Hi, everyone. How are you doing? Thanks for coming today. Oh, there we go. Yep. All right. Yeah. Oh, Jesus. That's uncomfortable. And that's recorded. That's great to know. So, hi. Thank you for coming to this launch lecture. I've been given a list of things to read to everyone today, and I'll be introducing Dr. Walker in just a minute. And I am going to pretend like I know what all of these things mean. So just some general announcements. Welcome for being here today. Quick show of hands. How many here are for four-credit students? Single person? Yes? All right, great. And how many are here for other, just other students in general, not for credit? Excellent volunteers to show up for this and then lastly interested guests right there in the back awesome so for four credit students the one you signed in and everything all right you're golden just a reminder during this talk you turn off your phones toss trash away before you leave the next lunch lecture is on April 3rd. And I think, do I save this until later? Does anyone know if I talk about the 2024 seed grant winners? Either way. I can say that now. The 2024 seed grant winners are Yanni, I'm not even, Lu, because this is, I heard someone who knows how to pronounce it in the back. Huijin Oh, Moini Riley, and Richard Wong. So I think we can give them a round of applause for that. So today we have Dr. Bruce Walker, a professor of psychology and interactive computing. He runs the Interdisciplinary Sonification Lab that studies human-computer interaction, issues in non-traditional contexts like cockpits and vehicle displays. His research interests include a bunch of stuff, including sonification and auditory displays, driving, trust and automation, technology adoption, human AI robot teaming, which is what we're going to be talking about today, Navigation, wayfinding, creativity, VR, AR tools, and assistive technologies. And I only think that's the end of the list there. So Dr. Walker teaches sensation and perception, engineering psych, HCI, research methods, auditory interfaces, and assistive technologies. He's the past president of the International Community of Auditory Displays. He's also the director of what we're going to be talking about today, which is CHART, and the Center for Inclusive Climate Communications. He's on the steering committee of another center, which is CRANE, the Center for Research and Education in Navigation. In addition to academic research, leading to over 250 journal publications and proceedings, Dr. Walker has consulted with NASA, state and federal government, private companies, and the military. Dr. Walker is also an active entrepreneur, founding startups and working on projects related to COVID diagnosis, skin cancer detection, mental health monitoring, and gun safety. Gun safety and digital altercation. Yeah? All right. So, please, everyone get help, give a warm round of applause to Dr. Bruce Walker. I'm delighted to be here. I'm going to walk around, let me know, just wave your hand if you can't hear me. But that long, unnecessary introduction, I know most of you. So today I'm going to talk about this new center that we're launching called the Center for Human-AI Robot Teaming Chart. I'm going to give a bit of the history and the overview of it give you a flavor of some of the projects but then I'm going to dig into one of the projects that kind of exemplifies how we approach this and how we want to approach this topic going forward so this is a consortium and I'm going to read because this is a lot of text a consortium of researchers laboratories and affiliate organizations dedicated to transforming the synergy between humans, artificial intelligence, and robotics. This is based in the School of Psychology but this is an international organization already. We have members from 17 different universities and five countries so we're we're already quite representative of the space. We're just getting started but we're pretty excited by what we're doing. So what are the missions? What are the aims? What we're trying to do here is really understand and elevate, harness the elevated capabilities of advanced technologies to derive societal betterment, enriched education, and propel scientific discovery. Yeah, mumbo-jumbo. We want to make people, technology work together for the greater good, right? Right? Sounds kind of like everyone in the room, right? Okay. Embrace a wide spectrum of domains and backgrounds, fostering collaborations that transcend traditional boundaries and disciplines. Sounds kind of like iPad. Sounds kind of like iRIM. It totally is, right? But But our emphasis is on the technology and humans partnering when that technology has some kind of agency. So it's not just about people using something. It's not about human computer interaction. How do I make my device easier to use or more user friendly? How do I actually become a partner or team or collaborate? So that technology needs to have some kind of agency. That's what sets Chart a little bit apart from the rest of the HCI, user-centered design, iPad, iRIM family. We certainly integrate diverse perspectives from psychology, engineering, computing, robotics, and beyond, public policy, all of these things. We're really coming at this from an ecosystemic design perspective. So it's not just that you build a thing. That thing's got to work within the environment. It's got to work within, you know, the systems. I can make a great car, but if it doesn't fit the roads, then it's no good, right? So, and finally, lots of aspirations towards solutions and having real impact. So we don't pretend that we are very different from the rest of the HCI and IPAT world, but we do know that we need people, technologies with agency, that's the thing, and leaning really into teams science. There is, in fact, a science of teams, and it's generally the domain of IO psychologists, industrial and organizational psychologists, and so we really want to lean into that as much as we can. Here at Georgia Tech, of course, this is very much the iPad iRIM chart space and we're delighted that this talk and our efforts in the broader sense are supported both by iPad and by iRIM and we thank both of those organizations many of our members of chart are and will be also members of iPad and iRIM it's just a natural overlapping space all right so that's That's just sort of some of the framing of what we're trying to do. In terms of our org.chart, see what I did there? We have a steering committee with myself, Meng Yau Li, Professor Meng Yau Li, and Professor Chris Weiss, who introduced me. We're all in psychology. In addition, Sydney Scott-Cheroni is here. She's our Chart Fellow. I'll point out that the School of Psychology has generously given us some seed funding to get us off the ground, including funding for a GRA for some years so that we have that helping us. So it's really great. We'd love to see some more people here. So if you're interested in being part of CHART, if you're interested in being in the leadership of CHART, would love to talk. Absolutely. So we're just getting going. This is a launch. So there you have it. Okay. Our members are already from most of the schools, all of the colleges, if I'm not mistaken, even if they're not listed here. here, academic members from many universities that we know of, including European universities. We have industry members, and I'll say that these particular, at this stage, these members that we list as industry members are individual members who happen to work at those companies. so HRI and mass robotics and torque and Intel and Delta are not yet industrial affiliates I don't want to overstretch what we but we are delighted that the people who are part of our group bring knowledge from all of these areas to to what we're doing all right so that's just kind of all of the text heavy what we're up to and it shouldn't surprise anybody what we're doing right and you You know, we're engaging in, first of all, those of you who know me, I use this word all the time, friend-raising, right? Friend-raising is the first thing you do, and fundraising is the last thing you do. In between, we're doing research and design and education, tool-building policy, all of those good things, right? Fund-raising will naturally emerge from that. We're happy to have some seed funding. We are applying for other grants. Many of us are already collaborating on separate grants, not necessarily to run the center, but for our research. So the friend-raising is the first and most important aspect of this. All right. Well, we have all of these members. We have lots and lots of projects that people are already doing and have been doing. Sometimes it's because it's what they do, and that's why they're in chart. Sometimes there are projects that have emerged from collaborations based on connections that were made in the context of chart, right? This is just a very brief overview of some examples. Misuse, disuse of automation. We have other people studying trust in automation, perceived risk, bio-team dynamics, adoption of flying cars, anthropomorphism and voice agents, right? How, you know, how lifelike do you want your agent to be and what does that do to your trust? The Hercules mapping and wayfinding system, this I-Fection project from Germany is all about AI and affect and emotion. Graph ingestion engine is an interesting way for people and AI systems to collaborate in parsing graphs. And robotic guide dog is one example I'll talk more about. This is, again, just a very brief sample of some of the many projects that are going on from our researchers. And I'm happy to share more information about these projects or other projects with anyone who's interested. We have slides for all of these as well, but today I'm going to focus on the robotic guide dog Because it really is one of those projects that is kind of at the heart of chart Surprise myself sometimes All right Okay, so robotic guide dog the people involved in this as you as you see many of you know professor Sehun Ha or Greg Turk or Clint Ziegler, many of you know Terry Kim, who's one of our many outstanding graduate students, and many, many other MSCS, MSHCI, undergrads, lots of people have been working on this project already, and as you'll see, it has multiple facets. So we want to, in general, enhance the quality of life of individuals who are blind or visually impaired by creating a robot guide dog? What's wrong with real dogs? Right? Well, just like anything, we're not trying to replace. We're trying to provide alternatives because there are certain circumstances when it makes sense to have a robot. So why do we have a robot? But this particular form factor is actually very useful. Let me talk about some of the challenges of dealing with an actual dog, a canine, of this form factor. It can be a year and a half wait time. It takes a lot of time and money to raise and train the dogs, you know, a couple of years. And then, of course, all of the maintenance that's required. And then when the dog retires, because they retire after a certain number of years, then anything that they have learned about you and that you have learned about them has to be relearned with a new guide dog if you get one, right? Guide dogs go with you. They go in the car. They go on the plane. All those kinds of things. But there are certainly lots of challenges associated with them. All right. Now, when looking at this question of robotic guide dogs, many, many issues have to be understood. There's some what we might consider microscopic things related to the robot. How does the robot stand and move? And some of that is basic robotics, and how do you get a four-legged thing to not fall over yes but there's also subtleties about how that robot walks and moves as it relates to the humans as it relates to the environment as it relates to the other people and we'll talk about some of that things about navigating bi-directional communications how do you communicate with this with this robot the task of co-navigation is a collaborative effort. When you think about an actual guide dog, the guide dog is instructed and is sort of supervised by the human, by the handler, but that dog has some autonomy, right? And they have some responsibility. There's this learned trade-off or sharing of responsibility between the handler and the animal. How do we make that happen in in the case of a robot? There's issues of explainability, right? So if you're what if you have a real dog and it's leading you or you're walking down the sidewalk and then it sees a squirrel, right? Most of them are trained well enough that they're not gonna go after the squirrel but they're gonna look so you might feel and sense that that dog is sort of distracted but that's understandable it's comprehensible right how do we understand the peculiarities and the quirks of a robot the whole notion of explainability and how does it prioritize? How does it decide what to do and what not to do? Lots of issues related to the design and the features and the ecosystemic fit. So I'll go through a few of these and hopefully make it clear that this is all about people, technology, and teamwork. And that interplay in the ecosystem for all of the stakeholders is a great example of how chart projects can move forward. All right, so here's an example of an overview, a system guide. There's lots of parts, and I'm not going to go through this and parse it, but I just want you to understand that this is a complex system. So we're talking about complex multi-agent system here with learning and training and shared responsibilities. There's different kinds of tasks and subtasks, policies that we have to build, that kind of thing. You train the dog to walk towards the middle of the sidewalk so that you can walk to the right of it, and you are essentially walking on the right-hand side of the road, if you will, right? we walk on sidewalks like we drive on the roads unless unless you're one of those ****** delivery robots right next time what look at them they're all over the place and they're in the middle of the road they could have a policy of shorelining to the right walk like a human right that would allow them to interact with because you're like whoa right that's our experience of those robots not zesty not zesty at all so so we have to understand all of the basic subtleties now in this particular case you're seeing on the left the standard gait that the robot, the way the robot walks out of the box, as it will, right? One thing that you'll notice on the left when the robot moves is it has this very high-stepping prancy gait, not because it's fancy, but because that does two things. One is that it helps it avoid trip hazards. It also gives the system more time to plan where the next foot is going to go. and so it's a it's sort of a from an engineering perspective it's optimized right also what's the technical term noisy as **** all right say I'm a blind person I've got my robotic diet guide dog and it's doing this I'm already blind why do I want to hear this noisy thing preventing me from hearing the environment. That's a poor design. That's not a good robot human integration. Also, if I'm walking in a school or down the corridor or a hallway and the thing is making that racket, that's not fitting in the ecosystem. So there's research involved and doing some reinforcement learning efforts, we can make the robot walk quieter. So you see that's a very microscopic thing where we're optimizing for humanness, for quiet. Not necessarily for stealth but for being reasonable. So that level of care and attention and listening to our users and people saying, well, I would never have a robotic guide dog if it was that noisy. We have to go back to first principles of sort of the programming to get this thing to operate in a more acceptable manner. So these two gate controllers differ in their noise and their compliance level. And, you know, so just an example. Okay. Well, we also can ask how should this robot, after we've figured out how it walks, how does it move and how does it turn? Right? So you could, if you look at the original robot, they walk, they prance, right? And they turn like this. So they're hopping in place. Well, if I'm beside it, the thing is turning right into me. And now we've got to go, whoa, okay, I've got to move like that too, right? But if I don't, if I'm blind, then that's not an acceptable thing for me to kind of get out of the way. If I'm walking along, I don't want my robot to have to go prance, prance, prance, prance, prance, prance, prance, prance, prance, stop. Prance, prance, prance, prance, prance. That's not how people walk. We walk and we kind of take corners right, so we need to teach the robot Not only how to stand and walk but also how to move in a graceful way that is compatible with walking beside someone Right, it's like if you've ever walked beside mom and holding mom's hand and then mom all of a sudden stopped and turned That would be a problem right you have to navigate and and operate in a way that is consistent so we have whole programs of research to figure out how do people actually move with a real dog and how do you train a robot to do that as well not surprising once I figured out how to move I've got to figure out how to avoid obstacles because there's a trash can there's a desk there's a fire hydrant there's something in the way other people and it can be a very dynamic situation on the sidewalk or wherever I am. So we need to make sure that the navigation and wayfinding capabilities of the robot are comparable to the wayfinding capabilities of an actual dog. Understanding what is my overall goal. In this case, I'm going from left to right, and there's an obstacle. I can deviate to the left or to the right, but I still have to figure out if I deviate to avoid this big round obstacle, I have to get back onto my intended path. And how do I recover? How do I repair that path planning? That's not super easy, but we leverage what we do know about robotics. We know what our communicative intent is, go to the end of the street or go to my office or whatever. I can set a far destination and let the navigation system deal with the subtleties. So lots of interesting things there. again, ultimately, the human is responsible and is the supervisor, but seeds control, seeds some kind of capabilities to the canine or to the robotic assistant, and we have to understand that collaboration and the back and forth. One of the things that's identified here is intelligent disobedience. You may watch a person walking with a guide dog, and if they happen to be walking towards the corner, the edge of the sidewalk, there's the crosswalk, if it looks like, if it seems to the dog that they're going to step into the road, the dog may very well turn in front and physically block, right. So that's what we consider intelligent disobedience. You might say, let's go, come on, let's go, and the dog says, uh-uh, but they don't say, uh-uh, right. A robot could, but they But they actually physically block. Now, that expected behavior, that intelligent disobedience, is one of the reasons that the dogs have a certain form factor. If you had a little tiny chihuahua guide dog, right? And it turned in front of you, you might very well trip over it, right? But the fact that you've got this dog that is knee height or above, it can turn in front of you and you're going to feel it and you're not going to trip over it as much and it's robust enough that it's going to prevent you from walking out into the road. So understanding how to move, how to get from start to destination, deal with the obstacles, but also some of these other kinds of behaviors like preventing someone from walking into the road, that's not easy. It's truly a collaborative kind of thing that you learn with your canine, with your dog, and we have to figure out how to do that with a robot. All right, well, check, check, check. All these things are things we can work on. Well, how do I actually communicate with this dog or with a robot? One of the things you'll see is that the typical guide dog has a harness, and the harness is used to hold a handle, right? The handle is the real haptic interface between the human and the animal. It's typically a rigid harness, and it is coupled to the dog by the harness in a somewhat rigid manner. So if the dog turns to the left, because of the way the geometry of things, I'm going to feel that in my wrist. The whole handle will turn when the dog turns, right? So I have to be able to feel that movement. If the dog turns to the left or the right or slows down or stops, I need to be able to feel what's going on there. There's also usually a leash. So if you're going in between tight spaces or, for example, into an airplane, you would not necessarily use the harness, but you would let the dog lead and you would hold the leash and walk behind it. So the dynamics of how you stand next to the dog differ. We need to figure out all that in the case of the robot as well. So understanding how the harness and the leash serve as a way of communicating in a haptic manner between the dog or the robot and the human is important. We have other ways of communicating. So here's an example, just as a bit of a sidebar. When I first became involved in this project, Sehun Ha and Greg Turk had already been working working on some of these things, and had been using these robots for some time, and I was somewhat cavalier about how I approached these robots. I figured my mental model was that, oh, if I nudged it, no biggie. but Greg Turks is like whoa don't touch the robot because as some of you who are in this space no it's possible for that robot to freak out and thrash these are very powerful mechanical systems and it's it's not a toy right they're not built for nudging they're not built for well if you walk beside an actual dog many times people will sort of rub their the outer part of their left knee against the the back hip of the dog so that helps them know how far they are and that's where you know when the dog turns they're kind of coupled at the knee as well but we're not touching these robots because they'll freak out well from a human centered design perspective freaking out of robot is not acceptable so I put my foot down in a prancy kind of way and I said I want to be able to nudge this thing I want to be able to push it right and not have it freak out and thankfully Sehun and Greg are the Georgia Tech kind of people who who say, we can do that, it's not what it does out of the box, but we can make that happen, and here you see it. I can nudge that thing, and it'll go. It gets it, right, we've had to build a classifier, and it takes the movement and the joint angles and so on, and figures out what the communicative intent was, and it's like, okay, I got it, I'm moving, right? Well, I want to be able to do that not with my knee because you're not necessarily going to do that, but can I do that with the harness? Can I just pull back on the harness and have it stop? Absolutely. So I pull back, and it recognizes that input. The classifier says, oh, that is a stop signal, and it stops. Right? And this one is slowing down. So I'm not pulling back to stop it, but I'm pulling back enough to slow down. So this is communicating in a very natural way that doesn't require you to verbally say out loud so that everyone else can hear you, right? Slow down. You're walking too fast. Let's go. No reason we couldn't do that as well, and we're working on those kinds of things. But you get the idea that this is really the way it has to be. This is the natural human-robot collaboration communicating in a very physical, normal, natural way, but that's not how robots are built. So Georgia Tech kind of people have to step in and say, we got this. There's another one here. This is slow and it wasn't slowing down, I want to slow it down. Okay, it slows down and say now stop all right slow slow slow okay and I'll stop it looks so easy and and I put on my psychologist hat and I say well of course right that's that's how it ought to be and all the robots are like roboticists are saying this is not easy okay but it can be done there you go all right so you get the idea that we need to communicate in many naturalistic kinds of ways not just using speech and not using typed commands and so on lots of other things we could talk about here in communication but I want to change gears a little bit and talk about modeling we want to understand and be able to model the relationship between the the robot and the person how where is their physical spacing, how wide are they, how long is this team, what happens when you're going through a narrow place like a doorway, what are the dynamics here, what is going on with all of this. So understanding different ways of moving and co-moving requires us to do some modeling. One of the things we have to do is collect a lot of data, and many in the room here, many at Georgia Tech are interested in teaching robots from humans, right? How can robots learn from humans? In this case, it may be robots learning from dogs. So we can bring actual dogs, guide dogs, with their handlers into a tract space and do motion capture and collect all sorts of data about that. and then use that to train the robot so that it behaves in ways that are sort of natural and acceptable. We can get all mathy on it, but here's some examples of the different trajectories that we can apply. So we can try to understand what is happening with the robot in the blue line here, and then what is the sort of squiggly movement of a human, and it turns out that people actually don't move in a very smooth way. We think we're walking straight, but in fact, we're often not. So understanding the realities of how a human and a robot or a human and a guide dog work together and the complicated dance when there's some kind of a turn involved, all of these things are really interesting. Some subtleties that you, when I tell you, you'll be like, oh, duh. But until I tell you, they might not make a lot of sense. So when I'm walking with a companion, whether it's a guide dog, a robot, or my wife, we're holding hands or whatever, or I'm holding the harness, right? If we make a right-hand turn, the companion has to walk faster, right, Because they're walking a longer path, a larger radius, right? So that person naturally has to walk faster and keep up. And I have to sort of slow down, right, and accommodate. But when we're making a left turn, then the companion has to slow, and I have to speed up and catch up and go around. We all know this from experience, but it's not until you actually have to train a robot to do it that you understand it's not the same to walk straight, left turn, right turn, so these are some of the things that we collect the data in order to build models. We can build quite sophisticated models that have sort of different mathematical relationships between the robot and the human. In this case, we see a model known as, or that we call a delayed harness, right? And it's just an example of how we can take the modeling strategy, take data, and be able to program and predict. All right. Let's talk about some other kinds of communication. You can communicate with nudging, with the harness, but also with commands. What should the robot understand? command? Should I speak in German? Wait, where does that come from? Well, it turns out that many of the military dogs, not guide dogs, but the military bomb sniffing dogs and the security dogs are trained to respond to commands in a specific language other than English. Why? So that if the handler tells the police dog or the military dog to attack, right, the other person can't say, stop. Right? So they have their own language. which the animals are smart enough to, and they have learned to recognize the voice, but you understand that they speak in code sometimes, and including a different language. So this is some of the things we could consider. For a guide dog, we just generally don't do that. But what are the kinds of commands? So now we have to look into this in a linguistics way. So we bring in that side of psychology. We break down the set of commands and tasks and then figure out what is the best strategy. Until fairly recently, we would create a voice interface where you would speak some commands. Stop, go, turn, faster, slower. and maybe there was 10 or 15 or 30 commands that the robot would understand. Nowadays, we just throw an LLM at it. And I can just say, well, let's skedaddle, and in theory, it would get a move on, right? So that's a more natural kind of thing. So understanding how people communicate is something important, right? But we can parse this out and understand that eventually, no matter what happens between us, I have some communicative intent, I have some goal like walk faster or go to my office or sit down or whatever, get under the seat. And the robot has to turn whatever I say into some kind of action sequence programming, right? so we have to figure out how to make all that work all right well this previous slide is assuming I am talking at the robot but the whole point of this talk today is that it's not me talking at the robot it's me collaborating with the robot in which case the robot could very well say stuff to me it's a conversation what would that be like what would the robot say what can the robot know right certainly navigational assistance or safety and caution advisories environmental context saying it's crowded here the sidewalk is busy some of these things I would be able to understand myself if I could hear the sounds of people but not necessarily if I'm out on the sidewalk and there's a lot of traffic noise I might not be able to hear how many people are actually walking on the sidewalk itself so there's an opportunity for the the robot in this case identifying whether you're walking into the men's room or the ladies room right often those doors are right side by side and their only differentiator is a little sign. So lots of examples where the robot can certainly and usually does have cameras, outward facing cameras. It can know where it is and can communicate information in lots of different ways. It can go far beyond what the canine guide dog can do. So the canine guide dog is not able to tell you about the weather report or if it's going to rain in 15 minutes. Should you put your backpack or your? So there's some of the examples. If I'm going to walk and it's a 15-minute walk from my office to the bus stop, I want to, and I have a dog and a backpack and stuff, I want to know if I should put my jacket on first and then my backpack, right? And maybe put the little cape or the jacket on the dog before I leave. otherwise if you get out there and it's raining oh no you're on the sidewalk and you're trying to figure out get your jacket on and this dog and whatever so anticipating these kinds of things and we can certainly use AI and machine learning your nest thermostat can understand learn your habits your comings and goings why can't your robotic assistant right and then be connected to the weather connected to the temperature the rain all of that good stuff sorry and deliver ads yeah well yes we'll get there all right so not only is it the behavior the technical the communication stuff but what does this thing look like so this is going down the road of delivering ads here's a question should a robotic guide dog be smooth and plasticky or furry let's take a little poll who says smooth and plasticky all right who says furry who says I don't know why why smooth and plasticky very psychological of you thank you yes also you might want to clean it right you might not want to have the pollen get in the fur of your of your robot right no fleas yes absolutely now some people said furry who who wants to volunteer why furry it's like a real dog cute cute and not scary right yeah so there's lots of also very psychological yeah for sure so here's a fun fact we don't give canine guide dogs to children very often they're not the greatest in terms of being responsible yet but also they spend most of their time in school. If you took a big furry dog into school, it's a pet as far as all of the other kids are concerned. Right? Oh, let me touch your dog, let me touch your dog. And we adults know that this is a working animal. And we don't go up to, we've learned as much as we want to, we don't go up to working dogs and pet them. Right? But kids can't help themselves. But what if we made the pediatric guide dog spiky so that the kids, other kids, have no interest, right? It would prevent, it would likely lead to those kids not messing with the guide dog. What would be another effect of that? those other kids would not want to have anything to do with the kid who has the spiky guide dog so again social psychology context, ecosystem understanding all of that we have choices about how we do these things so we do some research we ask a bunch of people who have dogs who have lived with people who have dogs who don't know anything about guide dogs and we get lots of responses. Many people will say, oh, I want this, I would think this preferred height of the thing should be small, like 12 inches, like a little chihuahua, right? And many people say it should be medium, some people say large. This is another example of sometimes users don't know what's good for them. So participatory design that we like to do has its downfalls when we ask people to make design choices and they're not experts. So we have to collect data but also understand that even if we were to build a large Labrador sized dog, a robot, we would then have to educate people about why that's the right size as opposed to to the Chihuahua that they might have preferred, right? It's like why we have to educate people why a large SUV is the correct vehicle for driving in the middle of the city, right? I jest. So lots of things about the look in the field, the behaviors, how should it walk? Should it be like tough walking? It's tough walking. or whatever tough walking is, right? How should it behave? Should it be assertive and say, we're going through the crowd. Get out of my way, right? Or should it be hesitant? No, you go. Should it be Canadian? I can say that. All right, so lots of things that we want to think about here. It's outer surface. We can ask people about whether they want it to be metallic, soft, fluffy, plastic-y, or anything else. Fluffy, right? And people will say all of these things. Well, when we're asking people about their choices for the design and the outward look of things, and we see this kind of variability, what comes to mind? Customization. Personalization. What if I had a smooth robot that I could have when I wanted a smooth robot, and then I had some clothes, some kind of jacket thing that it could wear, right? Well, that sounds absurd, except we know from Becky Grinter and other people's work that people will dress up their Roombas, all right? All right, so lots of possibilities here, but what happens when you change the look and feel of something? One of the reasons that white canes are white and always look the same is so that people know what it is, right? Now if I have a robot that is sometimes furry and some robots are spiky and some robots are smooth I may not recognize that this is a blind person with a assistive robot and I might think it's their toy robot their pet so we have yet another experience of ecosystem fit all right well one thing that didn't surprise us is more battery more better right again we know this in in terms of automated vehicles this is the concept of range anxiety right and people with disabilities are are absolutely concerned about their technology failing so this is something that we we need to be aware of how do you control it voice push buttons and so on lots of possibilities so again we can look at the sort of industrial design of this and some design students have been involved in this it's been very interesting to see what we have so bottom line is we end up with some summary of the things that it can do the way it looks the way it walks all of that some of these are easy like we can make it make a robot that's furry some of them are hard like making it turn properly and avoid obstacles and not run into me and not freak out but in all cases we have to design this collaborative pair the the robot assistant and the human have to work hand-in-glove to accomplish this goal of getting from place to place not disturbing other people being able to understand what's going on and and do so in a way that is fun and and and exciting and and not a chore all right so I'm going to circle all the way back to chart bring this home if I can and And that is this robotic guide dog is an example of an interdisciplinary research team tackling many gnarly challenges, right? They're gnarly because people are complicated. And we live in complicated worlds in society and environments and stuff. So anytime there's a robot or a technology that is actually integrating and collaborating with a person, it's going to be gnarly. All kinds of issues of design, function, communication, adoption, abandonment of technology, policy. Do you all remember with a little chuckle, ha, ha, ha, when someone said, oh, we're going to have scooters. You know those scooters, but they're not going to be on the sidewalks. That's the chuckle. Right? There was a policy decision. Someone said, safety first, right? Those scooters are going to be other road users, like bicycles. They're in the bicycle family. Now they're not, right? So what is the reality of how technology and people work together in connection with all of these other elements of society, right? how will you take your robotic guide dog onto an airplane it's got to go through TSA it's got to go through the scanners you're gonna fold it up and have it go through the x-ray machines maybe it's got a big battery that looks like something other than a big battery right that's policy stuff that's the practical realities now does a robotic guide dog count as a carry-on can it go in the overhead compartment right these are the realities of people and technology coming together in a collaborative way and in sort of co-dependent co-reliant All of this is exactly what Charts seeks to tackle, and all I would say is, come join us. It's fun. And with that, I'll say thank you. I'm happy to take some questions with whatever time we have. And Chris, do you want to moderate, or do you want me to- No, I can walk around and hand off the mics. Oh, please do, yeah. If you just want to project, you can. oh yeah we're doing online okay great talk thank you so it's clear you thought deeply about the design space and in some cases you talked told us about choices you made and gave intuitions about why at a higher level it seemed like in some cases you said we're gonna stay as close to a natural guide dog as possible. In other cases, you diverge from that. For example, you used a robot dog rather than some other kind of robot, right? Or rather than a quadcopter, which you could have had, right? On a string. But in other cases, like dogs can understand speech commands, but they don't talk back to you. And you chose to let it talk back to the human. So how did you decide when to stay close to the biological role model and when to expand things or to go away from that? That's a great question and I think everyone heard that. I'm going to punt and say that this is conveniently a research project and it's not a product. It's not a thing where we have to make final decisions and put it in the marketplace. So we can, what if it could talk? Well, what would that be like and we can have some students run with that what if it were furry that's explore that space so it's an exploratory and that's the great thing about the university is that we can do that and we don't have to necessarily you know get it right because at this point we're we're researchers right that kind of gets us off the hook a little bit but it allows us to do different things now there are certain constraints for example the the alien go robot is a four-legged quadrupedal canine like thing and we're interested in how would something like that integrate with people and be a collaborative partner we're not interested in designing a robot from the ground up right so there are some places where it makes sense for us to use what's already there and just accept those constraints for now someone else might decide well maybe six legs is better but for now we're like all right we got what we got right it's kind of like yeah yeah just this probably be a last one because we have three minutes until the next thingy starts all right yeah this could shoot can be a quick question yeah thanks Bruce for this and I just have one question for the center, the chart. I think it's great, the idea is great, and it seems to work truly interdisciplinary. But I do see people, this center might involve people from very different backgrounds, people doing robotics, people doing AI, and people from psychology. And sometimes when people talk to each other, I feel the language is very different. and the background, they come from different training background, so I wonder if you see any challenges there and maybe potential solutions, how to maybe truly integrate people from very different areas altogether. Yeah, absolutely. I think many of us in the room are very familiar with this concept of interdisciplinary teams having language problems. Not necessarily the language, but the communication, and the words that you say and the way you refer to problems and the way you conceptualize things can be different. This is a symptom or a challenge of any interdisciplinary project. This is not new to chart, but it's also something that's very well understood in teams science, and it's really one of the communication or the cores is this communication. So we want to leverage and bring in what is already known, train teams of people to be teams, train people how to be on an interdisciplinary team and be part of that effort and sort of fix those communication issues. So education is part of the chart mission and not just educating students, for example, but educating researchers how to be on interdisciplinary teams. All right, I think we're going to have to wrap, but I also want to point out that we do have in this room, right after this, for those of you who are interested, There is a meeting to address the...