AI-trained domestic robots： from science fiction to reality？

Domestic helper bots are coming. Billions is being poured into this new AI frontier. A global gold rush underway. Go make us a cup of coffee. I'm on. But how close actually are we to the sci-fi dream of a robot butler? Holy! Never seen this in my whole life. Actually, let's start with another task and we'll come back to this. We'll come back to this. Good accuracy. Oh, spoke too soon. So this is it. But it's a normal house on a normal street. But that kind of makes sense, doesn't it? They're building a robot for homes. Hi, Natasha, how are you? I'm good, how are you? Very well, thanks. Can you describe what we're seeing here then? So you've got the robot off camera. There's someone wearing some of the similar kit. How's it all working? So we have a teleoperator who is wearing a suit to teleoperate the robot. Ideally, it should have its own mind. I mean, it sees a mess. It should be able to go and like clean it up because it thinks it's messy. And it should know what to do. Can it do that? Right now, it is getting trained to do that. Yeah. As an MIT robotics dropout, Bipasha spun out a student project quickly convincing investors to put money into her vision. We believe today people have two aspirations, usually like a car and a house. In the future, they'll have three aspirations, a car and a house and a robot. A few miles away, also in Palo Alto, another company has trained its bot to do some tasks completely on its own. Wow, it's a home. Tony Chow began working on his home robot while at uni but dropped out to build Memo full time. Hi there, I'm Memo, your future home robot here at Sunday. So can it see me around it? Is it? Right now, it doesn't react to you, but it can perceive the world with all the cameras and other sensors. Can you just go make us a cup of coffee? It's after 2 p.m. Would you like me to go ahead and make that cup of coffee for you? Yes, please go ahead. I'm on it. Making your coffee now. So Tony, can I just check with you? There's no one that we can't see doing this for the robot. This is all autonomous. Yes, it's one single neural network controlling the whole body movement. It was all going smoothly. Then Tony paused the demo as he saw something was off. Maybe we need to restart this and restart this. I think there is a... Can you stop it? I think we might need to do it again. I saw there's some peeling off here. Sorry about that. Let's just try again. Maybe I'm overly cautious. Oh, holy. I've never seen this in my whole life. Okay. After a reset, Memo smoothly cleared away the table. There's obviously work to be done to get it ready for homes, but there's no doubt this is one of the most capable bots in the world. How have you got the robot to be this advanced? How have you done it? Normally, the way we train AI is to teleoperate the robot in people's homes or in other places and gather the data and train the AI. And what we're doing is different, as in we don't need robot data to train the robot. And instead, we build these gloves. And people just wear the gloves in their homes and collect data for us. And that gives us really diverse data because we now see like more than 500 homes. And also all the different ways they go about doing the task. Sundae's army of robot teachers are all being paid to do repetitive everyday tasks, a reminder of the human drudgery underpinning how these machines learn. The staff allow cameras and sensors to record their every move at home. But Sundae plans to start shipping Memos next year. So if the bot needs to be remotely controlled, it will raise privacy considerations for customers, as well as questions about safety in a domestic setting. Like there will be kids running around. Kids will be climbing on the robot. So we really want to make sure that we put safety as a most fundamental consideration when we design these robots. Some robotics companies are already out in the wild, gathering training data in public. Here it is. Isaac's been trained by watching hundreds of hours of laundry folding. For the last couple of months, though, it's been doing it on its own for real, here and in six other sites in San Francisco. Deployment is the strategy. Nothing counts if it's sequestered away in a lab. This is the very task that we'll also be doing in homes. And here we get to prove that it works in the real world, bring down the time per article of clothing that we mold. Here's a T-shirt. We'll do a live demonstration. There we are. OK, so this is a lot slower than a human. So some people might be watching this saying, oh, that's not very impressive. But can you tell us why it is impressive? It is impressive because this thing can run all day. You often hear that robots are best suited to tasks that are dull, dirty or dangerous. We want to make clothing folding optional for people. Oh, here we go. Hang on a second. Struggling with a little bit at the end there. You know, we kind of care about getting a presentation right. That is way better than I would do it. When you first put this robot here, how long did it take to do one T-shirt? I think that we were looking at 2-2-30. So already, what, in how many months? That's in a month and a half. In a month and a half. You've got down to 1-30, just over 1-30. We've planned to start shipping Isaacs to homes this year. They say it'll be able to tidy up and fold laundry, but it's not clear how much of that will be autonomous. I actually didn't know how to do it properly until I saw this. Across town, at Physical Intelligence, they're not interested in building robots themselves. They're focused on creating the AI software that can be used to make any robot capable of doing chores autonomously. The rest of AI, like in CHAT-TBT, for example, we're used to training these models on an internet of data, but we don't have that web of data for robotics. We want to be able to actually kind of essentially breathe intelligence into any sort of physical embodiment, whether that's kind of a standard typical robot, a humanoid robot, or even something that looks closer to an appliance, for example. Like all these companies, Chelsea and her team are developing their AI software using reams of videos of humans doing chores. Their approach is proving successful. Investors like AI giant OpenAI are on board. Another tech giant, Tesla, is deep in development of its own humanoid robot, Optimus, but it's not yet clear what it can do. And it's not just in the US where people are getting excited about human-like robots. Chinese company Unitree is already dominating the market with its G1 robot, seen here in a demo given to us by its UK seller, Scan. This is happening right now. People are buying them, people are coding on them, and people are learning. It's entirely operated. It's not autonomous at all, as of yet. It's hard to get into China to film, but the humanoid robot industry is so hot there that the government recently warned a bubble might be building set to burst. What's really interesting driving around and speaking to all these robotics companies in Silicon Valley is there's a lot of confidence. They think they are building the future and they're certain it's going to be a huge moneymaker. But there's also some nervousness as well because there's lots of things that we wanted to film and we wanted to show, but we can't. They are scared because, of course, this is a race, a race to bring to market a domestic robot. A company that hopes to be the first is Onex. The Norwegian-founded company moved here last year and millions of dollars are riding on its bot, Neo, becoming a hit. We've just walked up to see Neo in action and someone is operating Neo right now and watering all the plants. I've just been told this is what they do all day. So there are people in the office operating Neo with a VR headset, carrying out tasks and testing all day long. Good accuracy. Oh, spoke too soon. So interestingly, Neo is struggling with this particular handle but the team here change the handles all the time because they're trying to train the robot different ways to open different handles. They had it there. Founder Bernt Bornich is pushing hard to get Neo out to homes. He says thousands of people have already pre-ordered. So what can Neo do right now for me in my home? Yeah, so let me tell you what Neo does in my home. So does a pretty good job of like tidying. So like resetting my living room and my house in general. Wiping surfaces, like cleaning the counter, like these kind of things. And that's all someone with a VR headset? It's a mix. So in my home, we have a lot of data. So a lot of the stuff in my home can get automated because we have data. Periodically, someone kind of steps in and helps if the robot does not know exactly how to move on. Early adopters will have to be comfortable then waiving their privacy. But Bernt promises that Neo will be able to carry out many tasks autonomously by the time they start shipping it. Your first set of customers then, they have to be quite wealthy and they have to be willing for Neo to make mistakes, essentially. Well, I don't think they have to be quite wealthy. It's thousands of pounds though, isn't it? I don't know how much it is. Well, it depends. 20,000 pounds? Yeah, like are most people who have a car. So you liken it to a car? For a very large subset of that, it will be way higher value than having a second car. Getting the right customers is more important in the beginning. Distribution with the right customers, so we can really use these amazing people that are early adopters to help us make this work. Outside of the tech bubble, some, including the International Federation of Robotics, think it could take 20 years before domestic bots become truly useful and accepted. But we have heard that before about other futuristic AI technologies. In some ways, brand new revolutionary technology does just sort of creep up on you. If you think about driverless cars, I've been talking about driverless cars in my reporting for a decade at least, but now they're here and they're just sort of normal. In many cities across the US and in China, driverless cars are an everyday sight. AI robotics companies are convinced their tech will go the same way, and not only will their devices be truly useful, we'll all eventually want one in our homes.