00:00:00:00 - 00:00:32:21 Unknown This projet is applying AI, Autonomy, and Digital Twin for Radioactive Environments and those embodiment. We have the PI is Brett Bonn. He is the senior manager of nuclear operations for Boston Dynamics. Brad’s primary focus is to work with customers and partners to ensure their robots keep people safe, reduce cost, PPE waste, and improve mission effectiveness in hazardous scenarios. 00:00:32:23 - 00:01:09:22 Unknown Also, we will have a co-lead. Co-PI Brian Ringley is a principal manager at Boston Dynamics where he leads digital twin product development, specifically industry facility mapping to support enterprise asset management and anomaly detention for preventative maintenance. Before coming to Boston Dynamics, he was a Senior Construction Automation Researcher at WeWork and a Senior Associate on the design technology team at Woods Bagot. 00:01:09:24 - 00:01:45:16 Unknown Please help me welcome Mr. Hunter and Mr. Brian. Good morning, everyone. Thank you so much for your time. We appreciate the opportunity to speak here today. A quick point of order. So, Brian, really unfortunately has had a bit of a delay due to a family loss. He said he is going to attempt to join. So we may have a portion of the presentation where I will extend my discussion portion and then take some questions depending upon his schedule. 00:01:45:18 - 00:02:38:23 Unknown So we appreciate everyone's understanding in the last minute changed, their life occurs. So I hope everyone's having a good morning and I appreciate the chance to speak to you all. So we were asked to come and join this presentation as a representatives of where robots have be used in the real world and to discuss and share some of the stories that we've experienced firsthand where technologies like AI, mobile autonomy and dynamic sensing have really kind of changed the way that a lot of people operate in hazardous environments and also in general industrial environments, environments that may not be directly dangerous but might be unpleasant. 00:02:39:00 - 00:03:12:06 Unknown And there are tasks within these environments that people could do, but frankly, their time would be much better spent elsewhere and they don't want to do these dull, dirty or dangerous tasks. So I'd like to share some of the history around robotics operating inside of nuclear environments, but I'm not going to spend much time there. I'm actually sure that everyone here knows even more about the history of operating inside of hazardous environments than I ever could. 00:03:12:08 - 00:03:43:04 Unknown I've been a radiation worker. I've operated spot inside of some of the most dangerous places in the world, some of the most well known nuclear accident sites in the world. And it is something that, as excited as I am about the technology, I find that I'm sometimes even more excited by the places that this technology has taken us and the things that we've been able to accomplish there. 00:03:43:06 - 00:04:14:02 Unknown So I'm going to talk a little bit more about how this evolution has taken place. Robots really did, I think, have some of their primary origins in situations like nuclear environments and because we're looking at a situation of humans can simply not go there and yet we must. There are things that need to be seen. There are things that need to be manipulated, moved, removed. 00:04:14:04 - 00:04:45:13 Unknown How do we do this? And the concept of creating a remote presence for a human being inside an environment that a human being could never go, really began the concept of remotely operated CD manipulation systems, remotely operated sensors and cameras. This was necessity and this particular type of environment was the you know, we talk about, you know, necessity is the mother of invention. 00:04:45:15 - 00:05:20:15 Unknown I sometimes add a little corollary to that of desperation is the catalyst for genius. And we see some of the most incredible environments when we pair this sort of necessity with people who are extremely versatile engineers and want to innovate into areas that no people have ever done before. And in the nuclear world. These early stage robots, they were so heavily dependent upon human beings for managing and controlling every little aspect of movement. 00:05:20:17 - 00:05:51:10 Unknown How they perceived the world was entirely through the human eye as just being relayed through cameras and individual joints where effectively under human control. This was very, very granular teleoperation. And as a result, it required very discrete and very skilled human operators to be able to accomplish these tasks. And it was something that had a a very narrow segment. 00:05:51:12 - 00:06:19:12 Unknown You also saw a lot of highly specialized configurations. These robots would be set up with exact tooling and a very, very low level design that was purpose built for a particular task. And as a result, they didn't have a lot of expandability to other areas and they were also extraordinarily expensive. And this was kind of where this world began. 00:06:19:14 - 00:07:10:10 Unknown And as time has gone on, parallel technologies have been developing in multiple different sectors and there's kind of a natural evolution that's been taking place within the world of robotics thanks to the massive miniaturization of highly powerful compute systems, we've been able to adapt software controls to more and more hardware systems. We're seeing a merging of the command and control systems that can be software either implemented or simply software supplemented, and that's giving us mobility and perception on a level that robots have never really seen before in the world. 00:07:10:12 - 00:07:59:23 Unknown The thing is, too, we've seen developments in sensor technology. Cameras have massively improved in the industry, but then beyond cameras, we have sensors that can detect heat, sensors that can detect radiation in a visual scale. Sensors that can detect three dimensional objects map those three dimensional objects. All of this data can also be fed into computer systems to enhance that perception of the world and then present that perception to a human operator or to someone who isn't an operator, someone who is an engineer or an operations manager or facilities engineer in a way that they are able to interpret and manage it. 00:08:00:00 - 00:08:42:22 Unknown We've also got better wireless technologies. We've got the ability to send massive amounts of data over a radio connection without the need for a very heavy cable or a fiber optic connection. We've got relay radio systems, mesh radios where we can extend the range of communication systems in ways that were never thought of before. Combining all of this together with what we kind of use as the catchall phrase for A.I. is making it possible to apply machine learning and additional alga rhythmic analysis to the data that these robots collect. 00:08:42:24 - 00:09:15:14 Unknown And that's enabling us to do more than just repeat, teach and repeat automation. We have the ability to actually have automation systems that can react to their world, respond to their world, and make decisions on how they operate within that world based on the parameters that we provide. We talk a lot about intelligent mobility at Boston Dynamics and those of you who've seen some of our YouTube videos, you're familiar with some of our robots. 00:09:15:16 - 00:09:36:23 Unknown We have our somewhat famous Quadruped Atlas. This is a hydraulic demonstrator. This is the humanoid. This one has just recently retired in favor of the electrical atlas and the one that we have commercialized at the time. At this moment, four legged mobility is spot. And we'll be talking a little bit about how spot's been used in this space. 00:09:37:00 - 00:10:01:07 Unknown But I want to kind of clarify when we say intelligent mobility in those early robots, when we responsible for very low level controls of everything that the robot does, which tread is rotating in which direction, at which time, or if it's a legged mobility, are we directly telling the robot, okay, I need its foot to go right over here? 00:10:01:09 - 00:10:33:10 Unknown We've been working on ways to abstract away the specifics of how a robot maneuvers itself through an environment from the operator and the operator is effectively simply giving the robot trajectory. It says, Go this direction, walk up those stairs, go forwards, go backwards, sidestep all of the placement of the feet, all of the understanding of which surfaces could be stepped on what surfaces or not. 00:10:33:10 - 00:11:08:02 Unknown Step a ball. Where are the obstacles in the environment? How do I navigate around those obstacles? All of that is now capable of being handled by the robot, so the operator is able to focus upon what's around them and not have to incorporate that into how the robot moves around them. So again, this is an evolution of what we've developed over the many years with kinematics and controls, but then combining it with perception. 00:11:08:04 - 00:11:41:05 Unknown So it becomes this effortless movement. So classically people were really surprised when they would assume they would have assumptions about what spot was doing underneath the hood, and they would be absolutely shocked when I would tell them that our first iterations of Spot had no air whatsoever. There was no machine learning, there was no iterative, algorithmic data based training in the earliest models of spot. 00:11:41:05 - 00:12:12:22 Unknown It was actually model predictive controls, and it was understanding that taking a step operates somewhat like an inverted pendulum. Over time, though, we really started to learn that when you have low dimensional problems, those predictive algorithms can be made extremely reliable. But while stepping is something that could be described as a low dimensional problem, it can actually be prescriptive. 00:12:13:01 - 00:13:01:24 Unknown Written out. Walking is an entirely new dimension, and we found that by applying iterative machine learning within simulation to certain components of the overall stack of software that makes a robot be able to walk had an absolutely incredible difference on how that robot maneuvered. And so in our case, we had a, a, a large dimensional problem of when the robot is going to move its leg and take a step, how fast does it need to move its leg to the new position on the surface that it's going to try to come into contact with How high does it need to lift that leg? 00:13:02:01 - 00:13:27:24 Unknown How quickly should it make a correction? If that leg appears to be moving, then when it's in contact with that surface? These are all things that, given sufficient time we could absolutely accomplish creating algorithms of our own to do this. But because it was so high dimensional, we began to apply machine learning to that individual aspect of it within simulation. 00:13:28:04 - 00:14:04:16 Unknown And the result is that we found the robot would walk in a ways that we would have taken so, so much longer for us to develop. The top image that's animating is kind of one of our classic little humorous moments, but it's a real world example of what I'm talking about. You can have a robot slip on a banana peel, and the reason was the robot did not have a way to quickly respond to a situation where its limbs suddenly had a different coefficient of friction with the surface they were coming in contact with. 00:14:04:18 - 00:14:35:17 Unknown This was something that really didn't have a quick, easy way to solve. Using kinematics down to the bottom. What you're seeing is where we've got our latest version of the software. This is where we've incorporated AI into certain components. So remember, this is actually not completely and totally AI Spot has not learned to walk from scratch. This is a mixture of prescriptive algorithms and machine learning. 00:14:35:19 - 00:15:08:17 Unknown And so now we can actually lubricate a ultra smooth surface with an industrial lubricant. And you've got the banana peel there, of course, for, you know, for old time's sake, and the robot's able to walk on to the surface. It can also step up onto and off of surfaces that are very slippery. It's a tremendous change. And this is an example of really where the application of machine learning is not macroscale. 00:15:08:17 - 00:15:36:10 Unknown I think that in the world when we say A.I., people assume that this robot is deciding what room to go into or it's deciding, Hey, I'll, I'll move this piece of equipment this way. The robot's actually not doing any of those decisions at all. That's all in the hands of the operator. It's all in the hands of the person who's programing an autonomous function with the robot. 00:15:36:12 - 00:16:10:17 Unknown Those are all very prescriptive and very within human control, where the AI we find to have the most valid and effective application is kind of in the microcosm, not the macrocosm, the small components of operation and autonomy, where it would take us potentially decades to get the reliability that we need for that one small component. This makes all the difference because AI is able to iterate upon all of the data sets that we put into it over time within simulation. 00:16:10:19 - 00:16:46:01 Unknown And we take that simulation and we apply it directly to how the real world robot operates. And the result is just tremendous. Walking robots are fantastic, but ultimately if they're just walking, they're not really providing any help, they're not providing any value, they are just walking. What this type of intelligent mobility does for us is give us the ability to go places that any human can go with the exception of swimming, climbing ladders. 00:16:46:03 - 00:17:18:11 Unknown This makes it unlocking any kind of terrain, especially in the cases where we're dealing with nuclear decommissioning. Flat floors or perfectly aligned stairs are just not the expectation. The reality is that we're dealing with all kinds of hazardous tripping situations, laws that could potentially slip or collapse underneath us, all kinds of challenges to what would be traditional robotic mobility and even challenges to a person. 00:17:18:13 - 00:17:46:11 Unknown BE We see situations where even a human being would be having to take extreme care to navigate this environment, even if we could enter it in the first place. So the robot is bringing us to a city to these locations and giving remote situational awareness once we're there. The robots enhanced sensing capabilities. When I say that, I say it in terms of the devices that are being utilized on the robot. 00:17:46:13 - 00:18:18:10 Unknown You know, when people will ask us or can spot create a digital twin, well, the robot can bring the LiDAR sensor and the software into an environment that is then used to create the digital twin. Ultimately, the robot is kind of that literal moving part of a very complex and complete solution to create that digital twin that also counts for things like radiation sensing and radiation mapping, which I'm going to go into detail on in this presentation. 00:18:18:12 - 00:18:59:17 Unknown So the mobility leverages A.I., but then we also have intelligence in how we take the data that the robot collects and apply it to the world. I like to talk about the concept of non repetitive autonomy, the idea of an automated system that's able to adapt to an environment. And while it is performing a very prescriptive task, the nuances of how that task is implemented is being handled by an intelligent software and computing system that is able to make subtle changes to how it operates. 00:18:59:19 - 00:19:47:02 Unknown Great example on the left here is where Frame at Home has made an application of a surface contamination detector with spot. This system is actually called Amtrak. I actually can't remember exactly what the acronym stands for. The description of it, however, is pretty straightforward. The robot is able to enter a room autonomously, scan the three dimensional environment and present that to the operator, and then the operator simply selects in that 3D environment what flat surfaces, whether it's walls, floors or even curved surfaces like containment barrels that they want scanned for surface contamination. 00:19:47:04 - 00:20:24:02 Unknown The robot then scans through and checks the surface at a distance that's prescribed by the operator. So the operator, for example, might say that I want you to scan this surface from two centimeters away and the robot, using a combination of their software and our manipulation with our API, is now taking that 3D model that it has created for itself and moving at surface contamination, sweep at a designated speed at a designated distance, and then it will highlight in the software where contamination has been detected. 00:20:24:04 - 00:20:46:13 Unknown They've actually even adopted an optional add on where there's a small spray paint adapter where the robot will actually spray a marker anywhere it finds contamination so that once the sweep is done, a human being can actually visually see on the wall or on the containment barrel, on the floor surface, on a truck where that contamination was detected. 00:20:46:15 - 00:21:09:14 Unknown So you can see it in the software. But then you also have a real physical world representation of where the container contamination was. So while the operator is dictating what areas to scan, the robot is abstracting away all of those low level controls where an operator might need to be using a joystick with very fine grained control to make sure they stay a good distance away. 00:21:09:18 - 00:21:36:22 Unknown They have the proper movement speed, all of that's handled by the robot itself. So we start to see that automation that's high level decision making all within human hands and yet the application of intelligence from the computing side and from the software side has made that process much more efficient and much easier, much more accessible. Of course, then we take that information that's being collected by robots that are conducting autonomous surveys. 00:21:36:24 - 00:22:07:00 Unknown And for example, daydreaming is something that we hear about from a lot of our customers. They want to be able to take a visual survey that includes getting the state of equipment. Well, that might be a thermal scan. They might need to understand how hot these transformers or these motors are from the visual indicators. However, they don't want to necessarily take time to go through and read all these gauges manually, not just in terms of the walking distance, but even just reviewing the images. 00:22:07:02 - 00:22:52:24 Unknown So AI has made it so that that image can actually be interpolated, so we can actually interpret the value that the gauge is indicating. So what the operator receives is a trend line of what that gauge actually indicates. We can apply this to a lot of different sensing capabilities and use cases. So if there is a particular valve that has a visual nature to it, we can actually see the travel of a valve that could be incorporated into machine learning to have an automated system to indicate to an operator a toggle that was open, that was closed rather than the operating need to look at the actual discrete image itself or a video feed and 00:22:52:24 - 00:23:20:19 Unknown try to say, okay, is it open or is it closed? It becomes much more fluid. And so again, that human operator, they're the ones deciding what do we do now with this gauge at this value? What do we do with these valves in these positions? This is why I like to see A.I. as a one small component of a much wider solution and wider system. 00:23:20:21 - 00:23:48:19 Unknown So at this point, I would be handing this over to Brian Wrigley, and he's going to talk specifically about the application of robotics to digital twin. I believe, however, that he is still in route. I'm here actually. Oh, you're here. Yes. Fantastic. All right, everyone. Brian Ringley who is in charge of our product management, specifically focusing on digital twin. 00:23:48:21 - 00:24:12:01 Unknown I am not the principal digital twin expert at Boston Dynamics by any dimension whatsoever. So Brian thankfully has as joined us to cover this portion and then I'll come back in to talk specifically about radiation use cases once he's gone through. So Brian, take it away. Thank you. I'm going to switch which screen is. Oh, I was I could have. 00:24:12:02 - 00:24:49:11 Unknown That's fine. All right, great. I can I can see it through the camera in the room that it is displaying properly, which. Oh, see, There we go. Yeah. So I'm going to talk a little bit about the there's something really interesting about mobile manipulation robots and their abilities to both build twins through their perception and measurement of environments and then actually to also control environments from twins or manipulate environments from twins by virtue of their manipulation capabilities. 00:24:49:11 - 00:25:15:13 Unknown So it's a really fascinating, I would say, hardware moment in the kind of the history that is digital twin, which is, you know, largely a software methodology. So the the digital twin, one of the things that, you know, my background is architecture and construction before I got into construction robotics and then and then essentially all sorts of robotics across domains. 00:25:15:15 - 00:25:54:20 Unknown But I've always been interested in, you know, complex 3D modeling, building information modeling facility, information modeling, as well as reality capture, laser scanning, photogrammetry and things like that. And one of the things that I've always had trouble reconciling is the digital twin is really a product level concept that originated, you know, arguably with work that was being done at NASA and thinking about the affordances is of new software capabilities, simulation in particular to be able to kind of test sense and refine a feedback loop for the design of things like rockets and engines. 00:25:54:22 - 00:26:36:12 Unknown But people kind of picked up that there were some core tenets of the twin that actually applied to the facility level. And when you frame it that way, you you start to think of it as a facility lifecycle management tool. And that is my extremely, I think, boring way of describing what the digital twin is kind of in, in our domains thinking about robotics and industrial environments, specifically with some of the radiation things we're doing like decommissioning, you know, that's, that's at one end of the lifecycle, whereas we have customers that are measuring construction progress at the beginning of the lifecycle and you know, everything. 00:26:36:12 - 00:27:25:15 Unknown And then most of our customers are really trying to monitor operations of a facility which is happening in between. So so again, it's it's about how do we kind of take take these things that originated in a product lifecycle management context and really understand how those principles apply to adding value in kind of a facility level context. So like I said, you've got design and construction on one end, you've got decommissioning on the other, and then operations, you know, whether it's this is an example of like a commercial real estate twin where you might be sensing anything from air quality to like tenant access to our kind of industrial customers where you're sensing things like 00:27:25:17 - 00:27:56:13 Unknown how hot a motor is with the level of a gauges things of that nature. But in all cases you're starting with sensing and you're saying there is an advantage to using mobile robots in these situations because for whatever reason, fixed sensing either doesn't work or isn't enough. So in design and construction, fixed sensing doesn't make sense because there is no fixed environment yet it's an environment in progress. 00:27:56:15 - 00:28:22:07 Unknown So you introduce mobility that makes sense there. On the decommissioning side, you have a lot of areas that are pretty damn dangerous and you know, either the fixed equipment is going to hold up there or just the fact that somebody has to go in in the first place and install it is a problem. And then even in operations where fixed and sensing is actually quite successful for a lot of our industrial customers, it doesn't quite fill all the gaps. 00:28:22:07 - 00:28:47:01 Unknown It's often tied to a particular machine or it's kind of premised on the fact that it needs to be providing information at like 20 hertz where you might just want to 3D scan, you know, once, once a month. Right? So the cadence of that feedback is a lot different. It might make more sense to have one sensor moving around your environment rather than instrumenting your environment with many sensors. 00:28:47:03 - 00:29:11:16 Unknown So again, when I think about the twin and facility lifecycle management, what I'm thinking about is how do we combine the right software methodology with the affordances of our robotic products? And you know, from where I sit, it's really hard to separate the software product from the hardware product. When I'm talking about robotics, I'm talking about software and I'm talking about software, I'm talking about robotic. 00:29:11:22 - 00:29:37:21 Unknown But if we go and talk about what are the actual pieces of the twin software or methodology and thinking of it again as a facility level phenomenon, then you have a physical environment with physical assets. The distinction is important and then you have your virtual model with virtual assets. And as far as I know, no one has a kind of like atom two atom digital twin. 00:29:37:21 - 00:30:02:00 Unknown So I did a little bit of blurring on the road because presumably that information fidelity is is not full and nor does it need to be. And then you have your various communication channels, right? So you have sensing to get the physical to the virtual and then you have control to get the virtual to the physical and a dogmatic or true, you know, digital twin would operate in both directions. 00:30:02:00 - 00:30:23:23 Unknown But, you know, I think it's really a case by case basis, too, where there's value in a given application. So we don't have to harp on that too much. So as I said, traditionally, what you have is a situation where you have very specific coverage of a facility because it's tied to fixed sensors. You have to predetermine where you're going to measure in sense here facility. 00:30:23:23 - 00:30:51:11 Unknown You have to make that decision and then you have to kind of maintain that installation. And these installations like traditional Iot, by and large, is very high frequency. You're my you're measuring things like airflow and water flow and pressure. And those things need to be measured many times per second to check for irregularities and to be able to alert when something is going wrong and what that results in is selected of resolution. 00:30:51:11 - 00:31:15:07 Unknown And that's fine if your application is premised on sensors. But when you get to the idea of the facility level twin that starts to leave some gaps in your knowledge about your facility. You might have areas where you don't understand what's changed in your environment or where levels of a particular chemical or radiation are outside of an acceptable threshold. 00:31:15:09 - 00:32:03:09 Unknown So the you know, one of the things that's been really fascinating about working with Spire is its ability to carry sensors on its back and take one sensor or one collection of sensors and move those around the totality of an environment or the portion of the environment that you care about without the need to predetermine the exact places where you need regular data to just say, I basically have variable coverage of my environment by virtue of controlling the roots of this mobile robot and and in that interface, because a robot is moving around and must know where it is in order to do that autonomously through localization, then you also get the ability to say, 00:32:03:09 - 00:32:23:18 Unknown well, now all of that data or those measurements, whatever that robot is capturing, that can be correlated with a specific location in your facility. And we could show you the robot map and some people look at the robot map, but the robot map is for the robot really. That's for the robot to understand where it is. You want to see something you're used to working with. 00:32:23:18 - 00:32:56:07 Unknown So I think you'll see it's common to have interfaces that have blueprints and then we correlate the blueprint location with the robot map location so that when the robot finds a piece of data that you've indicated you would be interested in something outside of a threshold, for example, then that will be placed in its location. So you get the ability to view the data in the facility context and then you can, you know, dial in to that and have this offline management. 00:32:56:07 - 00:33:20:08 Unknown So once that data is stored in the facility model or, you know, if we were just to call this the twin or an example of a twin, then that's a place where you can do offline editing. So if you decide, Hey, I actually didn't want to be I didn't want to be warned at, you know, 26 Celsius. I want to be warned at 30 Celsius because, you know, I'm getting warned too much. 00:33:20:08 - 00:33:51:10 Unknown It's not actually a problem. I can make those adjustments offline. And then the next time the mobile robot circulates through the environment to update the data set, it does that with better knowledge. So in that way there's this kind of feedback loop between how the robot is alerting you to conditions in your facility and then how you're managing when the robot alerts, because it should be a kind of a collaborative system where the robot is simply elevating the most important information to human operators so that they can do their important work. 00:33:51:10 - 00:34:09:09 Unknown So in a sense, it just makes sure that that human talent isn't wasting its time looking for the right things to be working on. The priority is essentially set by the system. So if we look a little bit under the hood for, you know, those that are kind of curious about how these things work, you know, spot in particular. 00:34:09:09 - 00:34:37:12 Unknown And in fact many kind of ground mobile robots use ground negation where they're basically, you can imagine like they're dropping little breadcrumbs. Like these are locations that I remember being in these were the features of those locations and then creating edges between those so that they can navigate that graph. So the more the robot crawls around and records your environment, you know, the more places it can go and the more routes it can generate to go to the places that you care about. 00:34:37:14 - 00:35:07:16 Unknown It then attaches what we call snapshots to that which are point clouds that it's generating through its light. Our sensor. And those things are tied to each waypoint. And you can see that they're not globally accurate, right? They're not meant for human consumption. They're not meant for like globally accurate navigation. In fact, you could argue it works kind of a similar way that our minds work, which is that I know the rough dimensions of the room around me and recognize that and therefore know where I am. 00:35:07:16 - 00:35:27:09 Unknown Could I sit down and draw like an entire, you know, floor plan of the entire neighborhood and have that be dimensionally accurate? No, probably not. That's not quite how navigation works. But we do understand that, as I mentioned before, it's important to be able to correlate this data to the way you think about your environment, the way you manage it. 00:35:27:09 - 00:35:51:03 Unknown By and large, that's blueprint. So while we supply these things, we know that we know that they need to be tied to other systems, to 3D models, to do 2D plans and to other types of data. So what you see on the right is, is another image of a 3D point cloud generated by spot default. It's extremely sparse. 00:35:51:03 - 00:36:15:17 Unknown It's intentionally decimated to be able to compute very quickly for the robot. And that's good. That's good for the robot to get its job done. That's not necessarily good for you to, for instance, take a radiation measurement. And every one of those points that might be too sparse. So this is why we also augment the robot with things like the bulk arc laser scanner. 00:36:15:17 - 00:36:46:15 Unknown So what's playing on the screen right now? We have partners at both Crytek and partners at LEGO Systems who make the B ARC scanner, and this is them with a combined system where you are measuring the space in three dimensions by laser scanning and getting, you know, many thousands or even millions of points of your environment. And then with created this technology of sensing radiation, you're actually able to embed those radiation levels into that. 00:36:46:15 - 00:37:08:02 Unknown So this this is really interesting to me because I think in typical spot operation, we're either doing discrete sensing and getting data points or we're going out for like a construction customer and we're measuring an environment. And this is this moment where the two things come together. There is no distinction, the environment and the 3D measurement of that environment. 00:37:08:04 - 00:37:32:12 Unknown That's also the data points you want to have a three dimensional model where essentially every piece of that model has embedded information in it. And in that case, you know, radiation levels for decommissioning work. So I think that's a really that's a really interesting place where these kind of two worlds for spot sensing collide. And it's quite specific to this industry, though. 00:37:32:12 - 00:37:53:13 Unknown You could also imagine it for things like sensing gas within a volume inside of an environment. So I think it is more generally useful as well. So if we go back to our diagram. Hey, Brian. Yes, we've got, I think 3 minutes left in our slot. I went a little slow because I wasn't sure if you were going to be okay because you're covering so many. 00:37:53:14 - 00:38:21:10 Unknown Yeah. So would it be okay if we switch to some of the examples of doing this in the real world? Yeah, of course. Yeah. That was basically kind of the end of where I was getting to. So yeah. Let's switch back over to you now. I'm really sorry to do that. No, no worries. All right. So I'm going to go back to my deck here. 00:38:21:12 - 00:39:06:19 Unknown So just kind of perfect timing, though, to carry over into how this type of technology is being applied to the real world. We're really amalgamating together these different technologies. We're putting together the concept of simultaneous localization and mapping 3D point clouds and then applying sensor data to those point clouds for both determining scalar data at individual locations, but also even to colorize those point clouds and make it so that a human being whose entire uprooting this imagery and this content is able to apply their own knowledge of what's going on to the environment. 00:39:06:21 - 00:39:35:23 Unknown So upper left image here, this is a colorized point cloud based on radiation levels. The upper right is a breadcrumb trail that's highlighting by scalar dose rate. That's actually an FC pad at a U.S. nuclear power plant site. And they had the spot operating autonomously on this route. And we deliberately said, don't tell us where the newest fuel is. 00:39:35:23 - 00:40:07:20 Unknown The robot's going to show you. And so they can immediately see it highlighted where the most active fuel in the casks is. So right away you have that immediate access to that. The bottom image is one of my favorites. This is actually another nuclear power plant in the U.S. that has been applying a gamma reality e 3D sensor for radiation to do a pre survey at power of an area of the power plant. 00:40:07:22 - 00:40:46:00 Unknown And then during a shutdown, they installed additional supplemental shielding. They then had the robot run the same route again, create another 3D model highlighted by radiation to determine if the additional shielding that they installed was effective at shielding that area. So our friends in the U.K. have been doing some pretty impressive things with regards to both remote manipulation and also digital twin at the Dune Ray facility, as well as the Sellafield facility. 00:40:46:02 - 00:41:31:12 Unknown So by entering that have been sealed off for decades with a robot that can provide both telepresence but also tell manipulation they've been creating all new Visual 3D twins of these environments highlighted by radiation dose rate and have been able to isolate exactly which pieces of equipment inside the facility are the most radioactive. So they're actually starting to be able to make long term plans about how they're going to disassemble this facility, what components they need to work on first, what are the surfaces that are the most radioactive, what are the components and the systems and the areas that have the highest dose rates? 00:41:31:14 - 00:41:58:12 Unknown One of my favorite things about this is the output that we are getting from the sensors. That spot brings there all of these stairs. These are actually where spots starting at the bottom goes all the way up to the top and creates both dose plane maps and also a 3D point cloud that's human readable. So it's this is beyond the point of just navigation. 00:41:58:14 - 00:42:36:23 Unknown The robot has that local subjective view of the world for navigation. But the combination of sensors are performing a loop closure mapping to create something that we as human beings can see as a really a model of that environment. Now, when you pair this with a directional radiation sensor in software, which is what our friends both at Crytek and Gamma Reality have been doing, the result is that you now have a it's like a I don't know how to describe it besides just a magic camera. 00:42:36:23 - 00:43:03:06 Unknown You have this ability to immediately see visually which pipes are the most radioactive, what areas of the tanks are most radioactive, what sections of the floor are going to require the most attention? During this process, the commissioners at Dune Ray have actually had multiple cases where the equipment they thought was going to be the most radioactive was in fact not. 00:43:03:08 - 00:43:43:01 Unknown And it was a different component entirely that had the highest dose rates. Fukushima Daiichi has been using spot extensively for initial entries into places like the operations, the fuel handling rooms, the fuel handling control rooms. This has been mostly dose rate surveys and visual surveys. And you have areas where the floors have collapsed. They've got doors that have been closed that the robot has opened because the robot can autonomously open a door instead of the operator having to, you know, manually go through and position a gripper or turn a knob. 00:43:43:03 - 00:44:26:05 Unknown The robot basically. Oh, that door right there, where's the handle? And then the operator simply says, that's the handle, open the door, and then the robot moves through that environment, Now enters another location, the Sellafield facility. You know, we were I think a lot of us here are familiar with the condition of that particular facility. One of the things that I really enjoyed about the idea of this assisted manipulation is that when the robots are operating inside of the hot cell and cleaning up soft waste, this is something that a human being can do because the ambient dose rates are low enough that an operator could go in or a worker can go in and 00:44:26:05 - 00:44:55:05 Unknown have a controlled amount of radiation they would receive. But they have to wear positive breathing suits because the contamination level is so high every time one of those operators goes in that entire suit, all the hoses, all the gloves, all of that is disposed of during every single entry. We filled up with spot 18 plutonium, contaminated drums with soft waste and various debris with it, these hot cells. 00:44:55:07 - 00:45:22:09 Unknown The thing is, they told us afterwards that for a human being to do this, that would have created another 14 drums of waste just from all the PPE that would have been utilized for the process. So not only did this go faster by an enormous factor, you have the ability to work in shifts without having to deal with donning and removing protective equipment. 00:45:22:11 - 00:45:49:03 Unknown This also prevent more waste from existing in the first place because that PPE did not need to be able to be used. While I'm not able to name the site, this is a location that we're very proud to be operating in where Spot was able to conduct a down posting operation or hundreds of different radiation readings were taken both by spot and other unmanned ground vehicles within the facility. 00:45:49:05 - 00:46:13:06 Unknown And then in the process, a lot of those UGVs got stuck with their wheels, having difficulty navigating in challenging environments and it was spot to the rescue. That spot has actually now received the highest total accumulated dose of any of our robots to our knowledge of 20 Sieverts. So 2000 REM. So this has been something that you can see myself in the corner. 00:46:13:08 - 00:46:49:06 Unknown This was a bucket list item for me in terms of being able to reach these sorts of environments remotely and genuinely make a difference here. So I wanted to make sure that we had some time to kind of talk about how this was being done in the real world. Hopefully this has been a beneficial presentation for everyone. I really appreciate your time and we're happy to take any questions and I'll be present for the roundtable at the end of the day as well for additional conversation. 00:46:49:08 - 00:46:55:08 Unknown And thank you for them. Thank you for done. I think we want to hold the question at the end.