness to the wild, multi-robot/human-robot interaction, real-time control, and better hardware ("human hands are still impressive"). He quotes Jitendra Malik: "robotics is also about dexterity and dynamics."
- Self-driving cars are the clearest example: decades of effort, still hard, because they need real-time closed-loop feedback, not just cheaper static perception + planning.
- Conclusion: the field must answer which performance direction to push, not just celebrate deployment cost reductions.
More from Embodied
- Anthropic previews Model Hardware Standard to let AI agents run lab instruments — ivan_bezdomny · 2026-09-16
- World's first 10,000-unit-scale humanoid robot factory starts production — yogthos · 2026-09-16
- First M5 Ultra Geekbench leak: 52,516 multi-core, ~35% faster than M3 Ultra — jedisct1 · 2026-09-16
- Fei-Fei Li's World Labs is hiring robot learning researchers to scale world models — YunzhuLiYZ · 2026-09-16
- Figure CEO Brett Adcock Shows Off New F.03 Humanoid Robot at Dreamforce — adcock_brett · 2026-09-16
- Berseth: Self-driving shows cost cuts aren't enough — real-time closed-loop control is the hard part — GlenBerseth · 2026-09-16