Tesla’s Optimus Robot Hits Human-Level Learning Milestone

Tesla says its flagship humanoid robot is learning more like a real human every day.

A recent breakthrough unlocked the ability for the Tesla Optimus to learn complex tasks directly from Internet videos of humans performing them, initially from first-person perspectives. This means a single AI robot brain can control many different tasks without needing people to manually teach them each step.

“If Optimus can watch videos, you know, YouTube videos or how-to videos or whatever and based on that video, just like a human, can learn how to do that thing,” Elon Musk said in an interview in CNBC. “Then you really have task extensibility that is dramatic.”

Video content has become the preferred way to learn over the last two decades after Sal Khan popularized the concept with Khan Academy. A 2024 survey by TechSmith, which created the video editing software Camtasia, found that 83 percent of people prefer watching videos over reading text or listening to audio when accessing instructional or informational content.

The rise of video-sharing platforms like YouTube and TikTok fostered a creator-driven educational ecosystem that appears poised to spill over into robotics.

Tesla just shared new footage of Optimus robots at its Fremont factory independently executing a variety of household and industrial tasks with skills learned from the video-based approach.

The robots also showed off their recently enhanced human-like movements at Tesla’s Austin Texas plant , during a visit by US Transportation Secretary Sean Duffy. A next-generation Optimus robotic hand, with a reported 22 degrees of freedom (DoF) can be seen in the background while a Tesla Bot looks on. Their discussion centered mostly around autonomous driving.

Tesla Optimus robot with US Secretary of Transportation Sean Duffy and Elon Musk at Tesla's Austin, Texas factory
Tesla Optimus robot with US Secretary of Transportation Sean Duffy and Elon Musk at Tesla’s Austin factory

The advancements come as Elon Musk shifts his focus back to his technological pursuits after a controversial run as head of the Trump White House’s Department of Government Efficiency (DOGE). Musk has admitted that his political activities have hurt Tesla’s bottomline after he took a public-facing role for Trump’s successful 2024 campaign.

He’s intensifying his focus on artificial intelligence and robotics as the US tech sector ramps up efforts to stay competitive amid China’s rapid advancements. The Chinese robotics industry has nearly doubled its market share in just four years, rising from 30 percent in 2020 to 50 percent today. As it seeks to lead the humanoid robotics market by 2027, the Chinese government has approved a state-backed venture capital fund of $138 billion to drive innovation.

Unitree H1 humanoid robots dance during the CMG 2025 Spring Festival Gala on Chinese TV
Unitree H1 humanoid robots dance during the CMG 2025 Spring Festival Gala on CGTN

Tesla recently filed paperwork in China to trademark its humanoid robot’s name. The formal legal move suggests Tesla is preparing for future sales of Optimus robots in China, where they’d compete with perceived robotics leaders like Unitree, UBTECH Robotics, EngineAI, AgiBot and rising EV manufacturer Xpeng.

Tesla's petition to trade mark Tesla Optimus in China
Tesla’s petition to trademark Tesla Optimus in China

In the US, their competition includes the well-funded startup Figure AI, whose flagship humanoid robot, the Figure 02, recently completed a 20-hour run of back-to-back shifts on BWM’s X3 production line at its Spartanburg, SC plant.

Figure 02 robot working on the BMW X3 production line in footage shared by Figure AI
Figure 02 robot working on the BMW X3 production line in footage shared by Figure AI

As noted by The Humanoid Hub on the X social media platform, Elon Musk previewed how the Tesla Bot’s learning would evolve on Lex Fridman’s podcast back in November 2023.

“Optimus will figure out how to do things by watching videos,” Musk said at the time.

The EV maker has hired workers to wear motion capture suits to train its Optimus robots. To transfer skills, they wear VR headsets while performing specific tasks so the humanoids can learn through imitation. The employees, referred to as Data Collection Operators, are paid about $48 an hour.

As Tesla moves beyond the labor-intensive approach, engineers are now working toward expanding its learning to third-person videos. Ultimately, they want Optimus to figure out how to fix problems on its own by watching YouTube how-to videos.

This is trickier because camera angles are different for every video, so the robot must figure out which motions are important and how to map them to its own body. Optimus would also need to determine which video is most relevant for the given task.

Musk has predicted that Optimus will be Tesla’s most profitable product of all time, emphasizing its transformative potential in both industrial and consumer applications. He predicts that billions of humanoid robots will greatly outnumber humans in the not-so-distant future, facilitating a post-scarcity economy.

Tesla aims to manufacture 5,000 Optimus robots in 2025, with most of the production toward the end of the year. Initial Optimus units will be for internal use. External sales are expected around the second half of 2026, with Tesla employees getting first access.

That data, combined with videos, years of real-world driving data from Tesla’s autonomous vehicle programs, serves as the foundation for training Optimus through reinforcement learning. Digital twins of Optimus robots train in simulations where they figure out how to do things through trial and error.

Tesla transfers that knowledge to physical robots via a process called Sim2Real, short for Simulation to Reality. Variations of the method have been adopted by most of today’s humanoid robotics firms, from garage startups like Palo Alto-based K-Scale Labs to Boston Dynamics.

Tesla’s Optimus has a neural network running directly on its hardware that makes real-time decisions like adjusting its grip or changing its walking path. It’s how the robot in a recent demo could navigate challenging terrain with its vision system disabled. The system is never perfect, and companies are refining methods to close what’s known as the Sim2Real gap.

Sim2Real concept illustration
Problems with robots crossing from simulation to reality has become known as the “Sim2Real gap”

As Unitree founder Wang Xingxing reportedly explained at a recent industry event in Shanghai, the lack of a mature, unified end-to-end AI system is the industry’s biggest challenge. The breakthrough would facilitate what some call the iPhone moment and others call the ChatGPT moment in robotics. The pressure is on for companies to get there, sooner than later, before investors lose patience.

Though he envisions a future where personal robots like the Tesla Bot facilitates an age of abundance, Musk cautioned during the recent Saudi US Investment Forum in Riyadh that a dystopian future is also possible.
The manufacturer is positioning Optimus as a general-purpose robot that can do everything from household chores to factory work to social companionship.

Tesla announced its intent to make fake humans in 2021, and the first prototype emerged the following year.
The automaker unveiled its second-generation iteration in 2023 with a slimmer design, improved hands, and more agile movements.