The Knightscope Machine | William "Bill" Santana Li

Okay, so what's on the machine? So maybe I'll go over here. 360-degree daytime, nighttime video, so we've got all the way around HD live streaming and recorded video. We've got hyperlocal conditions for CO2, temperature, pressure, and humidity. We can run 300 license plates a minute and run it against a database, and if we're working with law enforcement for that particular We can run it against the NCIC database and tell you if the car is stolen, stolen plate, or there's a felon tied to that plate. We can also do people detection. So we have one client that's a large mall here, open air between 10 PM and 6 in the morning. There should be no one wandering around there. If you see a person, raise your hand. All right. We're also able to establish a 2-way call. So if you press this button here, You set up a 2-way intercom call, so similar to maybe if you're at your university or like a college campus or a big parking lot, there's big Code Blue, I think they're called, emergency phones. So this is like the mobile version of that. And then you're able to also broadcast your voice. Let's see if this machine has it. This machine doesn't have it. The next batch of machines has a thermal camera, so we can do a thermal image. So obviously to detect a fire. Also, if you're going by the same pipe every day, why is the pipe going up 1 degree every week? That gets interesting. Or back to the people detection. So the camera's detected that it's a person, but if I'd done a thermal scan, it says 98 degrees, now I feel a lot more confident that it's actually a person. The other thing we've got is a rogue router detector. So I didn't know this, but at data centers, they've got actual humans walking around with a rogue router detector, making sure that no one's literally at the front doorstep trying to do a cyberattack. So there's 172 routers in the building. Why this morning are there 175? All right, so we can automate all that good stuff. OK, so how does the machine autonomously move around? This is— so someone joysticked it in here so we can control it, but the way these operate out in the field 24/7, I guess probably first comment, it's really easy to do a demo, it's really easy to build a prototype, but you want machines running out in the field outdoors 24/7, 365, in the rain, with cats, people, kids, dogs, and vehicles and trucks and all sorts of— that's a whole other level of engineering. So no software node crashes, no electrical issues, no water ingestion issues, no lighting issues. I mean, it's a really tall technology stack. Besides just the machine, then there's all the data that has to somehow get out of it. And then you've got to process all that data with the front end and back end and everything in between. So what we're doing is slightly complicated, but that's why it's also exciting. So how does the machine understand its environment? It's analogous to what maybe you've seen from Google on the self-driving car. So there's 19 lasers on the machine. There's 16 up here, plus or minus 15 degrees field of view, shoot out 100 meters, 360 degrees. So it's kind of like the laser pointer when you're a little kid, but every 25 milliseconds it's doing a scan of the environment. And so every time it shoots a beam and maybe hits the tip of that shoe, it's Saw something, so I'm gonna plot a pixel there. Pixel, pixel, pixel. And if I keep doing that, all of a sudden I can kind of see. So there's 16 up here, another one here, down here, and here that allow us to better understand the environment. Some ultrasound proximity sensors here so you're not kind of bumping into things if one of these doesn't capture something. Then we use the wheel encoders, so similar to the odometry on your vehicle, to see how far we've moved. An inertial measurement unit, so we know kind of degrees of motion and kind of acceleration, deceleration, turning, and that sort of stuff. And a lot of software to figure out the probabilities of where I am. So it does 504 calculations very frequently of the probability that I am in this location and I am in this orientation. And if all 504 calculations based on everything I just told you All the inputs from all the sensors says, yep, all our calculations kind of align, then OK to move. And then these also run 24/7 and autonomously recharge themselves. They run on lithium-ion batteries. They can run for 8 hours, but then kind of like an electric vehicle, you're sitting around all night waiting for the machine to recharge. So what we do is a coffee break approach. So every 2.5 hours it goes around and finds a charge pad, drops 2 electrodes, charges back up from 70 to 85%, and then goes around for another 2.5 hours and then finds a charge pad again and recharges themselves. And you can have multiple machines share charge pads so you can play who's on first and all that fun stuff. And then over time we get to add more and more sensing capabilities. So long-term what we want these machines to do is be able to see See, feel, hear, and smell. And one of the things that's coming this summer is— I'm really excited about— is a mic array. So this is putting 16 microphones on the machine and allows us to localize sound. So I heard cars coming from that direction, I heard people talking in that direction, I heard glass breaking in that direction, and I heard a cell phone going off in that direction. That could be very interesting information. It's pretty cool. I guess the opposite of voice recognition. So everyone working on voice recognition is trying to drown out all the noise in the background. All that noise is actually really interesting for us because cell phones going off at 2 o'clock in the morning on a Wednesday at a certain location seems fine. 3 o'clock in the morning, cell phone goes off on a Saturday, you know, not normal. So think of this as advanced anomaly detection, really trying to understand the environment over and over again. to see what might be abnormal. Alright, so I'm going to do what you're not supposed to do when you guys go start pitching your companies, and we're going to do a live demo. So instead of walking you through this, I am going to log in to the Nightscope Security Operations Center, which we call the KSOC. So this is live at our headquarters, but That's not too interesting, so I'm going to switch and we're going to go to one of our clients in Fremont, California. So this is Northland Controls. You can see there's 2 machines running around in their environment. This is live streaming HD video from one of the machines there. Obviously we can record and analyze all this stuff. We can go look at license plates. So this gets interesting where, you know, you can pick up a license plate, but what do you do with that data gets more interesting. So if it's a retailer, you can do parking utilization. So we know the number of vehicles we see every hour, we know the number of parking spaces we're monitoring, so we can give you low, medium, high utilization. Or one client asked for a parking meter, which is in the middle here. So any plate, any vehicle sitting there more than 24 hours, we want to get an alert. So these vehicles have been parked there for 1 to 12 hours. Let's see here. We can also go check out a thermal image. So if anything's over 400 degrees, we'll also kind of raise a flag. And then the signal detection that I talked to you about, this is that rogue router detector. Also detects mobile devices, so there's a MAC address ID of devices that are all within the vicinity, and so we can ingest the whitelist of all the people that should be in that building. And then one big tell for industrial espionage is someone walking around with 4 devices. In Silicon Valley maybe, but most people don't walk around with 4 devices, so if I detect A person, thermal scan 98 degrees, 4 devices, flag. So the other way to think about this is to provide security guards and law enforcement officers really smart eyes and ears to be able to do their jobs more effectively. So there's people detection. So this is all live at a client. If the network's cooperating, I'm gonna switch over to Draper. You have this one. I have this one? Okay, so there's machine 18. And look at all these smiling people who are now Being surveilled. That's just for fun. Let's click on this. So somebody inside a Draper. Who's that? Oh, that was me. Busted on the phone. You can also— Send a prerecorded message either manually based on time, like, hey, you're at the mall, it's 30 minutes before it closes, or by location. In case you're wondering, I'm a security robot. It's nice to see you here. And obviously you can customize that depending on the location. Happy birthday to you. Or you could do a live broadcast. So God forbid there's a live shooter situation and the dispatcher security team wants to be able to speak through a machine or through all 100 machines, you're able to do that here. So like a mobile PA system. Or you can establish a 2-way call that allows a 2-way conversation. communication between the machine and the operator, as I mentioned earlier. Okay, since nothing crashed, I am going to get out of this now since it's not supposed to do that. All right. We also provide our clients with an iOS and now Android version of the machines— sorry, of the KSOC that allow us to have a security guard have everything I just showed you in the palm of their hand. So now you have a 3 million square foot distribution facility, you've got 2 guards. Like, how are they physically supposed to be able to do this? But if they have the capabilities in their hand, then they'll be able to drag that icon that's on up here, drag that icon over here, and have the machines autonomously go to that location because they want to inspect something gets really interesting.