Elon Musk has outlined plans for an 'Optimus Academy' to train thousands of Tesla's Optimus humanoid robots through self-play and simulations. In a recent podcast interview, he highlighted the challenges of robot training compared to autonomous vehicles and reiterated a solar energy mandate for Tesla and SpaceX. The initiative aims to bridge the gap between simulated and real-world performance.
Elon Musk, CEO of Tesla Inc., discussed his vision for advancing humanoid robotics during a podcast interview with Dwarkesh Patel on Thursday, joined by Stripe co-founder John Collison. Musk introduced the concept of an 'Optimus Academy,' a dedicated facility where Tesla plans to deploy at least 10,000 Optimus robots, potentially scaling to 20,000 or 30,000, to engage in self-play and test various tasks in a real-world environment.
Musk emphasized the differences between training Optimus and Tesla's autonomous vehicles. While the company benefits from nearly 10 million cars on the road generating vast amounts of driving data, humanoid robots present greater complexity with 'dozens and dozens of degrees of freedom,' particularly in their arms, and limited opportunities for real-world data collection. 'You can’t equivalently just deploy Optimus that don’t work and then get the data that way,' Musk noted, acknowledging this as 'an important limitation and difference between cars.'
To address this, Tesla will leverage its existing AI infrastructure, including the same AI chips used in vehicles and a physics-accurate 'reality generator' originally developed for cars. This will enable millions of simulated robots in a virtual world, with the real robots in the academy helping to close the 'simulation to reality gap.' Musk explained that learnings from any single robot could be instantly transferred across the fleet, drawing parallels to self-play techniques successful in AI games like Go and chess but applied to physical tasks.
In the same interview, Musk reiterated Tesla and SpaceX's commitment to solar energy, stating they have a 'mandate to get to 100 GW a year of solar.' He added, 'We're going as fast as possible in scaling domestic production [of solar cells].' This hybrid training approach could position Tesla competitively in the growing humanoid robotics field, where companies like Boston Dynamics and Figure AI are also advancing.
