>> Welcome back everybody. Jeff Frick here with theCUBE. We're in Mountain View, CA at Knightscope, a really interesting company that's making autonomous vehicles. They're not cars, they're robots, and they're for security. And they're deployed and they're in use, I think it's 15 states or 14 states all over the country, just closed a huge round of funding. A lot of great momentum, and we're really excited to be rejoined by Cube alumni Mercedes Soria, she is the VP Software Engineer. Mercedes, great to see you again. >> Thank you for having me. >> Absolutely, well thanks. We had you at the studio last time so thanks for having us over here where all the action's happening. >> Yeah, you're welcome to come anytime. >> So for the people that missed the first interview, just give them a quick overview of what Knightscope's all about. >> We build autonomous security robots. Those are machines, they are running around autonomously, collecting video, collecting signals like thermal signals, the signal from your phone, and collecting a bunch of information that then is transformed into a webpage that a customer can see. So they get alerts to anything that is out of the ordinary. >> So, the application is often in a mall, or in a parking lot or some of these types of places where it's really an ongoing patrol that the robot does. >> Yeah typically what you want a robot to do is the monotonous work. So, at a mall, the security guard walks around all day long, and most of the time nothing happens. So when something happens, only then you want to be notified. Otherwise, it's just a guard that walks around. So that's the job that the autonomous robots do. >> And is there somebody monitoring all of the sensors and stuff all the time, or is it more of an alert system, or is it kind of all over the map? >> It depends on our client. For example, we always monitor all the robots. We get alert systems set up so if anything happens to a robot, we will be notified. But on top of that, some of our customers like to see their video 24 hours a day to see what's going on at their facilities. Some other clients only want the security presence, so they don't look at video. It really just depends on what the client wants. >> What's the big difference by having a Knightscope robot versus just security cameras that are just pointing and on all the time. >> If you have a steady camera, by default it just doesn't move, so you can't cover that much space. You're only going to see that one box that the security camera is covering. With an autonomous machine, you can take it wherever the crime is happening; you see something is wrong, you move the machine over and you take a closer look. So it does a lot more than just this one square that you can look at all the time, you can see everything around the machine, it's 360 degree video that is running 24-7. >> And how does it impact the way that the security people do their job? Let's stick with the mall example. When you've introduced a Knightscope robot, what's the right word, is robot the right word, is it robot? >> It's a robot, yes. >> When you've introduces a Knightscope robot into a mall situation, how does that change the way that people do their jobs? >> The robot will do about 70 or so percent of what a security guard does. But now the guard, instead of having to go and walk around the mall all day long, they get to do a more interesting job So now they're more interested in robots, they know technology, they get to know how to deal with a machine, how to interact with people. Those are things of a higher level. If the machine does all of the monotonous and boring work that the guard does. So at the end of the day, that guard does something that is a lot higher level than what they were doing before. >> Do customers typically have fewer guards, the guards just doing more higher-value, how does it impact their whole security system once they bring in a Knightscope robot? >> It could be one of two things. There are some places that our customers have zero, zero, zero patrolling, so they have nothing. So in that case, if the robot comes in, now you have security that you didn't have before. Some of our other clients, they decide that one of the robots is going to do the job of maybe three people, but those three people now are doing administrator work. So their work is to become of a higher level, so it depends on the client a lot. >> We've got a bunch of the robots behind us here in the shop, I'm sure we'll have them in the intro packet, the different ones. You've got four different ones. First off let's do some of the basics: What are some of the sensors, what are some of the inputs that they are collecting, and why do you have four different ones? >> A lot of why we have four ones is because we wanted to give the customer security regardless of their environment. The first one is the K3, that's an indoor machine; it's a smaller size, it weighs about 340 pounds. >> Small one and you say it weighs 340 pounds? >> Yes, that's the small one. >> No little kids are running up and tipping that one over I don't think. >> No, they're not. People have tried, but not yet. That is for indoors. It has all of the sensors that our other machines have, they all have a thermal detection. They have their regular cameras. They have inertia measuring units. They have lighters, which is what allows you to tell that there is something in your way, that you have to get out of the way. We have ASD, which is Automated Signal Detections. Your phone emits a signal when it's trying to connect to Wifi. We can detect all of those phone signals, and then we can log that into the server. All of the machines have that, it's just how they are using in a different environment. Indoor for the K3, outdoor for the K5, we have the K1, which is a static unit. That's typically going to be put in the door of the places we're going to monitor. And then we have the K7, which is the largest unit that we have so far, and it's going to places that a machine that's smaller cannot typically transverse. >> So that's the one that looks like a little Jeep back here. >> Yeah, in a wind farm, in a solar farm, these machines don't do very well, and that's when these machines go in. >> So definitely for outside >> Outside-- >> On the road, in the dirt >> Large companies-- >> It doesn't have to be in a parking lot, in a paved environment. >> Gravel, any different type of environment. >> What is the experience, why do these things work? Is it just because you have more coverage because you've got a robot that's going places you don't have enough guards? Is it the intimidation that someone is watching me now, you're bringing a camera into the parking lot where maybe it was kind of hidden behind a little wall? Is there a two-way interaction, do people talk to these things and expect them to talk back? Where do you see the most effective, why are these things effective? >> The reason why they're more effective can be summarized in one point: Security guards don't like to do their job. There is a 300% turnover rate in the security industry, people don't normally know that. So you're getting a new team every four months. People don't like their job, it's a job that is very monotonous, very boring. We're putting a robot there to do the same job, so you can free people to do something of a higher level. And that's the main reason why they work. >> I wonder if you can speak a little to where you're using machine learning and some of the deeper technology beyond simply putting a camera on a mobile platform. >> Some of our customers, for example, at night at the malls or corporate campuses, there isn't supposed to be anyone there at night. So one of the big applications that we have is we have an image, and from that image we can train our algorithms to detect people in that image, to detect faces in that image. All of that is done by machine learning. Because we have five years of data of images and people and we train our algorithms to say, this is a pole, or this is a person, this is a tree, this is a person. So we get to detect people in a really high accuracy level, about 80%. We also do the same thing with license plates. We train our algorithms to detect that there's a license plate in an image, you detect that there's a car first, then you detect that there's a license plate, and from there you detect all of the character in that license plate. And all of that uses machine learning, even to differentiate that there is a number one opposed to a letter L. All of that had to be trained as the technologies that we're using. For the future, we're going to use prediction algorithms in the way that, now we have data of what happens around the location where the machine is deployed to. We're going to be able to say, "Okay, this area has a lot of crime that happens "on a daily basis or however often, "you probably should go patrol over in that area." That is what we will do in the future. >> The other interesting thing is you don't sell these, these are not for sale as like, going to buy a car, you actually provide it as a service. So a very different business model, very much in line with what we see more and more, it's a service, people basically rent the robot with the monitoring service? Is that accurate, or are there lots of different flavors that they can buy? >> What we do is called machine as a service. To eliminate our customers having to pay a big amount of money at the beginning, they don't cover that cost, we do. But they pay us a monthly bill. Included in that monthly bill is the machine itself, all the parts, the monitor in there with the one on our end, all of the software upgrades, which we do every two weeks, and all of the hardware upgrades, which we do every six months to every year. All of that is included in that package. How the customers chooses to monitor their machines, that is up to them. We have agreements with two of the largest security guard companies: Securitas and Allied Universal, so they can do the monitoring for the customer if they don't have a security operation center. >> Clearly, you're operating in places where they already have security in place, they have systems, so do you integrate with the existing alert systems and the existing infrastructure they already have in place, do you guys just tie into that? I would imagine there's some industry APIs that you can feed into those systems or is it a completely independent monitoring that they have to do now? >> We did a little bit of the reverse of that, we built our system for it to be integrable. The way we wrote our code, a customer system developer can call our APIs and get the information from the machine that way, so all of that is finished so they can integrate with us opposed to us integrating the other, there's hundreds of systems out there. So if somebody wants to look at data from Knightscope that's already there. >> But you've got the open API into your data feed so they can feed whatever system. >> It of course is secure, you have to have keys and passwords and codes, and all of the information is encrypted So there's measurements that we've taken to make sure that the information is secure at all times. >> So you're a hardware company, you're a software company, you're a services company, you're doing AI. >> Woman: We're doing design too. >> And design, and autonomous vehicles. >> Yes. >> What did I miss? >> Production! We build. >> Production too? >> We build. Yes. >> That's right, I noticed this says on the bottom, reminds me of an Apple product-- >> It is designed and built in California. 85% of what you see in this machine is United States. >> Pretty amazing. So what's next? What's the next big challenge? I know the Seven is not released yet, is it just more form factors, is it different sensors? As you kind look forward from an engineering challenge, what is some of the next big hills that you guys want to take to move this thing along? >> The three next big hills that we have: Number one is getting the K7 out there and patrolling. Number two is concealed weapon detection, that has been requested by a lot of our customers. >> Concealed weapons detection? >> A lot of our customers are requesting that. And third, on the software side of things, the actual prediction of crime that could potentially happen. Those are the next three big goals for Knightscope. >> I would imagine that with the concealed weapons it's just more types of sensors that can see X-rays or whatever to get more visibility. >> Yes. >> The big V. Not necessarily visible light, but visibility from the machine. >> Some of those things have already been requested by our customers, because what we've build is actually a platform. We can add other sensors to the machine depending on the needs of a customer. For example, we have a customer who wanted the machine to be able to identify people, so they wouldn't have to swipe a card. Put the sensor inside, it's accessible, it's already there, you get your sensor and your information. What's the biggest surprise that you hear from customers after they've had one of these deployed for, I don't know what's a reasonable time, six months, so they're kind of used to it in their workflow, how does it really their world, what do they tell you? >> There are two things, number one is how quickly people like the machine, how quickly they go, "Oh, yeah, it's here and it's working." And then also how much crime has actually been eliminated. They thought, "Okay, maybe I have one break-in into "a car every week, well maybe it will just go to less." It goes down to zero. There's people who had lots of crime, and just by the machine being there, they get nothing, they get zero. So that was our surprise for them, and that was a surprise for us as well. That's how the effective the machine actually is. >> It's weird when I just drove up today, there was one right in the middle of the driveway as I was coming in. I was like, "Is it going to move, am I going to move?" It's very, much more intimidating than you might think. It's a presence, for sure. >> Yeah, and we have things like car backup detection, because the machine could be going down the street, and a car could be coming out of there, so we have to detect stuff like that so we don't get run over. All of those little things that we can think of, we have to do that a lot. >> Alright Mercedes, well thanks for inviting us over, it's fun to actually see the machines for real. >> Thank you so much for coming. >> She's Mercedes, I'm Jeff Frick, you're watching theCUBE, we're at Knightscope in Mountain View, California. (upbeat music)