Don’t ‘demean’ Tesla’s bots, even if they were piloted by humans, Nvidia robotics boss says

The prototype for Optimus was first revealed in September 2022 

Tesla’s Optimus robots received more skepticism last week when multiple reports confirmed that the humanoid robots working the company’s “We Robot” event were remotely controlled by engineers. Optimus robots reportedly danced, poured drinks, and interacted with attendees.

The realization that the robots were being, to some extent, operated by humans led some analysts and investors to call the display “dishonest” and misleading. Musk’s well-known propensity for hype may have played a part in the reaction.

A top robotics executive at Nvidia said that in this instance, Optimus deserves more credit than scorn.

“I don’t know how much of it was teleoperated or not, but I have to say, even if it is teleoperated, teleoperating a robot with that kind of control requires really sophisticated AI,” Rev Lebaredian, vice president of Omniverse and simulation at Nvidia, told B-17.

Lebaredian has worked for Nvidia for 23 years and leads the company’s Omniverse platform, which is simulation software created so that robots can practice reasoning and take actions without causing damage.

Robotics is Nvidia’s next “zero billion dollar market”, and the company is all in, Lebaredian explained. He acknowledged that all builders in the space have some skepticism to overcome.

“People should be skeptical. That’s natural with any new technology,” he said. But there is a real technological achievement in Tesla’s presentations so far, according to Lebaredian.

“Mapping your controls to the robot acting in the real world and interacting with others around — that’s a huge, huge advancement, and it’s not something that should be downplayed,” he continued.

Seeing robots interact with humans as freely as the coverage of the event showed is an “amazing advancement” even with some level of remote controlling, that shouldn’t be “demeaned,” the Nvidia executive said.

Both Musk and Nvidia CEO Jensen Huang, are focused on humanoid robots as increasingly advanced machine learning offers an unlocking technology for robotics. Musk said at last week’s event that Optimus will be “the biggest product ever of any kind.”

Huang often says the humanoid form factor will eventually allow for the widest adoption since they could one day function in any environment meant for humans.

Lebaredian told B-17 that the humanoid format can also help with the challenge of data collection. Replications of human movement are more readily available than the data needed to train robots in entirely new forms.

To train Optimus, Tesla pays dozens of employees up to $48 per hour to perform tasks in motion capture suits. Musk has claimed that two Optimus robots are performing tasks autonomously on Tesla’s factory floor. But the series of demos that don’t display full autonomy call into question the timeline the Tesla CEO has forecast.

In July, Musk said Tesla would produce a small amount of “genuinely useful humanoid robots” for its own factories and be ready to ramp up production and sell externally in 2026.

For this effort and others at xAI, Musk is a big customer of Nvidia. Earlier this year, Oracle CEO recounted a dinner where he and Musk begged Huang for Nvidia’s graphics processing units.

Huang recently praised Musk for building a 100,000 GPU supercomputer called Colossus for xAI in a matter of weeks. That cluster is intended to support Grok-2, xAI’s chatbot. But, if Optimus is successful, it will require an immense amount of computing power as well.

“If you can teleoperate a robot to behave like that, then you can also imagine, once you have a brain that’s autonomous, it can replace the human who’s doing those same controls,” Lebaredian said. “The robot’s actually capable of doing all those amazing things, whether it’s a human in the loop or an AI in the loop — I think that’s remarkable.”

Tesla did not respond to B-17’s request for comment on the reports of the remote controllers.