Even as artificial intelligence keeps breaking new ground in software, the real test is still playing out in the physical world. Teaching machines to function reliably in real environments remains far more difficult than generating text, images, or code. And according to Dr. Jim Fan, the head of Nvidia’s robotics business and co-head of the GEAR Lab, the robotics industry may be investing its energy in the wrong places.
In a recent post on X, Fan reflected on the past year in robotics and delivered a blunt assessment: despite constant buzz and rapid experimentation, the field is still fragmented and messy. In his view, robotics development hasn’t settled into a clear, effective path forward—and that lack of focus is slowing meaningful progress.
His criticism highlights a key reality about physical-world AI: robots don’t just need “intelligence” in the abstract. They must be able to perceive and interpret 3D space, understand depth and distance, and recognize objects from different angles under changing lighting conditions. They also have to manipulate items with precision—gripping, lifting, rotating, placing, and adjusting their force in real time without breaking or dropping things.
On top of that, robots need an intuitive grasp of physical rules that humans take for granted. Balancing, friction, weight distribution, momentum, and unexpected collisions all come into play. A task that looks simple in a demo can become extremely complex when attempted in a cluttered, unpredictable environment like a home, warehouse, factory floor, or hospital.
Fan’s comments also point toward another underappreciated challenge: building capable robots requires huge investments of time, talent, and resources. Unlike digital AI, where training data can be scraped or synthesized at scale, physical robotics often demands expensive hardware, carefully designed environments, safety measures, testing cycles, and continual iteration. That makes misdirection especially costly.
While he didn’t frame it as doom-and-gloom, the message is clear: robotics is not progressing in a straight line, and the industry may need to rethink its priorities if it wants to turn impressive prototypes into dependable real-world machines. As interest in humanoid robots, automation, and embodied AI accelerates, Fan’s critique is likely to resonate with anyone watching the space closely—because the next big leap won’t come from hype alone, but from solving the hard, physical problems that software can’t ignore.
