Figure 03 Humanoid Robot Explained: Future AGI Level AI Assistant ?

Now, walking into the living room to start an end-to-end cleaning task. This is pixel-to-action long horizon thanks to the Helix AI system. The Helix AI system uses a layer 1 which decides what to do and layer 2 decides how to do it. And thanks to its newly redesigned sensory suite and hand system, the robot moves with a new level of human fluidity. This is thanks to the actuators actually being two times faster, as well as the field of view of the cameras being 60% larger, 25% the latency, as well as having two times the frame rate.

And what this means in the real world is that it can ingest more data as a wider field of view, basically more intelligence. It can learn faster. And you can see as a direct result of this, the way that it moves and readjusts itself in real time. This is with no teleoperation. Now we can see it walking to the kitchen to continue this end-to-end task. And tell me whether or not you think that this is going to be waterproof. Because here when it turns on the water, it may or may not get slightly wet here.

Something to note is that these fingers have a new sensor that was built in-house by figure. And as well as being a new sensor, it also provides a softer, wider fingertip for better resolution when touching different objects. In fact, figure 3’s new proprietary sensors can actually feel as little as 3 grams on its fingertips or the equivalent of a paperclip. And altogether, the robot is 9% lighter than the figure 2, which translates to about 63.7 kilograms of body weight. All of this is being powered thanks to figure’s GoBig dataset. This is the largest, most diverse humanoid pre-training dataset, which allows it to carry out these real-world tasks, such as cleaning laundry.

Which brings us to the clothes of the robot. Firstly, it can be removed without any tools just with the zipper. The clothes can be washed in the machine, and they can be customized. Whether they are cut-resistant or fire-retardant, different companies or use cases can use different types of apparel to fit those specifically. And here’s an up-close demo of the robot using its perception system. Specifically, notice how when it grabs inside of the drawer, it would be blocking its own view if not for the cameras built into the palm, which allows it to continue its grasping. It’s more redundant, and it can adjust itself in real-time.

Now, notice that it’s able to press these buttons, which lends to optical character recognition being built into the robot. Next, the robot is seen folding laundry in real-time. And this is particularly difficult because the robot needs to constantly re-evaluate how it needs to make its next move. So its long-horizon task is changing in real-time, and it’s using these upgraded sensor systems in order to feel with its fingertips to know exactly what to do next. But the robot isn’t just for home use. Figure says that they’re actually preparing to produce up to 12,000 of these Figure 3s per year with the goal of producing 100,000 robots over the next four years.

And what they’ve done here is they’ve actually switched from CNC machining to die-casting, plastic injection molding, and stamping in order to achieve a dramatically lower cost profile of all the components. So what this means for you as a consumer is that as these robots continue to be sold, they’ll become cheaper with economies of scale. Here you can see it’s actually carrying out industrial use cases, wearing different garments specific for the task. And something that they also mentioned was that on the side of the robot’s head, you can see what number it is of the fleet for easy identification.

You can also see its toes are actually able to bend forward. This helps with its balance, and its feet use induction charging. Look at these shoes here. Also, the robot features these multi-softness surfaces, and Figure mentions that its soft materials actually prevent pinch points. You can see here that its garment seems to have some type of foam around the edges. And when it comes to the power of the robot, while we don’t know how many milliamp hours the battery contains, we do know that the battery has been upgraded with a new battery management system, cells, packs, and connections.

And it’s even certified by the United Nations 38.3 safety standard for lithium batteries. Now, we can see the way that it moves with all of these improvements put together. It has very fluid-like human movements. This is thanks to the robot being able to adjust in real time for each and every task. It also features a microphone that’s readjusted and four times louder speaker for better speech-to-speech interactions with humans. And here’s how all this new tech translates to a smarter, cheaper robot, not just for the home, but also for the office. Let’s start with the price. So, because Figure has invested in this brand new tooling, the stamping, plastic injection molding, and die casting, and they’re moving away from the CNC machining, it means that they’re going to be able to save maybe between 50-80% on the components.

Then, because of their in-house BotQ production of all their most important, expensive parts, they’re going to further bring down that cost profile, meaning that if they can achieve production of about 12,000 units per year as they intend to, they’ll probably get to that $30,000 to $50,000 range. And if they achieve the 100,000 units by 2030, which would be maybe $25,000 per year, they would almost certainly be able to achieve that $20,000 to $30,000 cost window. But how does the robot get smarter while it gets cheaper? Well, that’s because every single robot that Figure creates is actually adding to the meta-intelligence of the entire network each time that it goes back to charge because it’s transferring all those terabytes of information back to the network and teaching every other robot from its experiences, and they don’t forget.

So, this means that as the fleet expands, the macro-intelligence of the network grows faster and faster. But the question here is, how do you actually pay for this thing once it gets down to that car range of $20,000 to $30,000? Do you pay for it upfront with a loan or some type of a financing plan as you would a car, and then you own all the data, or is Figure going to maybe adopt an approach like cell phones, for instance, where the upfront costs might be low, but you have to continue paying on a monthly basis, for instance, in order to get the intelligence updates, the physical service on the hardware, et cetera, and then they would be able to own the data as well and then monetize that, build out bigger data sets for training, et cetera, really build into their IP interest.

But tell me how much you think this will cost in the rollout of 2026 and whether or not you would buy the first version, the Figure 3, or if you would wait till, say, the Figure 4 to let some of the kinks get worked out. But like and subscribe, check out this video here for more on the latest in AI news, and thanks for watching. Bye. [tr:trw].