Skip to main content
  1. Home
  2. Computing
  3. News

The Tesla bot isn’t superhuman yet, but it can make dinner

Add as a preferred source on Google
Tesla's Optimus robot stirring an oven pot.
Tesla

A week ago we saw Tesla’s Optimus robot showing off some nifty dance moves. This week, you can watch it performing a bunch of mundane tasks, though admittedly with a great deal of skill — for a humanoid robot.

Instructed via natural language prompts, the so-called “Tesla bot” is shown in a new video dumping trash in a bin, cleaning food off a table with a dustpan and brush, tearing off a sheet of paper towel, stirring a pot of food, and vacuuming the floor, among other tasks. 

I’m not just dancing all day, ok pic.twitter.com/rKs5WwKjvf

— Tesla Optimus (@Tesla_Optimus) May 21, 2025

The performance may not shake the world of humanoid robotics to its core, but it nevertheless shows the kind of steady progress that Tesla engineers are making, with the bot’s actions and movements becoming evermore complex.

Recommended Videos

Commenting on the latest clip, Optimus team boss Milan Kovac said in a post on X: “One of our goals is to have Optimus learn straight from internet videos of humans doing tasks.” Just to be clear, that doesn’t mean the robot will literally watch videos like a human. Instead, it suggests that the robot will learn from the vast amount of data available in those videos, such as demonstrations of tasks, movements, or behaviors.

Kovac said that his team recently had a “significant breakthrough” that means it can now transfer “a big chunk of the learning directly from human videos to the bots (1st- person views for now),” explaining that this allows his team to bootstrap new tasks much more quickly compared to using teleoperated bot data alone.

Next, the plan is to make Optimus more reliable by getting it to practice tasks on its own — either in the real world or in simulations — using reinforcement learning, a method that improves actions through trial and error.

Tesla boss Elon Musk, who has spoken enthusiastically of Optimus ever since the company first announced it in 2021, has claimed that “thousands” of the robots may one day be deployed alongside human staff at Tesla factories, taking care of “dangerous, repetitive, [and] boring tasks.”

The company, better known for making electric cars than humanoid robots, is racing against a growing number of tech firms globally that are intent on commercializing their humanoid robots, whether for the workplace, home, or perhaps some entirely new human-robot ecosystems yet to be imagined.

Trevor Mogg
Contributing Editor
Not so many moons ago, Trevor moved from one tea-loving island nation that drives on the left (Britain) to another (Japan)…
Dell UltraSharp 52 Thunderbolt Hub Monitor review: A workstation in a single display
A massive 52-inch display, Thunderbolt docking and thoughtful productivity features make the Dell UltraSharp 52 much more than just another oversized monitor.
Computer Hardware, Electronics, Hardware

See at Dell

Quick Recap

Read more
Google’s new Pixel Tag is slimmer than an AirTag and smarter at finding your stuff
Google didn’t copy Apple’s playbook after all.
Wood, Accessories, Bag

I'll admit, tile-style trackers rarely excite me, primarily because most just copy the AirTag and call it a day. Google's first swing at the category breaks that pattern a little. 

The Pixel Tag launched today as the slimmest tracker on the market, backed by a Bluetooth trick neither Apple nor Samsung currently offers.

Read more
Datamaxxing is the latest AI trend and for once, it could be healthy for your health goals
Feeding fitness data to AI can turn unexplained health metrics into useful conversations, but trusting it as a doctor is another matter
Woman exercises with her Apple Watch and Dexcom G7.

Datamaxxing sounds like the sort of wellness trend that would involve a spreadsheet, three supplements, and far too much free time. The basic idea is much more sensible. People are feeding data from their wearables into AI chatbots and asking them to explain what all those numbers actually mean. The Wall Street Journal recently documented several people building systems around exactly that idea.

There’s a genuine gap for AI to fill here. A 2026 Nature Communications study found that wearables are good at producing summaries, but much less useful when people want personalized answers about their own data. An AI agent built by the researchers reached 84% accuracy on objective numerical questions.

Read more