Humanoid robots won’t just look and talk like humans — they’ll also have to go to college to pick up some skills.
In a recent conversation with Dwarkesh Patel on a podcast, Elon Musk outlined Tesla’s strategy for training its Optimus humanoid robots, revealing plans for what he calls an “Optimus Academy” — a facility where tens of thousands of robots will learn through self-play and real-world practice. The discussion highlighted a critical challenge that sets humanoid robotics apart from autonomous vehicles: while Tesla has nearly 10 million cars on the road generating training data, robots face both greater physical complexity and far sparser real-world data to learn from.
When asked about leveraging Tesla’s AI infrastructure for Optimus, Musk emphasized the commonalities: “A bunch of things that we’ve done for the car are applicable to the robot. So we’ll use the same Tesla AI chips in the robot as the car. We’ll use the same basic principles. It’s very much the same.”
However, Patel pressed on a fundamental difference: cars have relatively simple actuators — steering, acceleration, braking — while humanoid robots possess “dozens and dozens of degrees of freedom,” especially in their arms. More significantly, Tesla’s autonomous driving system benefits from “millions and millions of hours of human demo data collected from just the car being out there,” whereas “you can’t equivalently just deploy Optimus that don’t work and then get the data that way.”
Musk acknowledged this as “an important limitation and difference between cars,” noting that “we do have, so we’ll soon have 10 million cars on the road. And so that’s hard to duplicate that massive training flywheel for the robot.”
His solution represents a hybrid approach to machine learning at scale: “What we’re going to need to do is build a lot of robots and put them in kind of an Optimus Academy so they can do self-play in reality. So we’re actually building that out. We can have at least 10,000 Optimus robots, maybe 20 or 30,000 that can do that, that are doing self-play and testing different tasks.”
But physical robots alone aren’t enough. Musk revealed that Tesla will also leverage its simulation capabilities: “Tesla has quite a good reality generator, a physics accurate reality generator that we made for the cars. We’ll do the same thing for the robots and actually have done that for the robots. So you have a few tens of thousands of humanoid robots doing different tasks. And then you can do millions of simulated robots in the simulated world and you use the tens of thousands of robots in the real world to close the simulation to reality gap.”
The Optimus Academy concept represents a pragmatic acknowledgment that humanoid robotics can’t simply replicate the data advantages of Tesla’s vehicle fleet. Instead, Musk is proposing a controlled environment where thousands of robots can safely fail, experiment, and learn — much like how AI systems have been trained through self-play in constrained environments like Go or chess, but extended to the physical world. Another advantage of having these thousands of robots practice their skills is that the learnings of any any robot can be instantly transferred to the rest of the fleet. The combination of real-world robot experimentation with massive-scale simulation mirrors the approach that’s proven successful in autonomous driving, adapted for the unique challenges of humanoid manipulation and mobility. This strategy could prove crucial as multiple companies — from Boston Dynamics to Figure AI — race to bring capable humanoid robots to market, with the training methodology potentially becoming as important as the hardware itself.