Tesla's future robot legions: First they learn to drive, then to walk

Yahoo

20-05-2025

Tesla is taking all the knowledge it's acquired from a decade of autonomous vehicle development to teach the Optimus, its new humanoid robot, how to interact with the world.
Essentially, the robot learns to drive before it can walk — or dance, as a demonstration that CEO Elon Musk posted on social media site X showed May 12. The goal is to create an autonomous robot through artificial intelligence.
Tesla is deploying its Full-Self Driving software to power a robotaxi service scheduled for June to enable the Optimus to move through the world with built-in cameras and an onboard computer.
'Tesla's the leader in real-world AI,' Musk said at a company meeting streamed on YouTube March 20. 'What we learn from the car, we translate to the Optimus robot.'
And if Musk's plan to put millions of robotaxis on the road is adventurous, his promise to deploy billions of Optimus robots with their own Full-Self Driving software is highly optimistic, analysts say.
After all, Musk has promised self-driving cars for a decade without delivering any yet. And robot development is still in its infancy, even if Tesla has a leg up on a variety of competitors in the robot space.
Sign up for the Automotive News Daily newsletter to get the latest top stories in your inbox every business morning.
'There's a ton that we can leverage from autonomous cars and having sort of a head start with all the technology that Tesla has,' said Heni Ben Amor, an Arizona State University professor specializing in robotics.
'But there are some pretty major question marks that need to be resolved,' he said, adding, 'I want to be careful that we're not overhyping things,' with regard to progress on humanoid robots in general.
Unlike robotaxis, humanoid robots will have to balance on two legs, manipulate their hands and fingers in complex ways and interact side-by-side with humans before they can be useful, Ben Amor said.
Tesla already has significant expertise in software, electric motors, power electronics and similar skills that make it a leader in the race for a general-purpose humanoid robot for labor-intensive tasks.
With billions of miles of data and an AI supercomputer in Austin, Texas, Tesla launched an Optimus pilot assembly line in California this year. It plans to make 5,000 robots this year, which Musk likened to a Roman legion.
Tesla plans to first employ the robot in its factories, expanding to outside companies as early as next year. Musk said Tesla could make 50,000 Optimus robots in 2026, scaling to 1 million per year by the end of the decade.
'Ultimately, I think we'll be making tens of millions of robots per year,' Musk said. 'It's like serious volume. Maybe 100 million robots a year.' He said Tesla employees would get first access to the bots for home use.
Some Wall Street analysts forecast that humanoid robot production will eventually overtake the legacy auto industry.
'Over the long term, we project that the market for humanoid robots to be materially larger than the global auto industry,' Morgan Stanley said in an April research note. The investment firm is bullish on Tesla stock.
'Our estimate for $4.7 trillion in global humanoid sales by 2050 is nearly double the aggregate revenues of the 20 largest global auto OEMs in 2024, a figure that could very well shrink,' the bank said.
Tesla is already utilizing what are essentially robot cars for useful work at its Fremont, Calif., factory, Musk said. New vehicles drive autonomously from the end of the production line to a parking lot for shipping.
A key test will come in June when Tesla plans to launch a fully autonomous taxi service in Austin, similar to Waymo, Google's self-driving car company that already operates in several U.S. cities.
In 2026, Tesla plans to launch a full robotaxi, the Cybercab, with no human controls such as a steering wheel or pedals.
Musk wants the Optimus to reach an entirely new level of capability decades down the road.
'The future we're headed for is one where you can literally just have anything you want,' he said. 'If there's a good or service you want, you'll be able to have it. What's key to that is robotics and AI.'
To be sure, it's not just Tesla that's promising autonomous robots for labor-intensive tasks to work in factories, do chores in the home and ultimately transform society.
Boston Dynamics, owned by Hyundai Motor Group, said in April that the automaker will buy tens of thousands of robots in the next few years.
Last year, BMW said it's exploring the use of humanoid robots in production and did a successful test of one model at its South Carolina plant. BMW is partnering with California-based robot company Figure.
But Musk is promising highly aggressive timelines and a vision of the future in which humanoid robots would eventually outnumber human beings and lead to global abundance for every person.
'The only company that has all the ingredients for making intelligent humanoid robots at scale is Tesla,' Musk said. The Optimus will be the biggest product in history and 10 times bigger than second place, he added.
As a well-financed, vertically integrated company with a trillion-dollar market value, Tesla has advantages over its startup rivals in the humanoid robots space, but it also faces difficult technical challenges.
The decadelong development of the Full-Self Driving software is helpful but not decisive, Ben Amor said.
'It definitely helps. All of the gains are useful and some of it is transferable to general autonomy of robots and humanoid robots in particular,' he said.
'But at the same time, there's a big chunk of challenges that are not addressed in autonomous driving for which dedicated algorithms and methods need to be created' for the Optimus, Ben Amor said.
AVs don't interact in the way a humanoid robot in a factory would, for example, balancing on two feet, carefully manipulating its hands and working closely with humans, he said.
'Seeing a human robot as sort of a car on two legs would not do it justice,' Ben Amor said. 'And I think that's kind of an oversimplification when it comes to the physical world.'
As a researcher who served as a visiting scientist at the Google DeepMind laboratory in California, Ben Amor is a cheerleader for robot development. But he warns that too much hype could lead to disappointment.
'From a really well-functioning prototype to something that's deployable and reliable in a company, you actually take five to 10 years and where we are now is we don't have the solution yet,' Ben Amor said.
Optimists on the Optimus include Tesla's former director of AI, Andrej Karpathy, now CEO of startup Eureka Labs. He was involved in the Optimus project before he left Tesla in 2022.
'In terms of transfer from cars to humanoids, it was not that much work at all,' Karpathy said on a No Priors podcast in September.
'One of the early versions of the Optimus robot, it thought it was a car. It had the exact same computer, it had the exact same cameras,' Karpathy said. 'It thought it was driving, but it was moving through an environment.'
Humanoid robots like the Optimus are likely to be first used in-house for tasks like material handling and only later in the refinement process will they be safe enough to interact with humans in a home setting, he said.
'I think a lot of people have this vision of it like doing your laundry, etc. I think that will come late,' Karpathy said. 'I don't think we can have a robot like crush grandma is how I put it. I think it's too much legal liability.'
Have an opinion about this story? Tell us about it and we may publish it in print. Click here to submit a letter to the editor.