# Figure's Robotics Breakthrough for Humanoids

An episode of [AI Chat: AI News on Claude, ChatGPT, Gemini, Grok, Perplexity](https://invisiblenetwork.ai/podcast/1669990482), 18 Sep 2026, 18 min.

Audio: https://content.rss.com/episodes/365073/3162922/ai-chat-podcast/2026_09_18_00_07_20_92c4b58d-bb28-4b6b-a2fe-e92a842149b4.mp3

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! In this episode, we explore Figure's latest advancements in humanoid robotics, specifically their ability to perform general tasks in diverse environments. This breakthrough opens up exciting possibilities for the future of automation and human-robot collaboration. Show Links • Get the top 80+ AI Models for $8.99 at AI Box: https://aibox.ai • How I Grow and Scale My Business with AI: https://www.skool.com/aihustle • Monetize your podcast audio and video data with Fiund: https://fiund.com/

## Transcript

[0:00] I just finished recording this full podcast episode
[0:03] and realized that my microphone was not turned
[0:06] on so I get to record it again and maybe I'll
[0:09] be more practiced or I'll add a little bit of
[0:11] extra insights in this hopefully it's even better
[0:13] if I talk really fast I've already said this
[0:15] all once so that's probably why today on the
[0:17] podcast I want to talk about the one of the biggest
[0:19] unlocks in humanoid robotics that we've seen
[0:22] to date a massive technological breakthrough
[0:25] so honestly something that I'm very excited about
[0:27] and this is all coming from Figure, one of the
[0:30] biggest humanoid robot companies that's competing
[0:32] against Tesla's humanoid robot. There's a handful
[0:35] of others, but Figure is one of the big leading
[0:37] ones right now. They've created basically a massive
[0:40] breakthrough that allows their robots to do generalist
[0:44] work. Basically, you take it and put it in a
[0:46] new environment and it's able to adapt now. So
[0:49] I want to break down how they're able to accomplish
[0:51] this. They built some really incredible solutions
[0:53] to problems that no one else, or I guess maybe
[0:56] other people have these problems. But before
[0:58] we get into all of that, I wanted to mention
[1:00] that if you haven't already, I would love for
[1:02] you to go check out AIbox .ai, which is my very
[1:06] own startup where you can get access to over
[1:08] 90 different AI models and you can work with
[1:11] them all in the same chat. If you're sick of
[1:12] paying subscriptions, $20 a month to 10 different
[1:15] platforms for audio, image, video generation,
[1:17] you can get all of the best models from Google,
[1:20] OpenAI, Anthropic, all in one place. And if you
[1:22] use something like Claude for... working something
[1:24] like ClaudeCowork, you can actually get it to
[1:27] generate images right inside of Claude or video
[1:29] or audio with the AI Box connector. So it starts
[1:32] at $8 .99 a month. We have higher tiers if you
[1:36] have really big projects, but I hope this is
[1:37] something that saves you a ton of time. I recently
[1:39] had a project where I got it to generate 1300
[1:42] images right inside of Claude. It was able to
[1:45] do the whole project for me. I didn't have to
[1:46] mess around with it because it was using the
[1:48] AI Box connector. And when it did that, I...
[1:51] saw a massive spike in traffic to my website
[1:53] from that product and from that project that
[1:56] I worked on. So if you want to give it a try,
[1:58] there's a link in the description to AIbox .ai.
[2:00] All right, let's talk about what's going on with
[2:02] Figure. So yesterday, their CEO put out a tweet
[2:05] where he said, hey, we have a massive... basically
[2:09] unlock that we're going to be sharing soon. I
[2:11] was on pins and needles and today we got the
[2:14] drop about an hour ago. And by the way, if you're
[2:16] watching this on, or if you're listening on Apple
[2:19] podcasts, you can watch this on video. We have
[2:20] video on Apple podcasts and I'll be sharing some
[2:23] clips from their actual robot, what it's actually
[2:25] doing, et cetera. But for everyone else, I'll
[2:27] explain. So they released a new thing called
[2:29] Helix 2 .5. This is their new AI model that is,
[2:33] and a new update to their robots that let it
[2:35] do general purpose cleaning, meaning in the past,
[2:38] the biggest problem that they had was that they
[2:40] could get their robot to clean a house in a familiar
[2:43] environment that had been trained on, right?
[2:45] It had seen training data of this particular
[2:47] house and with particular objects that had already
[2:49] been trained on. So if you threw a bunch of objects
[2:51] on the floor and the training data had those
[2:53] objects in it, it knew how to pick up those objects
[2:56] off the floor. But if you had new objects, if
[2:59] you had new environments that it was unfamiliar
[3:00] with, it was unable to do this. So there's kind
[3:03] of this generalized issue where you could get
[3:05] it. trained on specific environments, but it
[3:08] didn't go to all environments. They came up with
[3:10] a really creative solution, which I will explain
[3:12] in a second. But first I want to tell you the
[3:14] results of the solution. The results of the solution
[3:17] were that they were able to take this and they
[3:20] did a big update and to test this out, they rented
[3:22] 30 different Airbnbs across the Bay Area. So
[3:25] all homes that this robot, that the figure robot
[3:28] had never seen before, had never been trained
[3:29] on before, and they threw new objects in the
[3:32] home. So they would throw toys all over the floor.
[3:34] They would go mess up the beds. had never seen
[3:36] they would go put piles of like laundry down
[3:38] it was able to go clean the house pick up all
[3:40] the toys make all the beds fold all the laundry
[3:43] in completely different environments where it
[3:45] had never seen these different objects before
[3:47] so this is the generalized cleaning they were
[3:50] able to do this on 30 different homes to validate
[3:52] that this model actually works how are they able
[3:55] to get this result this is what i was really
[3:57] excited about so the way that these models get
[4:00] their training data typically and traditionally
[4:02] has been specifically for these robots there's
[4:05] a couple different ways one of them is through
[4:07] motion capture so essentially that's a person
[4:09] wearing a special kind of suit you use the same
[4:11] technology for a lot of animation but it's got
[4:13] little dots on the suit and you even have special
[4:16] gloves you can wear and you go and do a task
[4:18] with a camera watching how you do everything
[4:21] with all of the sensors so it understands exactly
[4:23] how you move. That can transfer to a robot really
[4:25] well. The problem is what I mentioned earlier,
[4:29] where it's just specific environments with specific
[4:30] objects that it's really good at. And if you
[4:33] wanted to get better, you'd need a much bigger
[4:35] data set. So Figure actually went to a bunch
[4:38] of different suppliers and requested data sets
[4:41] from them. One of the biggest is probably Scale
[4:43] AI that you might be familiar with. Scale AI
[4:45] is a company that was purchased by Meta. And
[4:49] they, I think, were purchased, about 50 % of
[4:51] the company was purchased for $2 billion. They
[4:54] had the CEO of Scale .ai move over and is now
[4:56] the CEO of Meta .ai. And so this was a huge company
[5:00] that did a lot of training data for OpenAI, Anthropic,
[5:04] Google, and a lot of Meta and a lot of other
[5:06] players. And of course, there's all the drama
[5:08] where when they had sole acquisition and OpenAI
[5:10] said, hey, look, we're not buying any more data
[5:12] from them because they're purchased by Meta and
[5:14] it's a competitor, yada, yada. So there's a bunch
[5:16] of stuff that went on. Scale AI specifically
[5:18] though, wasn't just perhaps buying data. What
[5:20] they were doing was annotating and labeling the
[5:23] data and they were doing it in bulk. So the reason
[5:25] why their company was worth so much money, they
[5:27] would go and get data. And for example, they
[5:29] would go and get like legal data. Then they would
[5:31] have lawyers go and annotate the data and mark
[5:34] it up and say, this thing means this, this thing
[5:37] means this, and kind of go and label it. And
[5:39] they would do this with a lot of different industries,
[5:40] a lot of different types of data. Figure went
[5:42] to companies like Scale AI and tried to say,
[5:45] hey, look, we need a whole bunch of egocentric
[5:47] data. Egocentric data just means you wear basically
[5:50] a baseball cap. and a camera that's pointing
[5:52] down, usually just your smartphone works, it
[5:54] points down, it has to be able to see your hands.
[5:56] If your hands are outside of the lens, then the
[5:58] footage won't work, but it's gonna see your hands
[6:00] and you go and do tasks. So you would go and
[6:02] clean your dishes, you would go and do the laundry,
[6:06] whatever. And if your camera can see your tasks,
[6:08] you can use this egocentric footage to train
[6:10] robot models. Companies like Scale .ai would
[6:12] go and gather this from people, build little
[6:15] data sets and sell them. This is an industry
[6:17] that I'm quite familiar with because I'm working
[6:18] on a project for my own. company it's called
[6:21] find .com f -i -u -n -d so by the way if you
[6:25] have any sort of we specifically Work with podcasting,
[6:29] audio and video and image data sets. So if you
[6:32] have any of those types of data sets, if you
[6:34] have a podcast you've run for years, let us know.
[6:36] Go, you know, reach out to me. I'll leave a link
[6:38] in the description to my website, Fiand. And
[6:40] we license that and we help you monetize it if
[6:43] you have data sets. But in any case, a figure
[6:45] went to these companies, lots of third party
[6:47] vendors, including Scale, and they said, hey,
[6:49] look, like we need to buy. this data and they
[6:52] said they couldn't get enough quantity. So they
[6:54] were basically bottlenecked by not being able
[6:57] to get enough supply of this data. So what did
[6:59] they do? Figure decided to go and just build
[7:02] their own product, which is called Index. And
[7:06] Figure's Index program is essentially where they
[7:08] will send you like a special... a special camera
[7:12] and hat that you wear and um if you're looking
[7:15] at this on on apple you can see like what this
[7:18] looks like it's kind of it's not really like
[7:19] a vr headset it's more like a band that goes
[7:21] around your head but there's a camera on the
[7:23] band i mean it's basically just The way I mentioned
[7:26] it earlier with like the baseball cap and your
[7:27] phone is sort of a crude example that does work.
[7:30] But like if you get bumped the wrong way, it
[7:32] could mess it up. And so with this headband,
[7:35] it's able to, you know, just film you and they'll
[7:36] say, hey, go do your laundry, go do your dishes,
[7:38] and they'll pay you for doing that. Now, there's
[7:40] a lot of different monetization methods with
[7:42] this. And they have a lot of different types
[7:45] of places that are capturing this data for them.
[7:48] But essentially what they were able to... do
[7:52] with this index program is they now have 4 ,000
[7:54] people a day that are submitting to their data
[7:58] set. They couldn't get a big enough data set,
[7:59] so they literally had to go build their own custom
[8:02] solution where you can order one of these headbands,
[8:04] go do your work, and you get paid for them. I've
[8:06] heard about programs like this from a number
[8:07] of different companies. There's Shift that went
[8:09] viral recently because they said, hey, look,
[8:11] we'll clean your house for free if you let the
[8:13] cleaners wear one of these hats with a camera
[8:15] on it so that it's training data for cleaning.
[8:17] There's a lot of different ways to do this, but
[8:19] at the end of the day, no one was able to get
[8:21] it at the volume. that figure needed so they
[8:23] built their own solution the figure index and
[8:25] now they have 4 000 contributors they took all
[8:28] of that data I think they said they have 264
[8:32] ,000 app downloads. They have 16 million videos
[8:35] that have been uploaded from 108 different countries.
[8:38] And they said that they've paid out $15 million
[8:40] to creators. So you could go today and join the
[8:44] index program and make money. As far as the payments,
[8:46] and I mean, I would say, go look into this. This
[8:49] is interesting. There's all sorts of tasks people
[8:50] doing, you know, bringing in their groceries,
[8:52] taking out the trash, washing their car, washing
[8:55] their stove, like raking rocks. I mean, there's
[8:57] all sorts of tasks that... picking strawberries
[9:00] on their website, inside of machine shops, folding
[9:03] pizza boxes, doing laundry, making quesadillas,
[9:07] like all sorts of tasks are demoed on their website.
[9:10] So there's a lot of different things that you
[9:11] can do here. That being said, from a lot of the
[9:13] estimates that I was seeing, they're paying like
[9:15] a dollar for a five minute video. So if you're
[9:17] like, you know, if you go take out your trash,
[9:19] it's like a dollar. If you go do this, it's a
[9:20] dollar. In some cases, it could be worth a lot
[9:22] of money. I saw other places that were just kind
[9:24] of paying $20 an hour, which by the way, They
[9:27] would go to find like warehouse workers and say
[9:29] like, Hey, you know, your warehouse worker making
[9:31] 25 bucks an hour, we'll pay you another $20 an
[9:33] hour just to wear this like headband camera around
[9:36] with you while you do it. And I think a lot of
[9:38] people would take that up. Now they're making
[9:39] 45 bucks an hour. And so it makes a lot of sense.
[9:42] So figure was able to take all of this data that
[9:44] they now have. And the model is, you know, not
[9:47] being trained. There's not like some big breakthrough
[9:49] in how they're doing it. They were just able
[9:50] to get better data. They put all of this data
[9:53] in, and this is basically what the big unlock
[9:55] was for them. So now the figure robots are able
[9:58] to go around and clean houses in completely new
[10:03] environments that they've never seen before.
[10:05] And they've basically unlocked this generalized
[10:08] cleaning. So where does, or generalized work,
[10:10] where does this go in the future? I think a lot
[10:13] of the data set that they have right now. is
[10:15] people around their house just random people
[10:17] i think they're going to get more into specialized
[10:19] data sets this is where they're going to have
[10:20] to focus so they might say like hey we're going
[10:23] to it's theoretically possible that they could
[10:25] say hey look we're saturated on cleaning the
[10:27] house our robot is the greatest house cleaner
[10:29] in the world we don't want any more laundry folding
[10:31] videos we beg you please stop there's a chance
[10:33] it would say that i think they'll just keep collecting
[10:35] data and just making an absolutely astronomical
[10:37] data set i think the cost for them is nominal
[10:40] compared to what this what this could unlock
[10:42] for them and so But I think where this will go
[10:44] in the future is they're going to, you know,
[10:46] inevitably put a bigger focus on like. trying
[10:48] to onboard people into their index program that
[10:50] are like mechanics. They've got a bunch of mechanic
[10:51] shops on. Maybe they'll make a deal with O 'Reilly
[10:54] Auto Parts and have, you know, O 'Reilly mandate
[10:56] that all of the mechanics need to wear this.
[10:58] Or they'll go in with, you know, like Walmart
[11:00] shelf stockers. Or you could imagine that they
[11:03] would have kind of like some corporate partnerships
[11:05] that they would create. And then I think they'll
[11:07] just go and find specialized industries where
[11:10] people are doing very specialized things like
[11:13] an AC, an HVAC person. They'll have electricians
[11:16] go and eventually... With this method, they'll
[11:18] be able to train humanoid robots to be able to
[11:21] do a lot of this work. Where does this go in
[11:23] the future? This is something I've been talking
[11:24] about for about two years. Like, if you know
[11:26] me, I've been saying that this is coming, that
[11:27] AI and a chatbot is not the end of it. The end
[11:30] of it is humanoid robots. Because for up until
[11:32] now, robots have always had this problem where,
[11:34] you know, they got wheels on their feet or they
[11:35] zip around a warehouse picking up pallets and
[11:38] crates. And you train the robot based off of,
[11:41] you know, basically for a very specific task
[11:45] in a very specific environment. And it's hard
[11:48] to get it to do much more than that. It's generalized.
[11:50] It's kind of like artificial general intelligence.
[11:53] It's we've got that for robots now and how they've
[11:55] been able to unlock it. It's just way more training
[11:58] data because these humanoid robots, I've also
[12:01] heard people say like, surely you could come
[12:02] up with a better form factor. Why are we doing
[12:03] humanoid robots? Why don't we just make robots
[12:05] that are like special shapes? It's because of
[12:07] the training data. We can just give them videos
[12:10] about humans do things and humans have built
[12:12] the entire world for humans, right? The way that
[12:15] the height of a doorknob, the way that you open
[12:18] your car, like everything is built for a human.
[12:19] So it would make perfect sense that a robot that's
[12:22] the most useful, it just looks and acts exactly
[12:23] like a human. It can be trained off of the exact
[12:25] same data set. So I think they're going to be
[12:27] purchasing more. I mean, they're going to be
[12:30] putting more investments into this. They said
[12:32] specifically on all of this that they're going
[12:34] to budget over $2 billion a year on compute.
[12:38] for training these models and training these
[12:39] robots so there's going to be a way there's going
[12:41] to be some massive investments that are coming
[12:43] into it from that end and i think we can just
[12:45] expect all of this to accelerate what these robots
[12:48] are going to be capable of doing and the industries
[12:50] that they will go into now What does this mean?
[12:52] I know a lot of people have kind of said like,
[12:54] I don't know, in alarmist ways, it's kind of
[12:57] the end of jobs. And what are we going to do
[12:59] when the robots, you know, take over and they're
[13:00] doing everything for us? This is my opinion is
[13:03] yes, we'll have a lot of robots doing a lot of
[13:06] things. But at the end of the day, just like
[13:09] Chat GPT in and of itself isn't doing your job
[13:12] for you. It's not running your business for you.
[13:14] You're still sitting there managing it. Even
[13:15] if you have agents doing things, you have to
[13:17] monitor your agents. You have to set up your
[13:19] workflows. You have to optimize them when something
[13:21] breaks. It's going to be the same thing with
[13:22] robots. I think at the end of the day, you're
[13:26] always going to need a person and an organization
[13:29] to oversee the robots, to manage what they're
[13:31] doing. I think there's a lot of interesting...
[13:34] concepts that i've been talking about for a long
[13:35] time and it feels like they're getting a lot
[13:37] closer one of them being you can imagine something
[13:39] like a landscaping company where you have a tesla
[13:42] cyber cab again getting very you know or or just
[13:45] a tesla has self -driving you have a robot go
[13:48] into there with like imagine you just give an
[13:50] electric lawnmower to one of these robots you
[13:52] have it go into a cyber cab and is this economically
[13:55] viable better question but also is all of the
[13:58] tokens that a lot of these companies spend on
[14:00] anthropic credits to rebuild software economically
[14:02] viable there's also a debate there that maybe
[14:04] people are overspending but regardless you could
[14:07] give an electric lawnmower to one of these robots
[14:09] stick it in a cyber cab have it drive around
[14:11] and let's say you had a lawn care business it
[14:13] could go do your lawn care route for you so I
[14:16] mean, this is pretty interesting. Now, who's
[14:18] liable if something goes wrong, if it actually
[14:20] mows over your rose bed? There's a bunch of theoretical
[14:23] problems where you might want to have a real
[14:26] human there and maybe you just have your robot
[14:27] to go and do things quickly while you're going
[14:30] and managing and watching it. So I don't know
[14:31] what the oversight of these things are. I don't
[14:33] know how remote control works. I don't know where
[14:35] this gets to, but you can imagine there will
[14:37] be full businesses that are running with these
[14:39] robots to the point where I think there's a lot
[14:42] of... There's a lot of the baby boomer generation
[14:46] that is retiring and selling businesses. It's
[14:49] kind of a famous joke or a famous thing in private
[14:53] equity where you're going to go and buy some
[14:56] guy's HVAC company or you're going to roll up
[15:00] 100 HVAC guy's companies into a big HVAC. conglomerate
[15:04] in a specific city and you're going to run them
[15:07] all more efficiently yada yada i mean there's
[15:09] a lot of these people that have run companies
[15:10] for a long time and they're retiring you can
[15:12] imagine the new business move here would be go
[15:15] take over someone's hvac company and if you don't
[15:18] want to go do preventative maintenance on AC
[15:21] units yourself. You don't want to go and get
[15:23] on the roof of Red Robin with a hose and spray
[15:26] out all their air filters. You could just get
[15:28] a robot to go do that. So you could acquire this
[15:31] company and have the robot go and do a lot of
[15:33] the work. You probably have a camera on that
[15:36] thing where you can review the work at the end
[15:38] of the day, make sure everything looks good,
[15:39] send out the bill. And it could be like... a
[15:43] new kind of sort of automated business. And this
[15:45] isn't for every industry. And there's a lot of
[15:47] things that you probably, you know, are very
[15:49] tricky, complex problems that you want human
[15:51] solving. But there's a lot of the mundane stuff
[15:53] that these robots could actually do. So I think
[15:57] this is a fascinating time to be alive. I mean,
[16:00] undoubtedly, there is so much more that's going
[16:03] to be coming out. We're at the cutting edge.
[16:05] I mean, no one has these robots yet. These are
[16:07] just the concepts and the tools that are being
[16:08] built. But you can imagine this is a absolutely
[16:12] colossal. I mean, when we talk about when we
[16:14] talk about like productivity being unlocked,
[16:17] humanoid robots are going to be it there. I mean,
[16:20] you get the AI, the intelligence from something
[16:23] like Astra, or open AI or cloud and you stick
[16:26] it into these robots. There is, you know, I mean,
[16:29] that unlocks so much and you kind of get past
[16:32] this issue that I think we have where you need
[16:34] to Where you need to hire mass amounts of people
[16:37] to do like low skill labor or low pay labor,
[16:41] right? Like you think of like warehouses and
[16:42] factories. I think a lot of that will be done
[16:44] by robots. So what happens to jobs? What happens
[16:48] to the people that get displaced by this? I mean,
[16:50] there's a lot of questions that go into it, but
[16:52] at the end of the day, when the washing machine,
[16:54] when the dishwasher was invented, I'm sure people
[16:58] that were washing it by hand complained, but
[17:00] at the end of the day, now they run the big industrial
[17:02] dishwasher and the restaurant can make more money
[17:04] and have more profits and jobs shift around.
[17:09] So it'll be interesting to see what happens in
[17:11] the shift, but at the end of the day, we will,
[17:13] as a society, undoubtedly become more productive
[17:16] with all of this. Guys, this was a fascinating
[17:18] episode. Thank you so much for tuning in. If
[17:20] you enjoyed this episode, number one, if you
[17:22] have any sort of data set, if you have a podcast,
[17:24] if you know someone with a podcast, send them
[17:27] my way to findfiund .com. I would love to help
[17:31] evaluate it and help them monetize their data
[17:34] set if that's something that they're interested
[17:35] in. And number two, go check out my startup,
[17:37] AIbox .ai, if you want to get access to... every
[17:40] AI model in one place for $8 .99 a month. All
[17:45] right, I'll catch you guys all in the next episode.

Shows are listed from Apple Podcasts' charts and directory, and Podcast Index where it is used. Every show and episode belongs to its publisher and plays from the publisher's own feed; this site hosts none of them.