# Pairing Docked Drones with Boston Dynamics' Spot for Security & Inspections at Texas Instruments - Event: NestGen ’26 - Published: October 2, 2026 - Length: 34:08 - Watch: https://nestgen.org/library/watch/t_CQa4zxo_M/ > The transcript of a recorded NestGen session, for reading or for use with an AI tool. It comes from the video's > captions; automatic captions can mishear names and terms. ## About this session Ground robots and docked drones usually arrive as separate programs, each with its own operators, workflows, and data, even when they are covering the same site for the same reasons. Teams end up running two autonomy stacks side by side instead of one. In this NestGen session, Texas Instruments and Boston Dynamics will share how docked drones work alongside Spot across enterprise security and inspection routes, and what operations teams should think through when combining aerial and ground autonomy on a single site. ## Transcript **[0:08]** All right, it's showtime. Hello. Hello everyone. Welcome back to another very exciting session here at Nesgen. Uh do let us know in the chat where you're joining us from. Uh for most of today, we've looked at how autonomous drones are shaping critical operations. This session adds a very new frontier. It takes a look at how autonomous drone and a ground robot can work in the same site together handling security and inspections that would otherwise put people in hazardous areas. So, please join me in welcoming **[1:10]** our two guests for this session. First, I would like to introduce Scott. Scott uh leads the is a robotics engineer at Texas Instruments and leads the program inside TI integrating autonomous drones and ground robotics into their site security operations. He brings a very fresh operator's perspective including the practical realities of scaling a program like this across multiple sites. And joining him today is Andrew. Andrew is the AI product manager on the Boston Spot. um and orbits team and of course uh as you all know spot is one of the first and the most recognized quadriped robots in the world. Andrew leads the AI strategy behind it and the intelligence that turns what spot observes into insights team can act on. So together both Scott and Andrew will walk us through how Texas Instruments deploys dog drones alongside Spot while drone is providing the aerial coverage across the site and spot performing **[2:15]** close-range thermal and acoustic in inspections inside the facilities. So before I pass it over to them to take a uh I mean to conduct the session, I have a few quick housekeeping notes. So please uh type in any questions in the Q&A tab during the session and we will get to them at the end. The session is also being recorded and will be published afterwards. Also if you do have any favorite moments please uh do share them on LinkedIn take a screenshot and use the hashtag neesjhat a few lucky people who might have a surprise in store. So with that, a very warm welcome to both of you and uh over to you uh Andrew. >> Yeah, thank you for having me. Um Scott, I'll let you kind of kick things off here um while I'm trying to share my screen. Um so let me do that. >> Very excited, Scott. Looking forward to this. It's very different from all the different all the sessions that we've had today which were **[3:16]** very focused on drones primarily and this is a different kind of intersection that we've seen today. >> Uh you're on mute Scott if you Yeah, >> that would probably be a little bit better. Sorry about that. So again um my name is Scott Carl. I'm working at Texas Instruments. Um, I'm on the smart manufacturing and automation team concentrating on robotics. Um, and really, you know, this talk is about drones and leg robots, but really what it's about is what they make possible. A facility that watches itself continuously and brings people in only when a decision is actually needed. I'll show what's proven, what's live, and where we're taking it. And I'll be clear throughout um, just to kind of try to show what's real today and what the vision actually is. And joined with Andrew. Um so we'll we'll transfer over into um Andrew will take the stage here in just a bit here. So the question that we're answering is what if the **[4:17]** facility could wash itself? Uh we're talking about autonomous site oversight, continuous verified awareness of every critical asset around the clock. So something that we're trying to do today with humans, but it's cost-effective. um it exposes people into dangerous situations and um you know we have the the variability when you come to a human actually doing the patrol. So today almost every industrial site understands that its own condition through the snapshot right. So you have a guards patrol an operator round an inspection that's happens on a calendar um but between these snapshots uh you're missing a lot of what actually happens on at your site. So the site sees itself, understands what it sees and brings in people only for those decisions. That's where everything that we're talking about is a step towards. All right. [clears throat] So the cost of the snapshot. Um so we have the safety exposure for people having to enter these unsafe areas. So the high voltage areas, hazardous zones um just to prove that nothing was actually wrong. Um we have **[5:20]** unplanned downtime. So we're not able to inspect things as frequently as you would like them to be or and in a continuous manner which means that you're going to be missing signals. Those missed signals lead up to unplanned downtime that costs money. Uh we have a lot of wasted attention. So specifically here at at TI we're u we have in the range of you know hundreds of thousands to to millions of assets depending on what we're looking at which translates down into thousands of alarms that come in every single day. Those alarms cannot all be rationalized which means we have to have some type of automation. Um there's a lot of false uh false alarms that come in and hours of footage that we consume. So we need a way to automate this entire process and go ahead next slide. So what that kind of question is a a quick question for for all online is what does that cost u your site the most today? Are we missed early warnings? We have some false alarms, people in those hazards zones, **[6:22]** all of the above. There's a lot of uh a lot of areas that we are trying to close um the gap in in the effort that we are moving towards automation with robots. Um these are some of the categories that we are addressing. Next slide, Andrew. Thank you. >> The poll is live for anybody who like to >> Sorry, Scott. I was just saying that the poll is live for anybody who would like to go ahead. Go ahead. I think we can. >> So today we're going through a shift. Um today we have snapshots. Uh what we're looking for is an actual oversight of our entire site. So what we we have uh typically in you know your typical enterprise situation, you'll have periodic rounds and patrols. um that's going into a continuous roundthe-clock awareness. We had reactive alarms. We're moving into predictive early warning and maintenance. We had raw footage and **[7:22]** telemetry. Um now we're moving towards the verified and prioritized events. People were the data watchers. Now people are the decision makers. It was a sightby-sight cadence and now it's one network that learns based off of all of the work that has been done from the other sites. So this is a compounding effort that we are are going through here. The next uh layer is the perception layer. So how are we actually collecting these new signals that previously we didn't? Uh to start with the the answer or the question that we're trying to avoid is how do we not have sensors deployed on every single asset because that's cost prohibitive. The answer to that is these two uh air and ground combined solution. For a deep view, we have the Boston dynamic spot robot that's going to have the in our version of spot, we have the spot cam, the sorama L642, and the LAR sensor in the back that's going to allow us to take thermal, **[8:24]** acoustic, and LAR inspections throughout our facilities and to take that deep scan to really confirm what is actually happening with these assets. We're also pairing that with the drone. So, we're using the Sunflower Labs dock drone right now. That's giving us an aerial overview um patrolling our sites looking for any security incidents as well as uh inspection or inspections that we may be able to do outside uh such as on tanks or things that would be requiring ladder. They can both see the whole site and the detail that matters. From here, I'll refer to them generically. The approach doesn't depend on any one vendor is what I want to concentrate on. Okay. What's proven? So on the aerial dock drones, we've had a near-perfect launch and docking uh reliability. This has been uh extremely important because we have replaced a human with a a tool. What that means is that tool needs to be run reliably um and consistently. So having **[9:28]** that docked drone always landing in the the docked drone always landing in the dock is important. U that dock was the dock [clears throat] and the drone was able to be up in the air on day one. So as soon as we provided the utilities and infrastructure that was required that dock was able to be uh performing as it needed to and performing the function that we wanted it to perform. We started off at a site we've expanded to four. We're looking at globally. So, we have a use case that we knew was going to be uh something that was needed. We proved it out. We're expanding and now we're looking at other use cases. It translates out over to the the ground um layer with the the Spot robot because now with Spot, we've had thousands of autonomous inspection missions that we've run successfully. Uh we are operating 247 365 with Spot. He doesn't have a [clears throat] period of the day where he's not active. So yes, he has to charge uh but he is being used throughout the day for various tasks and he's got a **[10:29]** consistently high mission success rate a month after month. That doesn't mean 100%. And that's something that we have to be realistic about. We're not expecting perfection. We are expecting as close to perfection as we can get. Now the hardware is proven. The value now lies in what we can actually get uh from that intelligence. >> [clears throat] >> And so we have a a path to autonomous site oversight. Um the first piece is we have to automate you know how these robots are actually collecting our signals. The next piece is where the signals that we're collecting from our robotics connect with the rest of the site signals. So our SCADA systems and our other um our other site alertness um systems that we have in place um are now connected. Now what we're what step we're actually at is the contextualization piece. So now we have to refine uh the fuse data to tune to turn sorry to tune the alarms into verified events. So we go from just collecting the data to analyzing that data to **[11:34]** understanding that the data is actually true and accurate and representing what it's showing. After this step, we're going into the predictive nature of it. So now we know what trends and what flags to actually look for and to be on the uh on the alert for as they're showing up so that we can jump ahead of scheduling these maintenance events happening uh before any downtime or unplanned downtime occurrences happen. The last step which is the real vision of what we're actually looking to achieve. Uh we have to start with the foundation though. But the last step is the self-directing guide where the system is going to take its own action. Meaning that your drone is going to go and inspect your site on a regular bas basis and once it finds something we'll go and deploy spot along with it to get a further investigation or more insight of what's actually happening. people will go from being the ones collecting the inspection itself to just reviewing what's been collected and making the actual decisions. This is something that must be done in this age of autonomy that we are going **[12:38]** towards u where we have a resource constraint on just about every aspect. We are being told to do more with less. And so we have to be able to have a way to not get rid of um a way of collecting information, but just improve on what's being collected. So that takes us to the intelligence layer. Uh turning robots. >> Yeah. So go ahead. >> Yeah. Thanks, Scott. Yeah. So I'll take over from here. Um, so I'm going to talk a little bit about Boston Dynamics and how we've partnered with TI to kind of help build this intelligence layer. Um, so we'll talk a little bit about the uh robots that we build and how we um build that uh how we take the data that our robots actually perceive and collect and uh feed that into our software system orbit. um which is taking everything that the robots see and then visualizing that into um a **[13:38]** meaningful visual layer for customers like Texas Instruments. Um so humbly I believe that you know we build some of the best worldclass robots. We have robots like Spot um but we also have our humanoid um Atlas and Stretch. All of these robots feed into our central software system which we call orbit um which I will show you guys uh in a little bit here. So spot focusing specifically on spot today. Spot is essentially our facility um management uh robot. And so what spot is doing is patrolling factories, patrolling warehouses. And with our industrial um sensor package that we throw on top of it, we're able to uh address the concern that uh Scott laid out earlier, which is we can't sensorize our entire factory or our entire facility, but you can have one robot and throw a bunch of sensors on it and now you can actually monitor thousands of different assets or pieces of equipment along your factory. And so this is what **[14:39]** Spot is doing. It's going out and performing these missions every single day or to Scott's point almost every hour of the day in order to get this information and then visualize it and contextualize it for uh for the user. And so we do that by taking all of that data and all of the images and um uh data that Spot is actually collecting and we throw an AI layer on top of it. And so what you're able to do with that is you're able to ask very basic questions like what is the reading of the gauge? Is there a puddle on the floor? Is my door open? And we've uh layered in Google Gemini into our product so that we could get more advanced intelligence into our product. Um some really cool things with that means one out of the box we just have a more sophisticated AI layer. Um, but we're also able to push updates to our customers and update that model without having to have users actually wait for um, a software update from the entire software system. The other thing is we're able to take the data that we collect um, and understand **[15:42]** the feedback that we get from users saying, you know, this inspection isn't performing well and then we can refine those prompts and refine the model to actually address some of those low performing inspections. So, what this means is that every team at your factory or facility can actually use Orbit and can use Spot. Whether it's a safety and security team that's trying to monitor the facility and make sure that it's actually safe and secure, if it's an asset monitoring team that wants to actually use Spot to uh take a look at these pieces of equipment that aren't sensorized, or if you want to understand material or 5S flow across your facility. Um, and so you can imagine each of these have very different use cases and very different AI prompts. Is the door open? What is the reading of the gauge? Um, are my pallet stack too high? And we're able to kind of build these sophisticated models that are able to actually address these like more industrialized um, questions that we see across our customer base. So, something that's really cool is um because we're collecting all of this data and **[16:45]** we have Spot uh monitoring the facility, we can essentially build the industrialized version of Google Street View. And so what that lets users do is virtually walk their facility from anywhere in the world. They can look at the route that Spot actually took um and just play back the mission that that Spot ran. And that lets you catch things that maybe you didn't uh initially catch when you first deployed Spot. You can see now that there's a leak on the floor and that there's hoses all over the place and uh at some point someone had left the door open or used the shovel to prop the door open. And so you can kind of get this um virtual uh playback of your facility, but you can also scroll through time and see how your factory settings might have changed um you know time over time. When we throw AI on top of that, you can start to do really cool things like uh rather than having Spot go to a specific inspection point and take a picture when you have to scale that out to a thousand different assets and inspections. I'm sure **[17:47]** Scott can speak to this. It's very frustrating, can be very tedious. And so, one of the things that we've started to do with with Spot is say, "Hey, we know these assets are important and we know that these customers care about these things. let's just have Spot look for them and automatically flag whenever it sees it. And so this is adding an AI layer on top of kind of this visual site history intelligence that we've built. And rather than having to set Spot up to take these discrete inspection points, you can now just have Spot patrol your facility and say, "Hey, I detected a puddle over here. I detected a person over here." And eventually we can do things like, "This is a gauge. let's auto author an inspection of this gauge because you actually don't have an inspection um of this piece of equipment yet. And so what that looks like, and again kind of going off of Scott's playbook, we want to be very real about what we have today versus what we want to build in the future. And so everything I've presented to date has been everything that we can do today. Um but our kind of blue sky vision of this is kind of this world semantic **[18:48]** understanding of your factory and of your facility. And in order to do that, um, we kind of have to build upon all of the layers that we've introduced so far. And what that looks like in that blue sky vision is Spot walking around and basically building a digital twin of your factory. We can already pull in data from PLC systems. Um, so we can understand things that Spot isn't actually seeing and collecting today, but we can manifest that within orbit. And then we can overlay what Spot is seeing with that system and understand, hey, that's fire extinguisher 15 in my asset management system. Um, that's tied to again fire extinguisher 15. And you can have those systems all talk to each other. And that builds really cool systems where you can actually have an MCP layer that says, I had a alert for my PLC system. let me send Spot out there and actually um validate that that alert is real rather than sending someone out onto the floor and check um you know a piece of equipment that might be either during a time that no one's actually on your site or it could be a piece of **[19:50]** danger. >> Cool. And I'll hand it back off to Scott. Scott, you're you might be uh muted. Right. So, picking off of what uh Andrew was stating about where Boston is today and where they're building, uh this kind of shows where TI has been able to ingest the new data sources. So, we are perceiving this whole site. uh we have thousands of more sensing points than we used to as of 12 months ago. Um so we have a lot more to look at but now we have the piece of it where we are contextualizing what we're actually seeing to see what actually matters. Um this is the frustrating part. This is a very long uh enduring **[20:52]** task that is very manual. It's something that we have to go through because to get to this decide phase, you have to know what you're actually acting on, which requires all of your subject matter experts to put their knowledge written down on a piece of paper so that I can show my systems where to actually look to find that information. Where we're going is that ACT piece go to that next slide. So kind of the process that we're looking at we uh we have this continuous loop starts with the sensing goes to the understanding the verifying the act and then the learn. So this has been in place um and it's how we've been doing and operating long before automation came into the picture. So this is more of just adapting an existing system um with this new technology that we have available. uh every verified event is going to sharpen each asset's baseline so that tomorrow's events are better than today's that one line is probably the biggest gain that is coming out of uh using this new system. **[21:55]** So from data watchers to decision makers um you know today that's where we're at. We we're essentially just watching as the sensing is coming in and all of the it's giving us the reports and we're making sure that it looks okay. But where we want to be is we're not looking at the raw data itself. We're looking at the events that are coming in that are have been defined as worthwhile to look into for our teams so that they can use their uh their time optimally um and focus on the root cause and engineering side of things rather than um just going and collecting you know data in the field. Um this is the c the the part that I care most about today. skilled people are spending their time as data watchers walking rounds and confirming that nothing is wrong tomorrow. They're going to be the the decision makers giving them that power back. So if your team got its rounds and footage hours back, how would you use it? Where would you reinvest that time? Could you spend your time doing some more root cause **[22:57]** engineering to to figure out that you have a a problem that you didn't understand or preventive maintenance um so that you you're not always reacting on things? Could you improve your processes? Um, these are all every single one of these answers is a a statement that has come from uh some piece of my team that said, "If only we had less time that I was spending on inspecting raw data, I could make more engineering decisions, the strategic integration path. Um, so our data is the moat. And what I mean by that is the moat isn't the model, it's the context. The analytics engines, the vendor AI models, these new robots, these are all pieces of a puzzle. Uh but the real the real answer is the data itself, the golden layer data. Um sight specific baselines, your asset history and the fused air and ground context. Um what that's going to help us to develop is the KPIs and devel and deployment playbooks that we have been building **[23:58]** and are putting into place. um which is defining what success looks like for TI. But that what I'm trying to to kind of insinuate here is there's a lot of products that are available. There's also a lot of internal development that we do built bot or both. The engine's a choice. The data is what the advantage is. So where we started, we start off at a site um about a year and a half ago. Andrew may uh remember back in our fab we had started off with the the spot robot and um we didn't understand where we really were. It took us a good six months just to understand how um how many things and opportunities we had left in front of us. We've now expanded out to four sites and we're compounding the the data uh the knowledge that we are learning from each of these sites and helping us to reduce that time further and further as we expand to other sites. every new site uh deploys faster and it makes the whole network smarter. **[25:00]** So some of the things that we learned um vendor metrics aren't business value. So when I first started at this uh in this role um it was very much so I want cool robots and I get to work with some of the the greatest companies in the world and that are developing some awesome things. So yes, I don't even know what I'm trying to accomplish with this. um I you know they come up with some KPIs and we meet them we're all happy except for the fact that TI um and any enterprise is going to have stakeholders that have to be aligned with and you know they want to see where the dollars are are spent and how much impact it's coming with. So proof beats the projection. Um the drones only scale because the pilot evidence was was very understandable. Um it's undeniable. uh every phase earns the next with the actual measured actual return. Integration is the long pole. Uh what I mean by that is the IT security, the legal ESH, government relations, your financing, your procurement. There's a lot of different teams that **[26:03]** we have to deal with internally in an enterprise situation and they all have to give approvals and they have very different question sets they have to go through. Those partners we engage from day one. It's a lot easier now that I have the experience of gone through this before, but that is one of the the critical thing to to keep in mind. Um, TI owned KPIs for the intelligence layers. So, these are a couple of the things that we are are critically looking at. So, we have a few people in houses zones. You know, we have uh more uptime and less downtime from uh earlier detection. The alarms themselves cannot just be alarms. They have to be real verified events, near zero false alarms. U and then the response hours for redeploying. Um our manual response time is being driven down because of the engineering work that we're able to do. So we do a little bit more work on the on the front end and saves us a lot on the on the back end. So from a from robots that patrol a facility to the site that watches itself, prove **[27:08]** the hardware, we're done. We own the data. We're live and refining it right now. And next, we need to elevate the people. So, thank you all for your time and thank you Andrew for for the help. >> Thanks, Scott. Andrew, this was wonderful. Andrew, were you saying something? Go ahead. >> Yeah, thank you all. Um, this was great. Happy to answer any questions that you all might have. >> Yes, I do have a few questions, so I will take them one by one. Um so uh I think one of the question that has come up quite frequently is that um um how do we sync uh the drone and the spot like for example if spot is uh [snorts] during its petrol is checking for an open door can it trigger a drone to go ahead and fly over and find a wider view. A lot of these questions are connected and then there is another question around so first let's address this like is that **[28:12]** integration possible today or maybe something that we're looking forward to. >> Yeah absolutely I can tackle maybe some of this and Scott feel free to chime in wherever you feel appropriate. Um you know we've built orbit as kind of a two-way facing product and um one thing that we keep in mind is we call it kind of a data highway. Um, and so what that means is we want Orbit to, yes, be a visual layer that you can go to and be your one-stop shop for everything that you want to do. We also recognize that's probably not always going to be the case and that there's tons of systems that you use to monitor your factory. Um, and so what we build Orbit to do is pull data out of Spot and let you push it to wherever it is that you care about, but also pull data in from any of your other systems that you might have. So in a world where you want spot to trigger your drone to go out and look at something, we have APIs that are built that let you take what Spot has seen and push that to your system or vice versa. Take something that your drone has seen or that your PLC system has seen and **[29:14]** then have Spot actually go to that location and perform an action. So again, it's kind of like this uh birectional highway that we um that we've tried to build orbit to be. >> Right. That adds to the next question for you Scott probably. U how do you provide so much of the context of how these enterprise systems are running? I think you already had a slide covered on that uh to these new systems like drones and dogs for AI models to even understand what is required and what is going on for them to become self-arning. So is there a way that you've already figured that out on how to feed context to these systems? Um, yeah, go ahead if you have something to share. >> Yeah. >> So, that's probably the most important piece of what I'm what I'm going to in generalize call automation with robotics. Your data pipeline has to be understood. These systems, they don't know, right? Um, you're going to have a drone, it's going to be inspecting things. It doesn't know what it's **[30:16]** looking at. So, to teach it what it's looking at, you have to integrate it into your systems that are already live. And if it's not um capable, right, if you don't have that, then we have to build it. So that's what we've learned from this experience. You know, there's a lot of standards that we have to to enact and make sure that are put into place so that we can transmit the data between a drone and between spot. Um you know, Andrew is right. It's it's a data highway. You know, there's a lot of information that comes in and a lot of information that goes out of orbit. Um it's really about how your you know the end user wants to make use of it and what you want to accomplish. Um right now there is not a there's not a full develop just you know press the button and it's released. There is a custom integrations and developments that's going to be happening no matter what you're trying to do. So >> right >> a lot of fun >> I think. Yeah, that's the most important part about getting the right context about uh what whatever the uh robots are seeing at the end of the day. And uh **[31:21]** there's an interesting question around distribution of responsibility. So what do you know when to send a spot for a check in or to send a drone? Is it just primarily indoor versus outdoor or are there more nuances to it? No, there's definitely some more nuances to it. I mean, it matters about what type of uh payload you're trying to get the inspection from. You know, if it's a thermal camera versus a LAR sensor versus, you know, your acoustic sensor. Um, so that's going to be the first thing that you have to kind of qualify. And then the next one is how close do you of an inspection do you actually want to get? There's some locations that spot ar is not going to be able to get to and vice versa for the drones. So there is a right solution and a right fit for every inspection. And I think kind of creating that deployment technique of I need that's a that's one thing that we're working today on is robot as a service from internal customers. If I have an inspection that I need, how do I make a request to go get a robot to go and do **[32:24]** it? because you can't assume that everybody that is going to be working, you know, with robots or working in an enterprise is going to know what the robot is or how it actually works. So, we're trying to standardize the way that these uh requests are coming in as well to to make that a little bit easier, >> right? I think that gives a very interesting perspective beyond just because I think one questions come in is it is spot primarily used for indoor applications. I think that kind of addresses that. So I think uh we have most of the questions covered and we're also running out of time. Uh so once again thank you both Scott and Andrew for joining us today at Nesgjen. It was wonderful to hear your thoughts especially I think the mental model that you shared about you know right now the stage that we are in and to going to a complete self-arning system and a site. I think that's the vision that we all have and uh it is it is uh going to be coming **[33:24]** true especially with all the genai capabilities uh that is coming in and uh all the the physical AI revolution that's right now in place. So thanks once again if any of you would like to share any last minute closing thoughts uh we would like wrap the session up. Yeah, thank you again for having me. Happy to connect with any for all right. Thank you so much and yeah, look forward to catching up with you both soon. Uh have a great rest of the day and uh to your lovely audience, thank you for tuning in. Bye-bye. >> Thank you all. >> Bye.