Why Is Physical AI a Venture-scale Business?

Bilal Zuberi says physical AI is a more venture-like model because one model can do a job and hundreds of other things, so the solution generalizes. A robot built only to pick an object up and put it down is not a very big venture-scale business.

Robotics used to be deterministic. A robot arm had to be almost hard coded for the grab, the pressure, the move, and where to put the object down, and every potential path lived in the code base, roughly like if-then statements. Robotics then became probabilistic with AI. Real-time vision models, robotic control models, world models, and VLAs identify an object, choose a path, and figure out in real time how to grab it without crushing it. The change made it much easier to solve more interesting problems and to generalize the solution. A bottling company or a glass filling company still needs the work done, but a single pick-and-place robot is not the venture-scale idea.

Kiva Systems, the early robotics company Amazon bought, followed markers on the ground. The robots were not super intelligent, but they did very good automation work. The newer picture is robots interfacing with the world and working in conjunction with humans, hopefully without smacking people or killing people, beyond driving a car inside an airport while already knowing exactly where the lanes are. A lot of capital has come into the space because the markets are more accessible.