I Taught my Spider Robot to Chase People... for Science!

Added: 2026-05-13

[00:00:00]I have a problem. My hexopod robot is too stupid and I really want it to chase people. Uh, not for any villainous reason. I just think it would be really funny. So, with a little help from Revo, who sent their new Inspire 2 3D scanner, I'm going to do a little programming and make my Hexpod smart enough to chase people.

[00:00:25]So, here's our robotic victim. His name is Red Hex, and I built him a little over 2 years ago. It's got six legs with 18 motors, and it's all based off of an open- source project called Chica Hexopod. Gradually over time, I've made a ton of improvements, like designing my own custom parts to go on top, and I got to bring it to a bunch of really exciting events, which was awesome. But over the last few months, the robot was just sitting in a box. So, when a sponsor reached out to give me a 3D scanner, I was thinking, "What the heck am I going to do with that?" until I realized a 3D scanner can also be a really fancy camera. And everyone knows that a really fancy camera plus a really fancy spider equals a spider that can see people. And you know what comes next?

[00:01:05]Before we figure out a way to mount the camera on top, we first need to repair the robot. Since I've had this thing for so long, some of the parts had literally snapped off or others were just being worn down over time. So, I went and found the strongest material I had in my closet and started printing replacement parts. If you're curious about the design of this hexopod, I definitely recommend checking out the open source project it's based on, and I'll have that linked in the description. After those parts were printed out, I went ahead and started taking apart each leg.

[00:01:32]Eventually, I got all the components replaced. But now, we needed to figure out how to mount the camera on top of the robot. Since the mount needed to be strong, I started with a little/420 mounting piece that had a small attachment on the bottom. Then I designed and 3D printed a mating component that allowed it to slot in and then attach to the head. Now the mount was ready, but before anything, I wanted to test out the scanner. So let's check out the Revooint Inspire 2, a super small, affordable powerhouse of a 3D scanner. It's got a whole bunch of sensors on the front that allow for precision 3D scanning. This thing is crazy. You can 3D scan almost anything fully wirelessly, and you can even attach it to your phone to scan stuff literally anywhere. Using the full field scanning mode, I was able to get highresolution 3D models of random objects around my room. The scans had highdimensional accuracy and even included color. This is super useful for digitizing things or gathering data on objects to make 3D models of, design, attachments, customizations, and so much more. So, click that freaking link in the description and get yourself one of the coolest scanners I've ever seen. And this video wouldn't have been possible without Revooint sending out this awesome scanner. Revooint asked me to talk about how it integrates into my workflow, but I'm actually just going to hack into it for the data from the camera feed. So, let's check out the Python program I wrote that taps into these video feeds.

[00:02:52]Now, to understand what this Python program does, I'll have to explain a couple things. The first one being how it's looking at the things in front of it and how it's interpreting that information. The Revooint scan app has two video output feeds, an RGB camera and a depth camera. The depth camera shows how far things are based on brightness. And the RGB camera is just a normal camera. The program I wrote takes sections of your screen and uses them as visual input to OpenCV, which is a machine learning model interpreting visual things. We're basically going to use high-speed screen capture and pull a frame into the engine. Now, seeing labels of what an object could be is kind of useful, but not really. What my program does is it splits the view into three little sections. These zones allow us to get a better idea of where the object in our field of view is actually located. Using both the depth information and the RGB camera, we can get a sense for where we should move, what we might want to avoid, and lots of other useful information for automating movement. All of that information is then filtered and translated to a movement vector. This movement vector acts a lot like a joystick on a controller. Then we just map the stabilized vector data to a physical mouse position. Now, why are we using a physical cursor position? Well, it's actually a little bit of a hack. This hexopod communicates between two Android apps, and one of those has a bunch of movement gates that tell it exactly how to move around. These have been optimized and programmed for quite a while. So, it's difficult to start from scratch on this front. Instead, I'm emulating the client app and just inputting things as mouse cursor clicks.

[00:04:25]Right. So, even though I've written a system that will work in theory, we actually have no idea if it's going to work in real life. That's because a bunch of variables could come to play.

[00:04:34]Although I've added filtering and smoothing to my control system, it's entirely possible that the camera or the YOLO V8 model is fully convinced that a stick in the park is a human and it's going to go chasing towards it. This is why advanced robotic systems include something called a fail safe that shuts the robot off by default when failures are detected. This is a super critical feature.

[00:04:55]>> There's no fail safe. Okay. Should it adjust?

[00:04:59]So, follow me.

[00:05:02]>> Oh, no way.

[00:05:09]>> Overcompensating issues.

[00:05:11]>> I'm so scared, >> dude. It keeps saying refrigerators.

[00:05:15]>> So, person far away.

[00:05:17]>> So, if I go in frame >> Yeah. The thing is it sees person and the feedback loop is so long that it's adjusting and it's overturning and then it stops seeing person.

[00:05:31]>> All right, so what you just saw was our first full test of the system and we've learned quite a bit. The main thing that we found was that the system would severely overcorrect when trying to place the targeted object in the center of its view. So much so that it would just turn all the way past the target.

[00:05:45]One of the reasons this was occurring was because of the narrow field of view of the camera. But there were a few other issues that I had ideas on how to solve. First, I clamped the maximum turn speed to just 50%. Which makes the bot less responsive, but it helps a lot with overshooting. Then, I improved the overall efficiency of this system to reduce delays between each connection. I did this mainly by removing complicated detection boxes. And then, I made sure that it was only actually looking for people instead of every single random object. With this updated code, I went out to test it again. but this time with some friends.

[00:06:20]>> Okay, what do you got?

[00:06:21]>> So, we've got toolkit, we've got a camera, backup battery, >> and then obviously we've got >> and that's all we need.

[00:06:31]>> Banana car.

[00:06:32]>> Dude, that is fire.

[00:06:33]>> That is a fire banana car. Okay, so the first thing we want to do is just get it running.

[00:06:41]been in the in the dorm picking up dust >> over the last one day.

[00:06:47]>> Hold it. Hold it. Hold it.

[00:06:49]>> Here we go.

[00:06:49]>> Oh, >> that's scary, dude. Oh no.

[00:06:53]>> So, the IP that we're connected to should be the IP and the server, but it's not connected.

[00:06:59]>> What?

[00:07:00]>> I forgot the battery for the camera.

[00:07:02]>> Oh my god. Are you serious?

[00:07:03]>> Yeah.

[00:07:04]>> I didn't mean to do that, >> dude. And when you're editing this, it's going to be so weird. Why is it slowly rotating?

[00:07:13]>> Look at it.

[00:07:15]>> I wish this was easy. I wish like building these projects was easy, you know? That would that would make things a lot nicer.

[00:07:28]>> Oh. Oh.

[00:07:50]Follow me.

[00:07:55]>> Okay. So, it's going away from you.

[00:07:57]>> The direction's inverted.

[00:08:00]>> Yeah. So, it kind of inverted >> theoretically. Okay. So on the tracking camera, it sees you guys. It says 59% threat detected.

[00:08:08]>> You can see your reflection in the laptop.

[00:08:17]>> Oh, I don't like that.

[00:08:19]>> So I have like basically a protocol that's supposed to when it gets close enough to somebody, stop chasing them >> and then move back.

[00:08:27]>> Instead move back and look at them. And I think what's happening is that it's seeing the depth camera and seeing the bright stuff over there and then thinking that that's someone really close and switching into the mode where it's supposed to be looking. After that test run, we found a whole new set of issues. And although it was tracking and following people, it wasn't really chasing them, which is the end goal. I already had a general idea of what fixes I needed to make. But before I did that, I had a side quest to complete.

[00:08:59]This side quest is Robots on Ice 2026.

[00:09:03]It is absolutely freezing out here. I brought Red Hex out here to see if it could actually walk on an icy surface.

[00:09:46]At this point, I'm honestly shocked how well Red Hex is able to walk on the slippery ice. Turns out it's kind of the perfect robot for this kind of application.

[00:10:20]After running the robot on the rink for quite a while, I took it off and honestly I was surprised. There was no visible damage. So overall, pretty successful.

[00:10:45]That's about it for this section. I hope you enjoyed this little um excursion to put Red Hex on ice. That was pretty awesome. But let's get back to optimizing the AI and integration. After Robots on Ice, I spent a bunch of time making the software actually decent. It took a while and a lot of testing, but eventually it started to work.

[00:11:06]>> Got to make it fast. Go that way. Go that way, I guess.

[00:11:14]>> Oh, Jesus Christ.

[00:11:18]>> It's overshooting a lot now.

[00:11:30]So now with every piece of the puzzle put together, we headed out one more time to finally achieve the goal of making my spider robot chase people.

[00:11:55]Did you resend the code through Jitub?

[00:11:57]>> Oh >> Start calling it Jubile.

[00:12:12]>> Too easy. Good rain.

[00:12:30]Heat. Heat.

[00:12:58]Whoa, whoa, slow down.

[00:13:09]It's coming for you, bro.

[00:13:12]Run.

[00:13:15]That's autonomous right there. Oh my god. We need an arachnophobe in here.

[00:13:19]That would be a pretty funny reaction.

[00:13:21]Someone like deadly is afraid of spiders.

[00:13:23]>> Please. It's important.

[00:13:26]>> It's important. Please.

[00:13:28]>> Wait, actually.

[00:13:28]>> Yes.

[00:13:29]>> Yeah.

[00:13:29]>> Okay. Give me like 10 15.

[00:13:33]>> No.

[00:13:34]>> No.

[00:13:35]>> No. You You have to come as quick as possible.

[00:13:38]>> All right. I'll >> 104. Is he poisoning a good boy?

[00:13:41]>> Oh, >> we'll see what he thinks of my little invention.

[00:13:44]>> Um, guys, I'm afraid of spiders. My worst nightmare.

[00:13:50]>> Wash your face.

[00:13:53]>> Oh, green.

[00:14:04]>> Follow me. Follow me.

[00:14:06]>> What's up, bro?

[00:14:08]Checking out, man. See the little thing?

[00:14:10]>> It's a robot.

[00:14:16]>> Of course, it freaking crashed.

[00:14:18]>> Oh, are you serious?

[00:14:23]>> Bro, what?

[00:14:33]>> I did it, bro. Want a hot dog, man? Of course you did.

[00:14:39]Okay. What's scarier? The robot or >> That's about all I have for this video, but if you liked it, you should subscribe because I'm making more awesome stuff like this all the time and you don't want to miss out. Shout out to Revo for the scanner and a major shout out to Make Your Pet for designing and publishing the Chica Hexopod project, which this whole Hexopod was based on.

[00:14:58]Hope you all enjoyed and I'll see you in the next one.