Rodney Brooks: co-founder of iRobot and longtime MIT roboticist has a blunt message for investors betting billions on humanoid robots: the returns won’t match the hype. He says the buzziest approach right now: trying to “teach” dexterity by feeding robots videos of people doing tasks is, in his words, “pure fantasy thinking.” Human hands are immensely complex, with tens of thousands of touch receptors, and robotics lacks the deep, decades-long data pipelines that enabled breakthroughs in speech and image recognition. In short, we don’t have the right data & sensors to copy human manipulation in general settings.

Safety is another hard blocker. Bipedal humanoids pump large amounts of energy into just staying upright; when they fall, they fall hard. Scale them up, and the energy (and risk) grows nonlinearly. That makes real-world deployment in crowded, human environments a much tougher sell than glossy demos suggest.

Brooks’s near-term prediction flips the trend: the most successful “humanoids” in the next 10–15 years won’t look human at all. Expect practical machines on wheels, with multiple arms and purpose-built sensors designed for factories, warehouses, and other structured spaces instead of vaguely human shapes walking around the office.

He also points to a pattern we’re already seeing in AI: tools that feel faster don’t always make teams faster. A recent developer study he cites found engineers took longer with AI assistants despite believing they sped up. Robotics, he argues, faces a similar perception gap. Impressive clips don’t equal scalable productivity in the field.

Where the money is going:
Big-name humanoid efforts have landed heavyweight partners and capital, but Brooks doubts those resources will shortcut physics, sensing, and data realities. Even strong teams that pair cutting-edge AI with hardware still have to solve touch, manipulation, safety, and unit economics before mass production makes sense.