Tesla Robot RIVAL Livestream: 5 Autonomous AI Robot GLITCHES? ($650,000 MECHA)

Ajouté : 2026-05-17

[00:00:00]Is this fully autonomous? And if so, what's happening here? Today on AI News, we're breaking down Figure 3's 8-hour live stream into the most curious moments. Plus, a brand new Mecca robot. But first, Figure just ran this 8-hour live stream to increase the odds that something would break, and it did. But before showing you that, the three-step workflow for figure 3 is to first detect the barcode, second to pick up the package, and third to reorient it so that the barcode is face down on the conveyor belt, which requires the robot to reason using camera pixels alone. And for speed context, humans average about 3 seconds per package when sorting, which puts Figure 3 around human parody, as you'll see here. But figure states these robots are working fully autonomously thanks to its Helix 2 in-house neural network that's running entirely on board, meaning AI inference is done on the robot itself.

[00:01:00]Furthermore, multiple humanoids areworked together to be able to swap out with one another to keep the system running 24/7.

[00:01:11]In fact, each robot works for 3 to four hours until its battery is low, at which point it autonomously requests another robot to swap in and continue where it left off, as you can see here, allowing the previous robot to autonomously charge itself in the meanwhile. And on top of this, Figure 3's multi-root coordination system acts as an autonomous failover strategy. So, now let's look at the five possible failures and glitches. Starting with number one, you see him pulling out these packages and all of a sudden he glitches out and he stands up straight. And this is where he's probably resetting his sensors and joints.

[00:02:01]And one curious thing about this is that you see that the flow of packages has stopped. So, can he control the upstream conveyor belt as well for when he does these resets? Then you see him standing here readjusting his body position to continue on with the task, searching for a grip, presumably using the cameras on his hands.

[00:02:30]And then you see he continues forward first with this yellow package and he's on his way.

[00:02:44]Number two, we have this example of what would probably be a failure. You see this box on the left falls right off the conveyor belt. And let's show that one more time. He's getting ready to move this white package and accidentally swipes it off.

[00:03:02]And example number three is at package 2496 and 683. It looks like a glitch, but the CEO said that the robot is actually touching its head in order to avoid hitting its other arm on the metal chute. And you see here in both scenarios, the same thing occurring.

[00:03:29]And number four, at package 4,150, the robot seems to be self-correcting when it hits uncertainty, which triggers this joint and sensor recalibration loop you see here, where it resets and continues onwards.

[00:03:48]And number five, you see here, he lifts up his arm to avoid the shoot. And then watch this red package here.

[00:03:57]Assuming this is, he reaches for the black package, but accidentally swipes off the red one. And then he realizes it, recalibrates, and prepares for continued work.

[00:04:26]And as a teaser, they're already starting the figure 4 as they've reached design lock and transition from the planning phase to the shipment of parts.

[00:04:35]And the CEO described this next generation as their most significant engineering leap to date, calling it a mind-blowing advancement over the figure 3. But in another engineering feat, Unitry just released a video of its brand new GD01 mecha robot, which looks to be about 12 feet tall and looks a bit scary to climb into. What if this thing fell over? It weighs 500 kg and apparently the cost is $650,000.

[00:05:08]And the company says that you shouldn't upgrade this thing to be able to carry weapons or do anything dangerous is what they're alluding to. As you can see here that this could be used for purposes that could be potentially unsafe. So user discretion is advised. But you can see it's here working remotely and it can even do this strange crab walk.

[00:05:38]It would be nice to see how the human operator should sit in it. Or maybe this is just something that's controlled remotely.

[00:05:48]So, comment down below what exactly you would use this machine for, if anything, and what could be some cool upgrades. Meanwhile, Generative AI just got an upgrade with Anthropic's agent view for clawed code, which provides a dashboard to monitor and manage multiple AI agents simultaneously. Plus, users can initiate background tasks using the /bg command, allowing the terminal to remain available for other work. And the interface displays session statuses such as running or waiting while allowing users to monitor or reply to specific agents. And the orchestration layer focuses on improving efficiency during multifile refactors or autonomous debugging. And finally, Google just unveiled Gemini Intelligence, which is a suite of AI driven upgrades designed to transform Android from a reactive operating system into a proactive intelligence system. and it focuses on automating tedious tasks and deepening the integration of the Gemini AI model across Chrome and Gboard with its highlights including intelligent autofill where it leverages personal intelligence so that it can automatically complete complex web forms by pulling relevant data from connected apps and Google services and this is an opt-in feature. Then there's task automation where Gemini can handle multi-step workflows like identifying a syllabus in Gmail and automatically adding the required textbooks to a shopping cart. Plus, there's Gboard Rambler, which is a new voice to text tool that converts natural messy speech like filler words and self-corrections into polished, concise messages while maintaining the user's personal style. There's also Smarter Chrome, which is a new auto browse feature that allows Gemini to navigate websites and book appointments. And finally, there's custom widgets, which is a create a widget tool that allows users to generate personalized mini apps with all of these features debuting in the summertime on Samsung Galaxy S26 and Google Pixel 10 before expanding onto other Android powered devices. Anyways, like and subscribe and check out this video here for more on the latest in AI and robotics news. And thanks for watching.