The Art Behind Human-Robot Interaction

Thanks for joining us today. I'm Shirley Goh, and these are my colleagues, Mario Bellini and Leland Hepler. Today, we're talking about ATLAS and human robot interaction. So as we're discussing, if you have questions, feel free to throw them in the q and a, and we might answer them. So a little bit about us first. I'm Shirley Goh. I work in product marketing at Boston Dynamics. Previously, I worked at MIT and Boston College in their research spaces. And prior to that, I was a staff writer and editor at The Boston Globe. Leland, can you tell us a little about what you do and your background? Yeah. I'm Leland Hepler. I'm a product designer at Boston Dynamics. I've been at Boston Dynamics for about eight years and been working in robotics for about ten years. Before that, I developed video games for about ten years, and I was an animator at Disney Studios for about fourteen years before that. Okay. Mario? I'm Mario Bellini. I'm the director of human robot interaction at Boston Dynamics. I studied robotics at MIT, was an entrepreneur for six years, and then joined Boston Dynamics eight years ago. I led the Atlas product team, through the journey from R and D through commercialization and looking at ways that we could build applications out of our humanoid, culminating in the launch at CES this year in January. Why don't we start off by saying what HRI human robot interaction is and how that manifests at Boston Dynamics? Yeah. Human robot interaction is is any experience our customers have with our product, whether that's in person with a physical robot relating to this machine through its lights, its sounds, how it moves, how it feels around you, through our operator interfaces, you know, if it's on a tablet or handheld controller, all the way through the software interface that you use like Orbit to tell robots to do tasks on your site or understand the data that you're collecting. So it's this full spectrum of interaction design, physical, digital, remote, in person. And how has that changed in the last few years? We're at this exciting inflection point where everything that's made, you know, products like ChatGPT or Anthropic or any of these LLMs that you're used to, anything that's made those really engaging, the ability to have Internet scale data condensed and accessible, the the fact that they talk and seem, you know, very realistic. All of that can directly translate into robots that aren't just following set scripts anymore. They can be reactive to what you say, reactive to how you're interacting with them, and be much more natural. And so we're kind of seeing this explosion of new robot and interaction capabilities riding the LLM train. And for those who don't know, can you tell us what LLM is? Yeah. LLMs are large language models. And so they're they're products like ChatGPT or Claude that take Internet scale data, basically, all the words that you can find on the Internet and create a big predictive model of what word to say next. And through that predicting next word and a bunch of, you know, fancy tricks on top of that, you can have these chat bots that chat just like people do and feel remarkably human. And so what we've been exploring is how we can take that capability and make our robots interactive with voice, but also give our robots the same kind of reasoning tools that these agents have explored, so they can interact with the world in a less prescripted, more organic way. Okay. And Leland, can you tell us about, Atlas and HRI and a bit about, your work? Like I say, I have a background in animation, and also, I've done a lot of work in video games doing simulated characters and using old school AI techniques for creating interactive experiences with characters in in in video games. And for me, HRI, and the the part of HRI that I'm really interested in is, you know, you're you're you're trying to externalize the inner interstate of that robot. The interesting part of Atlas is that it has this humanoid form factor. And so, there are these interesting questions that are being sort of figured out in real time of what is the line between where we, you know, treat a robot like Atlas more like a machine, and where do we tap into the vocabulary of human behavior to help people understand what it's gonna do next, to feel safer, to feel like it's more predictable. And like Mario was saying, there's a software aspect, but then there's also these these sort of classic user interface components on the robot, like lights and sound that, again, there's a whole pattern. There's a whole language of of light and sound patterns that are established in machines and in in graphical interfaces that have certain meaning. But there's also behavioral aspects that people are very familiar with how humans move around, what our patterns mean. And so, I think the interesting challenge with Atlas is figuring out what that nice line is of treating it like a user interface that gives information at this more contrived user interface level versus tapping into, you know, the vocabulary of movement to help people kind of understand what it's what it's gonna do next. A couple of the interesting areas in on the behavioral side are anticipatory cues. These are classic things that animators do, or, you know, if you're if you're animation in video games, you exploit all these same techniques. But usually there are, you know, cues that we do before we do a big move. We do a bit of a move back, and then we do a big move into that. Right? So another one of the interesting cues I think that we can explore with Atlas is, you know, when we're when we're thinking or when we're processing information, there are all kinds of standard things that, again, animators tap into frequently of how how do we behave when we're processing information. And, you know, some of the cliche ones are, you know, touching our face and looking up, you know, tapping our face in some way while we're thinking, and having someone understand that the robot is in a processing phase. These are some of those things that we're familiar with at a deep level that can be sort of tapped into at that behavior level to communicate the same idea, but in robotics terms. We finally have the tools to make robots like NPCs and video games, and to have them be expressive in real time with, you know, very natural feeling movement, as well as the way that they talk and all of the different state, you know, like lights and sounds and things like that. And so we're you know, Leland's got the perfect skill set for this, but we really are gonna be at this taking animation and video game tricks and bringing them into physical space with robotics to make these, you know, fully embodied characters that, you know, might be doing rote industrial tasks today, but you can imagine, doing tasks and experiences closer to people in the future. What is the goal behind that? Is it to make people feel more at ease with having robots around them? Is it to communicate things? It's a little bit of everything. So, you know, if you're using a product, you wanna have a great user experience. You want it to be, like, easy to use. You wanna like using it. You wanna feel moments of delight. You want it to do the task really well. And so for some things that we're expecting Atlas to do, like sequencing parts into a machine, someone that's near the robot just wants to make sure, you know, is it on track? Is there an issue that I need to help with? If it's done with this task, how do I have it do the next task in the queue? And so it's probably pretty like meat and potatoes, straightforward interactions. We want like a blinking light when there's a problem. We want Atlas to, you know, have an error message that's easy to understand if something goes wrong. We want it to ask for more work if it, you know, runs out of things to work on. But if you're having Atlas in a different application like, you know, closer to people, whether it's, you know, let's say a tour guide, even though that's not something we're we're doing with Atlas right now, you probably do wanna have the conversation and the character of the robot put someone at ease. You want them to not feel afraid when they're close to the robot or trust that the robot, like, sees them, understands what they're doing, and is gonna be acting safely. I think a lot of this comes to like building trust with this, you know, larger industrial machine and anticipating what it's gonna do next, having it follow through on that anticipation, being able to not have this kind of scary black box, have some kind of understanding of the robot's internal thinking. I think essentially, it's making the robot predictable. Right? And everything kind of flows from that. Your safety, your feeling at ease. When it's predictable, you know when you wanna stay back a little bit. You know so the robot can do something that maybe you would be uncomfortable around being that close to the robot with the danger involved or whatever. But if if you know where it's gonna move and you kind of have a mental model of what it's gonna do, it puts you at ease. But I I one of the interesting things I think with all of this is I think there's an interesting tension between keeping people safe around robots, and also making them comfortable at the same time. But it's almost like, come closer, but not too close. Like, what is that sweet spot? And we have we have a long history with with the Spot robot, our dog robot, is it's a very safe robot. No one's we've never really had any incidents of of anyone having any real problems being around our robot. But we recommend people stay two feet away and or two meters away, rather. And when people are around robots, lot of times they're like, oh, you know, especially like a few years ago when it was when it was more more novel. People were really certain people were really made uncomfortable, didn't like being around it, had a sort of a negative reaction. And then as people warm up, they become really comfortable and it's it's like in our office, these robots walk around all the time and nobody gives it a second thought. But the interesting thing is, you know, sometimes people will do experiments with the Spot robot that will kind of destabilize the you know, in an experimental context, destabilize the behavior, and it will start to be a little bit less predictable. And I've even found myself in those situations in our lab kind of starting to feel that discomfort again. Right? Because all of sudden, I don't know what it's going to do in a reliable I've seen this happen a thousand times. And so all of a sudden, there's like a deep thing in my brain where I'm like, even though that, you know, it would be really hard for Spot to hurt me in any way just because it's it's of its size and weight, I become uncomfortable because I don't have a mental model anymore for what's going on under the hood. And that's a really interesting aspect of HRI is how do you how do you make people not be uncomfortable about the around the robot? So for now at least, Atlas is designed as an industrial robot and not, for example, a home companion. In your work, how are you working to signify that to the people around it? Yeah. I I guess we'll start with with Atlas is six foot four inches on on a good day. It's a a large hulking machine. And so it it has a lot of physical presence. You know, if you look at the set of all humanoids in the world, Atlas is probably one of the largest and and heaviest. And and that was a deliberate decision. You know, when we were looking at opportunities for humanoids in in the world bringing Atlas to market, we saw that there were these really repetitive tasks that were really heavy, backbreaking work. And we're owned by Hyundai, and there's a number of these tasks in the automotive supply chain. And we realized that if we focused our humanoid on those industrial tasks rather than trying to do a humanoid that did every task from industrial to home to, you know, close companionship. Atlas is a very robot y robot. And and when I say that, I mean, it's it's a robot that isn't trying to look and feel like a person. It is a machine, and it is designed as a very efficient machine for handling, you know, large heavy objects in factory environments. And so it's got like big chunky joints. It's six feet tall. It is has this, you know, round non anthropomorphic head. We weren't trying to design a robot that looked and felt human. We weren't trying to design a robot that you could, you know, spend an evening with in your home. We were trying to design a robot that was this industrial tool that happened to have two legs, two arms, and a head. And by doing that, we have this design flexibility to not try to make it the most human like robot as possible to celebrate the mechanical aspects, whether that's through really purposeful movement that takes advantage of the full range of motion of the robot or abstract lighting patterns on the head, or industrial design cues that really are honest to the mechanical nature of the machine. Atlas is a great platform for us to experiment with, to try to build these conversational interactions, to learn how people want to relate to machines. But when it comes to the product, it's all grounded in the fact that this is gonna be, you know, a machine loading appliance or a part sequencing appliance, or this machine that industrials there's these industrial expectations tied to. Was the inclusion of a head on electrical Atlas part of that intent to kind of convey different things? Because the hydraulic Atlas, you know, didn't have a head. What was kind of behind that decision? The head was a great opportunity. So there was a huge internal debate of, you know, should we have a head when we go to the electric Atlas? And, you know, the head is just another thing that can break. It's, got two degrees of freedom as it looks around. So those are potential failure modes. And if we're just purely optimizing for for cost and reliability, we probably wouldn't have one. And so HD Atlas didn't have one. It just did gymnastics. It, you know, fell on its, you know, its bulb bulb all the time and didn't break. But we knew that we wanted to start to get into this world where the robot would be working among people, maybe not, you know, within two meters of them, but among them. And as people were used to looking at other people, there's a lot of cues that we get from the head and the face. If you're looking at something, we can anticipate that you're probably gonna grab it. If you're if you're looking at something, we can anticipate you're giving it attention. We're used to reading facial expressions. And so that was an opportunity for us to explore with Atlas. And our leadership team, you know, really buys into this, you know, ubiquitous robotics future. And we wanted the opportunity and the platform to start learning that today. And so ultimately, Atlas has a head, you know, in line with the it being a robot y robot. It's not a Tron face. It's not, you know, this like soulless void. It's a more abstract form that we're having a lot of fun from the design perspective turning that into, you know, expression vocabulary that the robot can use. I've never thought of Hydraulic Atlas as not having a head. I've always thought of Hydraulic Atlas as not having a neck. Right? Because like, when I look at it, I see like a dude that's like looking around like this. Right? And the interesting thing that I think psychologically, if you have a form factor that looks relatively human, we kinda look here for something that fits our expectation. And I sometimes find it to be surprising when people talk about the hydraulic as being headless, because I've never thought of it that way. Necklace, yes. Now, if I can additionally get the head to look at the right place at the right time, so in other words, if Mario and I are talking to the robot at the same time, and I talk to the when I address the robot, it looks at me, maybe it glances over and then looks back at me, then Mario speaks and it focuses on on Mario. When you see that happening with robots and when you see that happening with animatronics, what when you see that happening in an intelligent way, all of a sudden your brain starts to kind of believe in this thing in front of you. Right? So I think if you if you get good conversational back and forth, and you get good head look working, a lot of these other things like gesturing and and these sort of subgroups of of like behavior, if you get those working on top of that core thing at the right time, you start to flesh out a thing that feels pretty comprehensive. But I think that the the voice and the head look are probably like eighty percent of the effect. And if if you get your brain to buy into that, the rest of it is sort of the icing on the cake. And so the interesting thing with the the emotional aspect is I think, you know, when when if I look around at myself and other people in a meeting, interacting at work, right, there's very little, apart from conversation, intelligent head look, there's very little emoting going on in our faces. Our faces are pretty neutral. Right? Maybe there's a, hey, but apart from that, we're not going through the full range of of emotional expression. Most people only use a tiny little bit of that. And so, when you're talking about the facial cues, I think a lot of those things that we associate with like emotion that you can communicate with a face, I think most of that, if you get a sense of of the character and its emotional, the robot and its emotional state, that's gonna come more from the voice and what you're talking about. And then your brain is this is like a this is this is like old animation tricks or game tricks. Like, realize, oh, I can only control these externalized physical things. But most of the illusion is somebody's brain filling in the rest. And your brain fills in the rest when those big cues like the voice and things like head look are correct. And then all of a sudden, your brain starts to be like, oh, okay. I'm I'm believing in this. So then you start to create these narratives of what's going on in that character's head, their emotional state. And the face can be icing on the cake, but like, if we can make if we can get head look working well, and then maybe we use the lights to keep the face alive in sort of a blink type of a, hey, I'm serving the same purpose as your blink does, and that I know that you're still operational, like, that's kind of what it serves in a a in a robotic sense. I think the the weight on the head to carry, like, big emotional cues is not really necessary. I think an interesting experiment with Atlas is how much of, you know, if if if Atlas is thinking about something and processing something. Well, there's a UI convention of just, you know, dot dot dot dot, where, you know, a visual dot, like, just goes around load the loading screen convention. Right? That's a thing that everybody's familiar with that combined with something like maybe, you know, Atlas looks down, maybe Atlas touches its face. I I don't know whether that's necessary, but it's an interesting space to play with where like playing with behavioral and UI conventional cues kind of tuning up. But again, that's much more externalizing the idea that, hey, I'm pausing, not because I'm broken, but because I'm processing something for you. And that's the important part of the information. So when we talk about people working around the robot, how are you planning for that? How close or far away will they be? You know, how will they stay safe? How will they know when it's okay to approach? Can you tell us more about that? Yeah. So when we were, you know, bringing Atlas to market and thinking about, you know, what it would take to turn this into a successful product, flexibility is the number one, you know, reason why a customer would choose a humanoid over a traditional automation or or other types of platforms. And you can't have flexibility if you have to bolt in cages into the floor and relay out your workspace so that there's room for the robot to be completely isolated from people. And so we knew from day one that the robot had to work beside people, whether it's on a an assembly line or loading machines next to them. And we wanted to make sure that we were doing that in the right way. You know, we have two thousand spot robots in the world. And as you start to scale robots with customers, you start to talk to their safety teams and and what they care about. And so we knew that we needed a safe system, a safe system that safety teams on-site who have liability for the devices that they bring into their environment would be happy with, a system that would be certifiable, that you could follow international best practices for autonomous machines. And so we started very early on in building a team that could build a certifiably safe person detection system and set of behaviors on top of that if a person gets too close to the robot. And the initial boundary we're drawing around the robot is two meters. So we're expecting that ATLAS, you know, isn't going to need any fixed safety infrastructure and it's gonna work autonomously within or up to two meters of people nearby. If someone gets closer than that, it's going to turn and and look at them and and blink its light in a way that's like, hey, I see you. It's probably going to pause itself. And if they get closer to within a meter, the robot's going to start to, you know, render itself safe. So maybe like slowly lower itself down or put down the object it's holding or just stand stationary depending on, you know, how close the person is and and what the current behavior of the robot is trying to do is. Over time, we expect to kind of loosen that that boundary. And so we're purposely conservative. You know, right now, we wanna make sure that we're doing right by our customers and our team is focusing on how we can add as much value to our customer sites with Atlas. As we move into other applications, we're asked to work closer and closer. We've got all the the hardware infrastructure in place that we could start adding more sophisticated algorithms on top. So Atlas can eventually, you know, work very close to people. How do you plan for people to use the robot? Has that been determined yet? Will it be through Orbit? Will users have like a tablet or console like they do with our other robots? Kind of all of the above. And so we expect it to be tightly integrated with Orbit from understanding the world and integration in the customer systems perspective. But day to day interaction with the robot on-site could be anything from using a handheld controller to, you know, drive it to a new location or conversationally saying like, hey, let's go to work cell twelve and the robot walking over that way. A lot of our customers are in really loud environments. And so we're we're figuring out the right way to navigate voice in those settings. But, you know, it could be as, you know, semantic and conversational as that. Teaching Atlas new skills is probably gonna be either VR based or through some kind of egocentric data. So you either do the skill yourself a number of times to teach Atlas how to pick up a certain object or do a certain part manipulation that you want, or you wear a VR headset and control Atlas to do that. Over time, as we start building these larger world models that the robot uses, these larger VLAs, we expect customers will do less and less of that. And most of the interaction will, in a couple of years, you know, be very either conversational if you're near the robot, or entirely in orbit and some digital twin of the world. What other considerations did you have when you were designing the robot? The humanoid robot is the most complicated product you could, like, ever bring to market. Everything they teach you in in product and engineering school is to, you know, cut scope as much as possible, make it as as simple as you can, only focus on a single thing. And a humanoid is is only economically feasible if it can kind of do everything or like a whole lot of things within the market it's it's targeted at. So in bringing that into, you know, product world, it was thinking about how do we make it cost effective, how do we make it reliable, how do we make it easy for users to retask. And there's a lot of human robot interaction and and product design work in that retask bit. I think that that's where most of our attention is gonna be in the next couple of years now that we have, you know, the new production grade platform walking around and working. But within that platform are all of these decisions around part serviceability, minimizing individual parts, leaning on the Hyundai supply chain, making something that they can manufacture at scale. Yeah. On the aesthetic side of that, making a humanoid robot while it can do some work for you, because you can leverage familiar patterns of movement to make people more feel more comfortable, also can work in the opposite direction. Right? So there's a there's a concept in robotics, games, animation for years. It's kind of a little bit cliche at this point, but like the whole uncanny valley of movement, you know, which is basically like, I'm looking at something that my brain kind of thinks it understands, and then all of a sudden it just did something that seemed really unnatural, and I'm getting a weird sort of lizard brain ick or something from it. And one of the one of the I think, from a movement standpoint, one of the coolest things about Atlas is the fact that it looks like humanoid form, but it's not constrained by the typical musculature and bone structure of a human. Right? So whatever's the most efficient way to get from point A to point B is on the table. It can spin its torso, it can jump up in the air and the legs come down oriented in the other position, it can just start walking and flip its legs to the other side. But those are all aspects of the design that, at least initially, when people see them, some people have an uncanny valley reaction. And I remember in the comments, for the first for the first eAtlas video where it was on the ground with its legs over its head and it stood up from being flat on the ground in that really unnatural way. You saw a lot of that type of reaction in the comments. Right? I remember, like, when in the comments we were discussing at work the comments on that first eAtlas video where people were kind of freaking out. And I was in a meeting where I went back and I got the comments that people had put in the in the comments for the YouTube video, where the first time that Spot opened a door with its with its arm. Right? People had never seen that before. And people were losing their minds and they were having this very, like, oh, this is horrible to look at. It's like so creepy, blah blah blah blah blah. And now, it's like, you know, Spot with the arm is in children's museums, and it's pretty ubiquitous, and most people have relatively become acclimated to that. And but I brought those comments because exactly, I read them back, and then I was like, oh, these are from Spot six years ago. So there's a certain aspect of this that's not in our control, and robots just need to be there out in the world, and people get used to seeing these new patterns of movements. But that is one of the things that come along comes along with, you know, Atlas very specifically with humanoid mixed with these extra human sort of movement capabilities. One of the the weirder weirder transitions from moving from, you know, product world of how do we cut scope and think about this as as a purely viable commercial enterprise to design world where we're thinking about how people interact and experience the robot is is now we we spend a lot of time thinking like, what's the vibe of Atlas? You know, when someone looks at Atlas, how should they infer its characteristics from how it moves and it acts? And and one of the word that words that kept coming up when we were running these workshops with the, you know, design and leadership team was purposeful. You know, what Atlas does should appear intentional. And related to that is how it moves. Because it has all these extra degrees of freedom, it shouldn't just try to mimic how a person would move in that situation. It should purposely move that part as efficiently as possible, even if that means like spinning its torso around or flipping its knees backwards or doing, you know, whatever other contortion. Do you think that kind of unease applies when it's a robot with a human form or with an animal form like Spot? I feel like, you know, for our other robot Stretch, you don't get so many of those comments because looks more like a traditional machine. Yeah. Very much so. I think, you know, few people see something that looks like stretch in nature, so you don't have this, like, anticipation of what it's gonna do. And, you know, your lizard brain doesn't, you know, go off. I think Atlas being a very robot y robot gives us, you know, some freedom. And once you get over that, you know, initial YouTube video, oh my god, what's happening moment, you know, it starts to like look normal and it all kind of like tells the same like purposeful movement story. If Atlas was, you know, five foot ten and much more human proportioned and shaped, I think there'd be a different level of ick to that. And if it moved in the same way, we we probably would have a a, you know, more long lasting response. So earlier on, you had referenced moments of delight. And one of my favorite things on the Stretch robot is something that someone on your team, Christa Shapton, created. And it's this animation for when the last box is picked. There's this Pac Man like animation of Stretch going through the truck and kind of gobbling all the boxes. What kind of moments of delight do you two have planned for Atlas? There's nothing quite when it looks you in the eyes and says, I love you. But, yeah, for Atlas, moments of delight that we're interested in are, you know, anything from specific quirks of how it moves. You could even say, you know, once you get past the uncanny valley, the way that it stands up is kind of a moment of delight. There's a lot of little subtle things we can do with the lights, you know, make it look almost like it's opening its eyes when it's waking up. Really subtle things like that that don't get in the way of the, like, industrial utility, but add a little bit of like a a wink to customers. There's a lot more things that we can do in the user interface in Orbit. And it's great to have people like Christa on the team who can, you know, both make really easy to use interfaces, but then inject those moments that make people really love using our products. The opportunities that I'm most excited about are the ones that are kind of more on the subtle side. We saw this a couple of weeks ago. Mario's working on a little museum guide version of the Spot robot. And so we were testing on the Spot robot some of these things that we've been talking about, like the head look, the conversational AI. Hey, I'm over here. Come look at me. And But there were a few moments where we were standing there and the robot looked at me, and then looked at Mario. And it was like my brain believed that that thing had intelligence just for a moment because and again, obviously, I know it doesn't, but my brain's, like, lizard brain kinda thing went off and it sort of it kind of broke through. Even though we've been like doing this for like a couple of hours and get it and testing things, there were these few moments where those little things lined up, and I believed for a moment. And I think that's the most the thing that I'm most excited about is those are these more subtle forms of the delight, but it's when it, like, you create an experience that your brain believes, and it serves some sort of a purpose for you. I think there's a little bit of that magic there, where my brain knows that that's a machine, but like for a second, I believe I believe that in it as a sentient being. And I think that's totally on the table for Atlas because of its form factor, and those are the really interesting opportunities in the years to come as these things become more ubiquitous and kinda just out there. Well, we're gonna take a break for just one minute. Don't go away. When we get back, we're going to answer some of the questions that you've been asking us. All right. Now we're going to go through some of the questions that our viewers have been asking. So this is one of the top ones that we get all the time, which is when can I get an Atlas? Reach out. So we have a number of customers that we're starting to work with to understand what they want to use Atlas for, what the capabilities of Atlas on their site will be. We're ramping up production in our facility in Waltham. We have a lot of production partners through Hyundai that are gonna be starting to ramp up their production. We're gonna be doing our first tests outside of our Hyundai family next year, and then moving from proof of concept into pilot with those customers in the next couple of years. So in the two ish year time frame, you can start to have much realer interactions with the robot itself. But you can't really go from zero to humanoid. And so we'd love if you reached out, talked to our team, and started to learn what Atlas can do on your site, help our team understand what, you know, features of Atlas or of Orbit are most important for you. And Spot is a great onwrap into Atlas through our Orbit platform as well. We have another question. You were referencing talking to robots before. Will Atlas talk to us one day? If I could wave a magic wand, Atlas would only sound like Wall E and make a bunch of cute, indecipherable beeps. And that's kind of authentic to it. However, people are gonna be talking to robots and probably expect some kind of, you know, actual language back. And so we are gonna have versions of Atlas that talk and it might be limited by application or environment. If you're in a really loud factory and you can't hear it respond, that that's not as valuable. But we are absolutely moving into a world where people will be talking to robots and they'll be talking back. Yeah. It's all it's certainly on the table now. Whether it's in variant, like Mario was saying, one of the big challenges with factory environments is noise. And if you needed to get super super close to the robot for it to be able to hear you, that might not be a desirable thing from a safety standpoint. But being able to have cool conversations, I mean, one of the one of the interesting things is how far can you take that, just apart from a conversation, how far far can you take that to have the robot, you know, tell you about its state, tell you about what it wants to do next, have a kind of an interesting back and forth with it. All that stuff is totally possible with the technology that is exists today. So it's only getting better. We have another question about what happens to Atlas if its WiFi connection suddenly drops? Yeah. It will probably have some kind of alert that its Wi Fi connections dropped, but you won't really notice anything dramatic on the robot. So one of our key philosophies is that everything the robot needs to be safe and to be in control of its physical form should happen on board the robot. And so if the Wi Fi connection drops, it probably won't be able to do that next manipulation task you had lined up. But it would be able to safely walk back to its dock and use the dock's wired connection to tell you that, you know, something's up with the Wi Fi. We had a question also about what happens if you need to just shut down the robot if it's doing something unexpected. I think we've all seen videos of robots out in the wild, and they start behaving in an unpredictable way. I remember there was one where it wouldn't stop dancing. I don't expect Atlas to suddenly bust a move and be unable to stop. But if you needed to stop it, what's the plan there? Yeah. Well, Atlas doesn't have time to dance. And it will have an e stop button on its back. If you're in close proximity and you need to stop it, there'll be stop buttons on the controller or an orbit that you could use to stop the robot remotely. Purely in design prototyping HRI world, we're looking at different stop commands that you can give verbally or gestures like, you know, holding your hand up or crossing your arms. What's most important is that if if this is a safety function that you're relying on to stop the robot, that we could actually certify its performance. And so, you know, it might be cute if you see all like, you know, stop and the robot stops. But we wanna make sure that in a loud industrial environment, if you yell stop, the robot will definitely always stop. And that's a really hard guarantee to make. And so we're doing experimentation to see what those feel like in terms of user experience. But when it comes to actual customers, we want to make sure that, you know, anything safety related is actually certifiably safe. We have another question kind of about, like the hive mind of robots. When Atlas is on-site, does it share its learnings across all Atlas robots? Is that confined to just one customer? Yeah. So all Atlas robots at a specific customer, will be sharing a Hivemind together. And so the good news is that means that when one Atlas improves its manipulation capability or, you know, learns that it needs to adapt its behavior in a certain way, that you get that for free across all of your robots. We have different data collection programs and different partnerships that we have in place right now that your our team would love to talk to different customers about for whether data collection goes into the core model that we're using to control the robot. In which case, you know, all all of the training data right now is gonna manifest in every Atlas regardless of which customer. Or if it's gonna be post training data that's gonna be gated to a specific customer. And so it really depends on, you know, when you start to engage with Atlas and what, you know, terms you have with your your data team. We've learned that data is kind of the life blood blood of modern industry and our spot customers care a lot about their data. And so we're designing Atlas's training program and Atlas's data collection program respect that on the customer end. So on that data thread, someone else also asked whether the robot would be surveilling the people around it. And I know that people are, you know, concerned sometimes about automation these days and how much it's watching them and the ways that it will be used. Can you tell us more about that? Atlas will not be surveilling people with its cameras. We're gonna be collecting data to train manipulation tasks. And there's a light that's on the head that, goes on when that's happening. So it's obvious that the robot's collecting data. The rest of the time, it's just operating. We want people to trust robots, and they can't trust it if it's an ARC that's always observing them. We have a question about what the anticipated lifespan of the robot is. Atlas is designed to operate continuously for, you know, three plus years. Think more importantly, every motor on the robot's designed to be field serviceable. And so you end up with kind of this robot atheist situation where you might have done a field replacement on a motor or on a connector, but the whole robot's still up up and running. And this is something that, you know, we kind of learned the hard way in our early spot days how important it is for customers with large fleets to be able to maintain them on their own beyond the expected service life. Alright. And one more question. Can Atlas see people approaching from behind? Yes. So while Atlas has, you know, a head with a very clear directionality, it has sensors all around. So it's got cameras in the back of its head. You can see people approaching from behind. If someone were to approach it from behind, it would probably turn its head to look at them so they know that it sees them and then turn its head back. Well, thank you so much for joining us today. To all our viewers and Mario, Leland. Thank you. If you'd like more information about Atlas, please visit our website.