Here is a fun chicken-and-egg problem for your Friday: how do you train a humanoid robot when you don't yet have a fleet of humanoid robots out there generating data?
On Tesla Inc (NASDAQ:TSLA)'s third-quarter earnings conference call, Elon Musk said the company may need roughly 10,000 to 20,000 Optimus robots to generate training data through "self-play," in which robots practice tasks and use the resulting experience to improve.
To be clear, this is not a production target or a confirmed prerequisite for deployment. Musk estimates the 20,000 figure is what the company may need to create a training loop for a machine that does not yet have a vast fleet generating real-world data.
Tesla Cars Have a Data Head Start
Tesla's autonomous-driving program benefits from information gathered by vehicles already on the road. Optimus does not have that same starting point, Musk said on the call.
"A big difference with Optimus versus the cars is we don't have this vast fleet of cars from which we can gather training data," he said. Instead, Tesla needs to build robots and have them learn by doing.
Ashok Elluswamy, VP of AI at Tesla, reinforced the distinction: "For the car, we actually have the data from the cars' embodiment itself, but for Optimus, there's no existing fleet to get into."
That creates a different scaling challenge. More cars can provide more driving examples as customers use them; Optimus may first need a substantial population to collect examples of robots handling physical tasks.
Robot School
Musk's latest comments outline a proposed approach, not proof it already works at scale.
According to the billionaire, Tesla's "Optimus Academy" is in the works. The company would let robots practice a wide range of tasks, then use the data in a "flywheel" to improve performance across those tasks over time.
For investors, the 10,000-to-20,000 figure is less a near-term sales forecast than a marker of the infrastructure Tesla may need before Optimus can improve rapidly. Key milestones will show robots performing more tasks, repeatable progress from training, and a clearer timeline for deploying enough machines to support that process.
