Drover AI’s Alex Nessic, Co-Founder and Chief Business Officer, joins Ryan Chacon on the podcast to discuss the role of AI and computer vision within IoT. Alex begins by introducing himself and his company before talking about the limitations of GPS. He then describes how AI and computer vision works together to resolve the limitations. The conversation then turns more high-level with discussion around the future of computer vision and the possibility of moving into the consumer space.

Alex Nesic is a passionate evangelist for shared micromobility and LEVs in general, promoting their role in a sustainable urban transportation ecosystem. Alex is co-founder of Drover AI, pioneering the use of AI-powered computer vision for micromobility. Drover is his 3rd company in the micromobility space – Alex previously founded CLEVR Mobility and was an executive at Immotor.

Interested in connecting with Alex? Reach out on Linkedin!

About Drover AI

Drover AI is pioneering the use of AI-powered computer vision on IoT devices used in the micromobility industry. Their PathPilot tech is used by our customers to enhance their regulatory compliance while also helping them optimize operational efficiency. PathPilot uses a camera and Drover’s AI algorithm to detect in real-time whether a scooter is traveling on a street, sidewalk, or bike lane, enabling control of the vehicle’s speed in each area. At the end of rides, PathPilot also performs parking validation, helping improve the proper parking outcome cities desire and keeping the right-of-way unencumbered.

Key Questions and Topics from this Episode:

(01:21) Introduction to Alex and Drover AI

(05:11) Limitations of GPS

(10:02) How AI and Computer vision work together

(13:52) Future of computer vision and IoT

(16:40) Moving into the consumer space

(18:41) Roadblocks in the industry

---

Transcript:

– [Voice Over] You are listening to the IoT For All Media Network.

– [Ryan] Hello everyone, and welcome to another episode of the IoT For All Podcast, the number one publication and resource for the Internet of Things. I’m your host, Ryan Chacon. If you’re watching this on YouTube, please feel free to like this video and subscribe to our channel. As well as if you’re listening to this on a podcast directory, please feel free to subscribe to get the latest episodes as soon as they are out. All right, on today’s episode, we have Alex Nesic, the Co-Founder of Drover AI. They are a company that is pioneering the use of AI-powered computer vision on IoT devices used in the micromobility industry. We talk about AI and computer vision and their role in IoT. How does it work, how does all this improve safety across the board, what problems does it really solve, the future of computer vision and its role in IoT, as well as other challenges kind of seen across different spaces as it relates to micromobility and IoT, AI, computer vision, all that kind of good stuff. So a lot of value here. I think you’ll enjoy this episode a ton. But before we get into it, if any of you out there are looking to enter the fast-growing and profitable IoT market but don’t know where to start, check out our sponsor, Leverege. Leverege’s IoT solutions development platform, provides everything you need to create turnkey IoT products that you can white label and resell under your own brand. To learn more, go to iotchangeseverything.com, that’s iotchangeseverything.com. And without further ado, please enjoy this episode of the IoT For All Podcast. Welcome Alex to the IoT For All Podcast. Thanks for being here this week.

– [Alex] Thank you for having me.

– [Ryan] Yeah, it’s great to have you. Let’s go ahead and kick this off by having you give a quick introduction about yourself to our audience.

– [Alex] Sure, my name is Alex Nesic. I am a Co-Founder and Chief Business Officer at a company called Drover AI. And we are pioneers in the space of bringing computer vision, so cameras and edge-based artificial intelligence machine learning to IoT devices, specifically in the shared micromobility space. So think shared e-scooters and bikes, things like that.

– [Ryan] Fantastic. So yeah, why don’t you… Let’s elaborate a little bit more on what the company does. So take us through maybe a use case application of the technology in a setting that’s relatable for the audience.

– [Alex] Yeah, absolutely. So anything that’s deployed as a free floating asset, the way that these scooters and bikes are out in the field, requires an IoT module. And the primary purpose for that is to track and manage them. And the connectivity relies on GPS signal and a chip onboard, as well as cellular connection, making it a connected device, right? And so on the back end, the operator can track and manage where all of the devices are located, battery status, maintenance status, anything like that, a variety of other… And from a customer’s point, that’s how you act with the device through the app, right? That’s how you locate the scooter and the bike and rent it, you know, pay, all of that. It’s all managed through the IoT module. What Drover AI brings that that is really innovative is an additional layer of intelligence. Currently GPS, you know, has a lot of margin for error, especially in, in dense urban environments where a lot of these fleets are, are operated where tall buildings will distort the GPS. You know, and so your location capability is limited by what, you know, the, the, the accuracy of the GPS, which is, you know, purposely a fairly kind of reasonably priced off the shelf module. And so instead of, of just relying on GPS, what our system does is brings a camera to bear and edge based computing power to run our algorithms. And we provide what we call contextual location awareness, meaning that rather than relying on a precise GPS based location, we position ourselves more like a human who goes outside. And doesn’t say, what are my coordinates? You look around and you can see that the street is maybe 10 feet away, that you’re on the sidewalk, that there’s a bike lane there. So it’s in the context of a location. And so even if we happen to be in a, or the vehicle happens to be in a place that is devoid of GPS connectivity, or cellular connectivity, the fact that it’s all happening on the edge allows our IoT module called the Path Pilot to identify where it is in, in the context of its surrounding. So whether it’s on a street, sidewalk, or bike lane in which allows the operator to, to do a number of things. First, among them being regulatory compliance, it’s illegal to be on sidewalks with scooters. So our technology enables the control of the scooters speed or, you know, behavior and even surfacing audible notifications to the user to let them know that they’re in violation of, of a regulation.

– [Ryan] That’s cool.

– [Alex] We also help with, with parking, right? And these free floating fleets, the end of ride is a big concern because users sometimes leave them in the middle of the right of way, which is in violation of the Americans with Disabilities Act. And so it, it just helps our, our customers, the operators, manage these devices much more safely and responsibly and, and it helps satisfy the demands of the city.

– [Ryan] Fantastic. So one thing I wanna actually expand on, I, I do wanna dive into more of the computer vision and the role it plays in IoT, but you mentioned something about GPS. We talk a lot about GPS and different kind of technologies when it comes to tracking assets and, and things like that. But tell us, or, or give us a little bit more detail as to what, what problem, I guess, I guess what GPS doesn’t necessarily solve for itself and kind of your ability to kind of fit in with your technology and your offering in these densely populated areas in these cities. And tell us a little bit more about the limitations of GPS in that setting and kind of what you’ve been able to do to, to optimize and make it kind of plausible now.

– [Alex] Yeah, absolutely. So GPS, again, you know, off the shelf chip set will give you some accuracy anywhere between three feet, all the way up to 30 feet. And the issue there is that it doesn’t degrade gracefully when it is challenged. So if you open up your phone and you look at your little blue dot, google maps. If you find yourself in a, a city with really tall buildings, you’ll notice that that blue dot has kind of like this aura around it, or a halo that is large, that indicates it’s quite uncertain of its location. And then you kind of see it bouncing around until it gets a more and more and more precise, you know, idea of where it is actually located. But even in that scenario, you may be dealing with not just signal attenuation, but multi-path, which is basically when a signal is bouncing off of a building, a lot of buildings are glass and stuff. So you, you, your actual precise location or blue dot might not even be reflective of where you actually are. It’s interpreting where the signal is bouncing off of and putting you maybe across the street. So there’s all of these different challenges that GPS, you know, has a bunch of tricks up its sleeve, enhanced GPS techniques, which require earthbound base stations called RTK, realtime kinematics, which essentially is a distributed network of, of earthbound stations that would magnify or amplify a GPS signal for, for enhanced position. You could use cell phone triangulation, you could use dead reckoning and other sensors on board to try to basically stitch together the trajectory of, of something that’s moving and dead reckoning does that. But all of that kind of ultimately falls short in the most challenging areas. And, and what’s important there is, is that if you’re trying to distinguish, you know, GPS works well for these broad areas that you need to geofence. Maybe you want a geofence an entire city block or a, a university campus that doesn’t want scooters there. So GPS is good because it doesn’t matter if you’re 20 feet off, right? I mean, if, if you cross the boundary and, and the, the deactivation of the scooter only happens 20 feet later, it’s not an issue. But it is an issue if you’re trying to distinguish between when a user has left the street and entered the sidewalk, that is, you need like, you know, several, a couple inches accuracy. There to really play in that kind of finite space. And so what computer vision really does there is bring that, that camera and ability to contextually position itself and not rely on, on a GPS signal that might be pretty far off for all intensive purposes. And then the other part of GPS, right, that, that doesn’t scale necessarily rapidly is that let’s say you do achieve that kind of holy grail of accuracy of 10 centimeters or better in all environments, which is doubtful, but let’s just- Hypothetically say that you can, what then the challenge becomes like a really massive data management challenge, because you still don’t have the context of where you are. You may have pinned down your accurate GPS location, but you still don’t know where that puts you in, you know, where the street ends and the sidewalk begins. And so what needs to happen is it’s called a ground truth layer. You have to go out and collect that data and draw a whole bunch of lines on some type of, of backend Google maps or other to identify, you know, what that infrastructure is at those exact GPS coordinates. And so no city has that information about itself. You have to go out and collect it and create that database. You have to update it over time as cities evolve and sidewalks might fluctuate- Or other infrastructure fluctuates. And so it doesn’t scale easily. It’s a, it’s a massive data management nightmare across multiple areas and what computer vision and, and edge based AI onboard the IoT device can do is, you know, even if our device has dropped into a wholly new environment without having any ground truth, because it positions itself contextually, it doesn’t need that information. It can do so by, you know, leveraging a general database of millions of images across dozens of cities that may look reasonably similar to that. And so we can infer based on, off of that existing database where it finds itself and make that decision in real time.

– [Ryan] Gotcha. Fantastic. I’d love it If we could move out a little higher level here for a second. Just for our audience to get a sense of, when we’re talking about computer vision and the AI and how it all kind of works together. Tell us a little bit more about even outside of your general use cases And focus area, how that, how computer vision and AI kind of together are playing a role in IoT. How does that work? How is that improving safety kind of, as we’re talking about here about the different levels of accuracy it allows us to kind of get to, but just, just break that down a little bit further for us so we can kind of just understand it a bit better.

– [Alex] Yeah. I think it’s really, you know, cutting edge technology here in, in these kind of creation of, of computer vision based networks that are connected, right? I mean, you can see a very kind of relevant application is the RingCentral network, for example.

– And, and these are basically static sensors that are leveraging these cameras on doorbells or other security systems that are deployed, and they create this mesh network essentially where you, you can tap into them and through motion detection or other kind of intelligent sensors and sensor fusion onboard. Those have a really accurate insight into what might be happening in our neighborhood. You know, certainly traffic management and other deployed sensors already exist on, on lamp posts and, and stop lights in, in cities. Well, I think what’s unique about, you know, what we’re doing is, is deploying these on kind of mobile assets that move through cities very differently, obviously differently than any static sensor, but differently than cars do that might already be mapping a certain environment. And so cameras really bring that, that additional layer of context that without it, you, you wouldn’t really necessarily have that, that understanding or, or ability to, to take action. So an example would be, you know, Google maps leverages data from their vehicles moving through cities. But the average age of that information is around 18 to 24 months old. And so how do you plug in, you know, maybe more recent information and, and so we can maybe leverage our distributed assets moving through cities, just on consumer rides to say, Hey, at this timestamp and location, we have a data set. You might be able to, to tap into that’s more recent than what, what has been collected 18 months ago. So there’s, there’s a ton of, of really neat, connected camera networks that are being built out and, and that could be leveraged for different uses.

– [Ryan] Yeah. That’s super interesting to kind of think about just how we can utilize those distributed assets to collect information to just benefit a lot of different applications from a data perspective, right? Like with like Waze, for instance, you drive around in Waze and when Waze first started, they would literally just use the, the, the GPS location of the cars that had the early Waze app and say, okay, if they’re driving here, there must be a road here. And then eventually map software and stuff got integrated in and things like that. But being able to use these scooters and other types of distributed assets to collect information that we’re not able to collect as accurately is super fascinating.

– [Alex] Yeah. Well, and actually to, to expand on that. Your, the, the feed is delayed on the video. To expand on that, you can layer in a bunch of different things like object detection, right? Let’s say you want to do some infrastructure surveying. What are the conditions of my sidewalks? You know, what’s the foot traffic like here, obviously taking privacy into consideration here where we can mask and blur individuals, but, but yeah, there’s a whole variety of different types of machine learning datasets that can be used on the data that we’re collecting to, to gather additional insights.

– [Ryan] Absolutely. Yeah. It, it it’s super interesting stuff, which I guess it expands into a really good kind of follow up question is what does the future of this kind of look like? Like let’s expand on where it is now and, and the possibilities to kind of the future of computer vision, its role in IoT, and kind of where you see it going outside of just these use case we’ve been talking about.

– [Alex] Yeah. And, you know, our core competency and, and our, our initial customers are the shared micro mobility. And, and it is specifically addressing mobility needs and regulatory compliance and safety. Overall, our goal is to really, you know, enhance the adoption of micro mobility for a more sustainable urban transportation future that moves away from cars. And, and so that’s really where any further pursuits of, of monetization, of, of data that we’re collecting would ideally be used to subsidize the growth of, of micromobility, either through additional infrastructure that can be built out. Bike lanes and networks that can accommodate more of these types of, of vehicles. But yeah, you alluded to, we’ve already touched on some of them. Where the camera and, and, and IoT connected devices can drive insights related to urban trans or transportation planning, urban planning, you know. We can add other sensors on board, air quality sensors for example, that, that would be relatively cheap to, to incorporate into these distributed assets so that you’re collecting air quality samples from different non-static sensors, moving through a city. That’s always been fascinating. But if you, if you also want to go and, and understand where certain construction is happening, that might not be reflected. On, on any other city dashboard, right? We can start recognizing where, you know, cones are deployed and, and other traffic deviation patterns are, are occurring to inform, you know, mapping, right? And, and, and flow through, through a city. We can inform curb management decisions, right? If, if we can identify, you know, that, that a curb space is being used improperly, or, you know, again, this involves some public, private partnership, right? I mean, but we can also do the same thing that, that bus rapid transit lanes are, are doing where they’re leveraging cameras onboard buses. To automatically ticket violators, people that are, that are using the bus rapid transit lane when they’re not supposed to, we can do that for bike lanes, right? I mean, so if, as a city you’re allocating resources to build out a bike network, but people are using the bike lane as a delivery zone. That’s not effective. It’s not safe. It pushes people out into the street. We could leverage our vehicle detection as well as some form of license plate recognition and, and be able to send citations or at the very least warnings to people that repeatedly do that.

– [Ryan] Do you think it’s ever something that you can move into more of the consumer space as far as cars obviously have cameras on them, whether it’s dash cams. Or just cars, cameras within the vehicles, is there ever a stage in which that can be leveraged from a, from, for the data kind of in the overall benefit, whether it’s safety or just kind of general coordination of things. And I, I just don’t know if there’s necessarily like maybe some specific parameters on why that’s not maybe either legal or allowed, and that’s why it’s more public type opportunities for this data to be collected just outta curiosity.

– [Alex] Yeah. No, I think you bring up a good point. I think we’re, we’re all still evaluating opportunities in this space. You know, a company that comes to mind when you talk about that, which basically has adopted what, what you’ve described as a business model is, is Nexar where they leverage consumer dash cams that, that are deployed in consumer vehicles. And they subsidize the cost of those dash cams by having access to, or retaining access to the data, to be able to create these dashboards and, you know, understand different things and provide insights to cities or other customers. So I think there’s, there’s always been this fascination with how to monetize or what the value is of, of data that comes off of vehicles, all kinds, right? There’s hundreds of sensors on every consumer vehicle that’s put out there and companies like Autonomo, and Wejo are, you know, currently valued at, at very, you know, high valuations based on the premise that this data that, that they’re ingesting in part, you know, through partnerships with OEMs has value and, and, and different insights can be derived from them. So I think we’re still very early days in terms of what exactly can be done with different types of data. Both just sensor data, as well as, as vision based data. So we’re, we’re really excited about different opportunities.

– [Ryan] Absolutely. Yeah. It’s, it seems like the possibilities are pretty endless, which is really cool to see. Last question before we wrap up here is just from your perspective of the market and kind of how you guys are viewing the IoT space, what are some of the other big challenges you’re seeing companies face, maybe companies that you’re speaking to directly, companies that you’re just kind of, kind of keeping an eye on and just generally speaking, running into potential roadblocks with IoT adoption, deployments, things like that.

– [Alex] Yeah, sure. So one of the biggest ones now that I think a lot of companies in the electronics sector are dealing with their supply chain. And logistics, right? Availability of components. There’s been a strain on, on that caused by the pandemic and, and the ripple effect that that’s had. So I think certainly with the advent of IoT, different IoT projects that have become wildly popular. I mean, another one that comes to mind is, is Helium, which is a, you know, a LoRaWAN, long range WiFi network effectively of hotspots that, that has, you know, put a, a strong demand on, on kind of compute modules that are used in, in IoT devices. So, so I think, you know, in general supply chain and logistics is a, is a big one. Another one specific to computer vision is really navigating the privacy issues, making sure that you are as a company and, and as a technology CCPA and GDPR compliant with, you know, redaction of images at the edge before they’re ever shared with anybody. So we, these are the kind of obstacles and challenges that we have to take into consideration as we try to scale our business. Yeah.

– [Ryan] Absolutely. Yeah. It’s some super interesting, just kind of, when you go from supply chain to chip shortages, just everything going on in, in the space, how it’s affecting IoT, I mean, it’s affecting tons of different industries, but just, just seeing how, how that’s happening. It’s been a really interesting conversation with different guests have had on the podcast just to understand how they’re dealing with it and what they’re seeing from, from their own perspective since each area of IoT is feeling it, but maybe feeling it in a little bit different way. For sure. Last thing before let you go here is for audience out there who wants to learn more about what it is that you’re doing, the technology use cases kind of, maybe it applies to things that they’re working on, and they’d love to kind of expand further and touch base. What’s the best way to do that, to follow up and kind of stay in touch.

– [Alex] Sure. Our website is www drover.ai. It’s a good place to learn a little bit about what we do. I’m also on LinkedIn, Alex Nesic. You can follow me on Twitter @Alexnesic and the company is also on Twitter @droverAI.

– [Ryan] Fantastic. Anything new, exciting kind of coming out of Drover that we should be on look out for in the coming months. And just, just anything you can give us a sneak peek on.

– [Alex] Yeah. We’re working with more and more new customers. We’re deploying in larger scale in Europe.

– Cool, cool

– Largely with Voy, one of our biggest customers there. So we’ll be spending some time supporting the launch of multiple hundreds and thousands of, of path pilots out there.

– [Ryan] That’s awesome

– [Alex] And then we’re also, you know, kind of moving to our next generation product, which is gonna be integrated directly into vehicles, not sold as a retrofit add-on module. So that’s kind of the next step in our, in our transition is to insert ourselves further up in the supply chain and already be kind of a baked in option in, in the vehicle itself.

– [Ryan] Now let me ask what’s the value and benefit there, as opposed to being able to kind of do it as an add on. Is it just the deeper kind of relationship with the company and the product, or is there obviously feature and, and data benefits from that?

– [Alex] No, it’s primarily costs, right? I mean, because right now. Our out on module is, has some redundancy with what already exists. Every one of these vehicles already has an IoT module that has GPS and cellular connection. And so we also have to have that on our IoT module. So now you have two sets of GPS fees, two sets of cellular modems, et cetera. And so nobody wants to be paying twice for the same thing. So I think being able to, to integrate and at, at the edge with those sensors that already exist and effectively just leverage those onboard systems will be a much more cost effective way to introduce this. But, you know, a secondary module was, was really a, a decision that we took to, to be able to go to market as quickly as possible and prove out the technology at scale ‘ fully knowing that we would eventually be integrating into somebody else’s vehicle design. But if you, if you start off on that path, you’re at least 18 months away from any type of market deployment. So our, our choice was to say, Hey, here’s a device you can slap on any scooter in the field and we’ll work with you on that. And we got to market a lot faster by. By choosing that path.

– [Ryan] Makes total sense. Fantastic. Well, well, Alex, this has been a great conversation. Thanks so much for taking the time, very excited to kind of stay, stay in touch and, and see all the different things you have coming out and kinda the evolution of Computer Vision and, and how things are evolving on your end. So I would love to have you back. I’d love to kinda get you involved in some of our other series as well, that kind of align with what you have going on, but thanks again so much for your time. Really appreciate it. I think our audience is gonna get a ton of value outta this.

– [Alex] Awesome, Ryan, thank you so much for having me. It was a pleasure speaking with you and I look forward to any future conversations we might have.

– [Ryan] Sounds great. All right. Take care. All right, everyone. Thanks again for watching that episode of the IoT for All Podcast. If you enjoyed the episode, please click the thumbs up button, subscribe to our channel and be sure to hit the bell notifications so you get the latest episodes as soon as they become available. Other than that, thanks again for watching and we’ll see you next time.