Figure's Robotics Breakthrough for Humanoids

0:00I just finished recording this full podcast episode and realized that my microphone was not turned on so I get to record it again and maybe I'll be more practiced practice or I'll add a little bit more practice I'll be a little bit more practice I'll be more practice I'll be more practice I'll be more podcasting I'll be more practice I'll be more one of the biggest humanoid robot companies that's competing against Tesla's humanoid robot. There's a handful of others, but Figures is one of the big leading ones right now. They've created basically a massive breakthrough that allows their robots to do generalist work.

0:45Basically, you take it and put it in a new environment and it's able to adapt now. So I want to break down how they're able to accomplish this. They built some really incredible solutions to problems that no one else, or I guess maybe other people have these problems. But before we get into all of that I wanted to mention that if you haven't already, I would love for you to go check out AIbox.ai, which is my very own startup where you can get access to over 90 different AI models and you can work with them all in the same chat thread. If you're sick of paying subscriptions, $20 a month to 10 different platforms for audio, image, video generation, you can get all of the best models from Google, OpenAI, Anthropic, all in one place.

1:22And if you use something like Claude for working something like Claude Cowork, you can actually get it to generate images right inside of Claude or video or audio with the AI Box connector. So it starts at$8.99 a month. We have higher tiers if you have really big projects, but I hope this is something that saves you a ton of time. I recently had a project where I got it to generate 1 ,300 images right inside of Claude. It was able to do the whole project for me. I didn't have to mess around with it because it was using the AI Box connector. And when it did that, I saw a massive spike in traffic to my website from that product and from that project that it worked on.

1:57So if you want to give it a try, there's a link in the description to AIbox.ai. All right, let's talk about what's going on with Figures. So yesterday, their CEO put out a tweet where he said, hey, we have a massive basically unlock that we're going to be sharing soon. I was on pins and needles, and today we got the drop about an hour ago. And by the way, if you're watching this on or if you're listening on Apple Podcasts, you can watch this on video. We have video on Apple Podcasts, and I'll be sharing some clips from their actual robot, what it's actually doing, et cetera. but for everyone else I'll explain.

2:27So they released a new thing called Helix 2.5. This is their new AI model and a new update to their robots that let it do general purpose cleaning. Meaning in the past, the biggest problem that they had was that they could get their robot to clean a house in a familiar environment that had been trained on, right? It had seen training data of this particular house and with particular objects that had already been trained on. So if you threw a bunch of objects on the floor and the training data had those objects in it, it knew how to pick up those objects off the floor. But if you had new objects, if you had new environments that it was unfamiliar with, it was unable to do this.

3:02So there's kind of this generalized issue where you could get it trained on specific environments, but it didn't go to all environments. They came up with a really creative solution, which I will explain in a second. But first I want to tell you the results of the solution. The results of the solution were that they were able to take this and they did a big update to end to test this out. They rented 30 different Airbnbs across the Bay area. So all homes that this robot, that the figure robot had never seen before, had never been trained on before. And they threw new objects in the home. So they would throw toys all over the floor.

3:34They would go mess up the beds, beds I had never seen. They would go put piles of like laundry down. It was able to go clean the house, pick up all the toys, make all the beds, fold all the laundry in completely different environments where it had never seen these different objects before. So this is the generalized cleaning. They were able to do this on 30 different homes to validate that this model actually works. How are they able to get this result? This is what I was really excited about. So the way that these models get their training data typically and traditionally has been specifically for these robots.

4:05There's a couple different ways. One of them is through motion capture. So essentially that's a person wearing a special kind of suit. You use the same technology for a lot of animation, but it's got little dots on the suit and you even have special gloves you can wear. And you go and do a task with a camera watching how you do everything with all of the sensors so it understands exactly how you move. That can transfer to a robot really well. The problem is what I mentioned earlier, where it's just specific environments with specific objects that it's really good at. And if you wanted to get better, you'd need a much bigger data set.

4:36So Figure actually went to a bunch of different suppliers and requested data sets from them. One of the biggest is probably Scale AI that you might be familiar with. Scale AI is a company that was purchased by Meta. And they, I think were purchased, about 50 % of the company was purchased for$2 billion. They had the CEO of Scale.ai move over and is now the CEO of Meta.ai. And so this was a huge company that did a lot of training data for Open.ai, Anthropic, Google, and a lot of Meta and a lot of other players. And of course, there's all the drama where when they had this whole acquisition, and Open.ai said, hey, look, we're not buying any more data from them because they're purchased by Meta and it's a competitor, yada, yada.

5:15it. So there's a bunch of stuff that went on. Scale AI specifically, though, wasn't just perhaps buying data. What they were doing was annotating and labeling the data, and they were doing it in bulk. So the reason why their company was worth so much money, they would go and get data. And for example, they would go and get like legal data, then they would have lawyers go and annotate the data and mark it up and say, this thing means this, this thing means this, and kind of go and label it. And they would do this with a lot of different industries, a lot of different types of data. Figure went to companies like Scale AI and tried to say, hey, look, we need a whole bunch of egocentric data.

5:48Egocentric data just means you wear basically a baseball cap and a camera that's pointing down. Usually just your smartphone works. It points down. It has to be able to see your hands. If your hands are outside of the lens, then the footage won't work, but it's going to see your hands and you go and do tasks. So you would go and, you know, clean your dishes. You would go and, you know, do the laundry, whatever. And if your camera can see your tasks, you can use this egocentric footage to train robot models. Companies like Scale AI would go and gather this from people, build little data sets, and sell them.

6:16This is an industry that I'm quite familiar with because I'm working on a project for my own company. It's called fiund.com, F-I-U-N-D. So by the way, if you have any sort of, we specifically work with podcasting, audio, and video, and image data sets. So if you have any of those types of data sets, if you have a podcast you've run for years, let us know, go reach out to me. I'll leave a link in the description to my website, Fiond, and we license that and we help you monetize it if you have data sets. But in any case, Figure went to these companies, lots of third-party vendors, including scale.

6:49And so they said, hey, look, like we need to buy this data. And they said they couldn't get enough quantity. So they were basically bottlenecked by not being able to get enough supply of this data. So what did they do? Figure decided to go and just build their own product, which is called Index. And Figures Index Program is essentially where they will send you like a special camera and hat that you wear. And if you're looking at this on Apple, you can see like what this looks like. It's kind of, it's not really like a VR headset. It's more like a band that goes around your head, but there's a camera on the band.

7:23I mean, it's basically just the way I mentioned it earlier with like the baseball cap and your phone is sort of a crude example that It does work, but if you get bumped the wrong way, it could mess it up. And so with this headband, it's able to just film you, and they'll say, hey, go do your laundry, go do your dishes, and they'll pay you for doing that. Now, there's a lot of different monetization methods with this, and they have a lot of different types of places that are capturing this data for them. But essentially, what they were able to do with this index program is they now have 4 ,000 people a day that are submitting content to their data set.

7:58They couldn't get a big enough data set. So they literally had to go build their own custom solution where you can order one of these headbands, go do your work and you get paid for them. I've heard about programs like this from a number of different companies. There's shift that went viral recently because they said, Hey, look, we'll clean your house for free. If you let the cleaners wear one of these hats with a camera on it so that, you know, it's training data for cleaning. There's a lot of different ways to do this, but at the end of the day, no one was able to get it at the volume that figure needed.

8:23So they built their own solution, the figure index, and now they have 4 ,000 contributors. They took all of that data. I think they said they have 264 ,000 app downloads. They have 16 million videos that have been uploaded from 108 different countries. And they said that they've paid out$15 million to creators. So you could go today and join the index program and make money. As far as the payments, and I mean, I would say go look into this. This is interesting. There's all sorts of tasks people doing, bringing in their groceries, taking out the trash, washing their car, washing their stove, like raking rocks.

8:57I mean, there's all sorts of tasks that picking strawberries on their website, inside of machine shops, folding pizza boxes, doing laundry, making quesadillas, like all sorts of tasks are demoed on their website. So there's a lot of different things that you can do here. That being said, from a lot of the estimates that I was seeing, they're paying like a dollar for a five minute video. So if you're like, you know, if you go take out your trash, it's like a dollar. If you go do this, it's a dollar. In some cases, it could be worth a lot of money. I saw other places that were just kind of paying$20 an hour, which by the way, they would go to find like warehouse workers and say like, Hey, you know, your warehouse worker making 25 bucks an hour, we'll pay you another$20 an hour just to wear this like headband camera around with you while you do it.

9:37And I think a lot of people would take that up. Now they're making 45 bucks an hour. And so it makes, you know, a lot of sense. So figure was able to take all of this data that they now have. And the model is, you know, not being trained. There's not like some big breakthrough and how they're doing it. They were just able to get better data. They put all of this data in and this is basically what the big unlock was for them. So now the figure robots are able to go around and clean houses in completely new environments that they've never seen before. And they've got, they basically unlocked this generalized cleaning.

10:08So where does, or generalized work, where does this go in the future? I think a lot of the data set that they have right now is people around their house, just random people. I think they're going to get more into specialized data sets. This is where they're going to have to focus. So they might say like, Hey, we're going to, it's theoretically possible that they could say, Hey, look, we're saturated on cleaning the house. Our robot is the greatest house cleaner in the world. We don't want any more laundry folding videos. We beg you, please stop. There's a chance it would say that I think they'll just keep collecting data and just making an absolutely astronomical data set.

10:38I think the cost for them is nominal compared to what this, what this could unlock for them. And so, but I think where this will go in the future is they're going to inevitably put a bigger focus on trying to onboard people into their index program that are like mechanics. They've got a bunch of mechanics shops on. Maybe they'll make a deal with O 'Reilly Auto Parts and have O 'Reilly mandate that all of the mechanics need to wear this. Or they'll go in with Walmart shelf stockers. You could imagine that they would have some corporate partnerships that they would create. And then I think they'll just go and find specialized industries where people are doing very specialized things like an AC, an HVAC person, they'll have electricians go.

11:16And eventually with this method, they'll be able to train humanoid robots to be able to do a lot of this work. Where does this go in the future? This is something I've been talking about for about two years. Like if you know me, I've been saying that this is coming, that AI and a chatbot is not the end of it. The end of it is humanoid robots. Because for up until now, robots have always had this problem where, you know, they got wheels on their feet, or they zip around a warehouse picking up pallets and crates and you train the the robot based off of you know basically basically for a very specific task in a very specific environment and it's hard to get it to do much more than that just generalized it's kind of like artificial general intelligence it's we've got that for robots now and how they've been able to unlock it it's just way more training data because these humanoid robots i've also heard people say like surely you could come with a better form factor why are we doing humanoid robots why don't we just make robots that are like special shapes.

12:07It's because of the training data. We can just give them videos about humans do things. And humans have built the entire world for humans, right? Like the way that the height of a doorknob, the way that you open your car, like everything is built for a human. So it would make perfect sense that a robot that's the most useful, it just looks and acts exactly like human. It can be trained off of the exact same data set. So I think they're going to be purchasing more. I mean, they're going to be putting more investments into this. They said specifically on all of this, that they're going to budget over$2 billion a year on compute for training these models and training these robots.

12:40So there's going to be a way, there's going to be some massive investments that are coming into it from that end. And I think we can just expect all of this to accelerate at what these robots are going to be capable of doing and the industries that they will go into. Now, what does this mean? I know a lot of people have kind of said like, I don't know, in alarmist ways, it's kind of the end of jobs. And what are we going to do when the robots take over and they're doing everything for us? This is my opinion is, yes, we'll have a lot of robots doing a lot of things. But at the end of the day, just like chat GPT in and of itself isn't doing your job for you, it's not running your business for you, you're still sitting there managing it.

13:15Even if you have agents doing things, you have to monitor your agents. You have to set up your workflows. You have to optimize them when something breaks. It's going to be the same thing with robots. I think in, you know, at the end of the day that you're always going to need a person and an organization to oversee the robots, to manage what they're doing. I think there's a lot of interesting concepts that I've been talking about for a long time, and it feels like they're getting a lot closer. One of them being, you can imagine something like a landscaping company where you have a Tesla cyber cab, again, getting very, you know, or, or just a Tesla has self-driving.

13:47You have a robot go into there with like, imagine you just give an electric lawnmower to one of these robots. you have it go into a cyber cab and is this economically viable better question but also is all of the tokens that a lot of these companies spend on anthropic credits to rebuild software economically viable there's also a debate there that maybe people are overspending but regardless you could give an electric lawnmower to one of these robots stick it in a cyber cab have it drive around and let's say you had a lawn care business it could go do your lawn care route for you so i mean this is pretty interesting now what what you know who's liable if something goes wrong if it actually mows over your rose bed.

14:22You know, there's like a bunch of theoretical problems where you might want to have a real human there. And maybe you just have your robot to go and do things quickly while you're going and managing and watching it. So I don't know what the oversight of these things are. I don't know how remote control works. I don't know where this gets to, but you can imagine there will be full businesses that are running with these robots to the point where I think there's a lot of, there's a lot of the boomer or there's a lot of the baby boomer generation, right, that is retiring and selling businesses. I mean, it's kind of a famous joke or a famous like thing in private equity where, you know, you're going to go and buy some guys, you know, HVAC company where he's got, or you're going to roll up like 100 HVAC guys companies into like a big HVAC conglomerate in a specific city and you're going to run them all more efficiently, yada, yada.

15:08I mean, there's a lot of these people that have run companies for a long time and they're retiring, you can imagine the new business move here would be go take over someone's HVAC company. And if you don't want to go do preventative maintenance on AC units yourself, you don't want to go and get on the roof of Red Robin with a hose and spray out all their air filters, you could just get a robot to go do that. So you could acquire this company and have the robot go and do a lot of the work. You probably have a camera on that thing where you can review the work at the end of the day, make sure everything looks good, send out the bill.

15:41And it could be like this, a new kind of sort of automated business. And this isn't for every industry. And there's a lot of things that you probably, you know, are very tricky, complex problems that you want human solving. But there's a lot of the mundane stuff that, that this, these robots could actually do. So I think this is a fascinating time to be alive. I mean, undoubtedly there is so much more that's going to be coming out. We're at the cutting edge. I mean, no one has these robots yet. These are just the concepts and the tools that are being built. But you can imagine this is a absolutely colossal.

16:13I mean, when we talk about when we talk about like productivity being unlocked, humanoid robots are going to be it there. I mean, you get the AI, the intelligence from something like Astra or OpenAI or Claude and you stick it into these robots. There is, you know, I mean, that unlocks so much and you kind of get past this issue that I think we have where you need to where you need to hire mass amounts of people to do like low skill labor or low pay labor, right? Like you think of like warehouses and factories. I think a lot of that will be done by robots. So what happens to jobs? What happens to the people that gets displaced by this?

16:50I mean, there's a lot of questions that go into it. But at the end of the day, when the washing machine, when the dishwasher was invented, I'm sure people that were washing it by hand complained. But at the end of the day, now they run the big industrial dishwasher and the restaurant can make more money and have more profits and jobs shift around. So it'll be interesting to see what happens in the shift, but at the end of the day, we will as a society undoubtedly become more productive with all of this. Guys, this is a fascinating episode. Thank you so much for tuning in. If you enjoyed this episode, number one, if you have any sort of data set, if you have a podcast, if you know someone with a podcast, send them my way to fiund.com.

17:30I would love to help evaluate it and help them monetize their data set if that's something that they're interested in. And number two, go check out my startup AIbox.ai if you want to get access to every AI model in one place for$8.99 a month. All right, I'll catch you guys all in the next episode.