As if Elon Musk needed yet another reason to hate OpenAI. Figure, a startup that’s partnered with OpenAI to develop a humanoid robot, released a new video on Wednesday. And it’s truly heads above anything Tesla has demonstrated to date with the Optimus robot.
The video from Figure, which is available on YouTube, shows a human interacting with a robot dubbed Figure 01 (pronounced Figure One). The human has a natural-sounding conversation with the robot, asking it to first identify what it’s looking at.
“I see a red apple on a plate in the center of the table, a drying rack with cups and a plate, and you standing nearby with your hand on the table,” Figure 01 explains in a human-like tone.
The human asks for something to eat, and Figure 01 proceeds to hand him the apple. The human then introduces some trash into the environment and asks the robot to both explain why it just did what it did, while also picking up the trash. There is, of course, some lag time as the robot processes the instructions, but it otherwise executes the command perfectly.
“On it. So I gave you the apple because it’s the only, uh, edible item I could provide you with from the table,” Figure 01 says, complete with a disturbingly human “uh” punctuating the middle of the sentence.
For one last trick, the human asks Figure 01 what should happen next based on the placement of the dishes on the table. The robot says the dishes are likely to go in the drying rack next. The human verbally validates the assumption and asks Figure 01 to do just that, demonstrating incredible dexterity and precision in picking up the cup first and flipping it upside down before putting it in its place. Then another demonstration of precision comes when Figure 01 puts the plate in an appropriate ridge on the drying rack.
It’s really worth watching the full video, embedded below, to get a proper understanding of what’s happening.
You may be asking yourself, what’s the big deal? Sure, Musk announced his robot in 2021 with someone dressed in a robot costume, but hasn’t Tesla made incredible progress since then that could rival OpenAI and Figure? Not really.
Tesla’s Optimus can walk around, pick up an egg, and even fold laundry. That’s pretty impressive, right? Only if you’ve taken Elon Musk’s robot demonstration videos at face value. Take a closer look at that laundry folding video from January, for instance. Did you notice anything weird in the lower right corner of that supposedly impressive video?
Still don’t see it? How about if we add a big red arrow? Do you notice how the hand that comes in and out of view matches perfectly with the movements of Optimus’s right hand?
Yes, that appears to be someone using teleoperation to make movements that are then mimicked by Optimus. A human is essentially folding the shirt through remote manipulation, or what’s sometimes referred to as a waldo.
Humans have been making waldo robots since at least the 1940s, and putting teleop capabilities in human-like robots since at least the 1960s. If we’re being honest, the tech that Tesla’s Optimus demonstrates isn’t much more impressive than the audio-animatronics that Disney made for the 1964 New York World’s Fair, as you can see below.
The accidental revelation that came out of Tesla’s video, provided you were looking closely, is why Figure and OpenAI are essentially light-years ahead with their robot. Because Figure 01 isn’t using any tricks like teleoperation, as Figure’s co-founder Brett Adcock confirmed on Wednesday.
“The video is showing end-to-end neural networks. There is no teleop,” Adcock wrote on X. “Also, this was filmed at 1.0x speed and shot continuously.”
Does Musk and the Tesla team have the ability to figure this stuff out? Absolutely. But Adcock and the folks at Figure have taken a big lead, as the video from Wednesday clearly demonstrates. And Figure has assembled a team of people from companies like Boston Dynamics and DeepMind who clearly know what they’re doing.
Corey Lynch, who works on AI at Figure, was clearly proud of his work in a tweet on Wednesday where he described the latest advancements, while also noting Figure 01 isn’t teleoperated. That repetition by employees at Figure that their robot isn’t being operated by a human can only be interpreted as a dig at Musk, who recently sued OpenAI.
“Even just a few years ago, I would have thought having a full conversation with a humanoid robot while it plans and carries out its own fully learned behaviors would be something we would have to wait decades to see,” Lynch wrote. “Obviously, a lot has changed.”
It sure has changed, even from the relatively primitive days of 2015 when Darpa’s robot competitions were still using teleoperated machines. And we’ll have to wait and see if Musk’s team can catch up—to 2015.