We’re in the middle of a new surge of interest in humanoid robots. Startups have raised billions of dollars and are making impressive progress in creating generalized robots that can share environments with humans. But one company that has been making steady progress in the field even before the new wave of excitement around humanoid robots is Tesla.
In a recent interview with the No Priors podcast, Andrej Karpathy, the former director of artificial intelligence at Tesla, explained the key advantages that Tesla has in developing and rolling out its Optimus humanoid robots:
1- Superior models: Tesla has already done a huge amount of research in autonomous driving and has one of the most advanced self-driving technologies. Karpathy, who was still at Tesla at the beginning of the Optimus project, explained that all the knowledge acquired and artifacts created when developing the self-driving car technology transferred to the robots.
“Cars are robots, and Tesla is a ‘robotics at scale’ company,” Karpathy said. The robotics team easily transferred the self-driving models to the robot, and the robot initially thought that it was a car and was looking for driving space. With a little fine-tuning, the models could be adapted to human navigable spaces.
Tesla continues to amass tons of data from its cars, which it uses to update its computer vision stack to make it better at understanding the environment under different conditions. All of those findings will also become useful for the robots.
2- Superior expertise and infrastructure: When looked at from the lens of a robotics company, Tesla has all the right talent and tools to assemble robots cheaply and efficiently. Karpathy said, “The moment Elon said we’re doing this, people showed up with all the right tools… There was so much expertise for building robotics at Tesla.”
This again gives Tesla a leg up in the race. And any efficiency gained from creating the robocars and the robots that build the cars will immediately transfer to the humanoid robot venture. Even the processes such as labeling data will be immediately transferable to Optimus. And Tesla already has a complex and efficient pipeline for labeling its data.
Reusing its existing manufacturing components will enable Tesla to reduce the costs of creating Optimus robots at a much faster rate than other companies. You can see this effect in the impressive progress Optimus hardware has made since it was announced—and every robotics expert I talk to tells me that the difficulty of creating robotics hardware is underappreciated.
3- Superior deployment facilities: Although there has long been a vision of using humanoid robots in homes, we are still far away from that happening. And that is for multiple reasons, including safety (current robots can harm people) as well as economic reasons (would you pay tens of thousands of dollars for a robot that will do the dishes and laundry with mediocre precision?).
The first use cases are likely to be in industrial settings where human labor is exhaustive, expensive, and in short supply. These are settings where the robots will be honed and optimized while also doing productive work and generating revenue or cutting costs. Again, this is an area where Tesla has a huge advantage over its competitors. It has multiple very large factories, where it can deploy Optimus without getting into complicated agreements with outside partners.
“Your first customer is yourself… then you can go B2B and can go to companies with massive warehouses,” Karpathy said. “Once you incubate in multiple companies, you can start looking into B2C applications.”
Why humanoid robots?
There is an argument against general humanoid robots in favor of specialized robots that can do a single task. Karpathy’s answer to this argument is interesting: In the long run, a general platform that can do many tasks will beat many disparate platforms that are designed for specific tasks. So, you buy the humanoid robot as a platform that can constantly be upgraded with new skills and applications, which can eventually be a very cost-efficient solution to many problems.
In fact, in a recent interview at the All-In Summit, Tesla CEO Elon Musk shared his vision for the future of Optimus.
He projects that when manufactured at scale, Optimus will cost less than a small car, or around $20,000. The company will need two to three iterations to refine the robot while also using it in its own factories, which, according to Musk, will take five to six years. Then, they can scale to producing millions of robots per year. The timeline might change, but I think we will eventually move toward a world where humanoid robots become the norm.