Just last week, Tesla released a video of their Optimus humanoid robot dancing. While the moves were impressively smooth (video at 1x speed), I was critical that this wasn’t showcasing the commercial applications that the robot would ultimately be deployed to when available.
Today that changed with a new video of Optimus that directly addressed my concerns. Though the 1 minute and 6 second clip, we see the robot putting trash in the bin, using a vacuum cleaner, pick and place parts from a production line and more.
The idea of the dancing video was to show the robots range of motion and the hardware speed and capabilities. Today’s video gives us a much better insight into how Optimus will learn new skills, a key piece of information that has remained unclear until now.
At the bottom of the video, we see ‘Optimus is learning many new tasks, instructed via natural language’. This is huge. Being able to have the robot perform a task it’s never done before by using natural language, dramatically simplifies the process of learning.
In the example where Optimus places rubbish in the bin example, the prompt is simple “Pick up the trash bag, open the lid and place the bag in the bin”. That may seem a little cumbersome to need to describe the step of opening the lid, I suspect this will be inferred and occur automatically with future models.
Tesla didn’t stop with commercial applications, also showcasing some use cases that may play out at home. “Pickup the spoon and stir the pot”.. Again, in the future I expect a prompt (or voice command) to stir the pot, will be smart enough to understand you would need to use a spoon, rather than a finger or any other possible item to stir something in a pot, on a stove that is likely hot.
Again this shows Tesla Optimus’ dexterity which is really quite impressive and it’s worth noting, we’re still not seeing their final version of the hardware.
From opening and closing blinds, to ironing clothes, tearing off paper towel, to washing dishes, the capabilities here are seriously impressive. Speed is also clearly a work in progress with the video moving between 1.5x and 2x speeds. Given how fast twinkle toes is, I suspect this is also just a matter of more training to improve, rather than any technical limitation.