Embedded - 534: I Think We'll Add a Head

Episode Date: September 17, 2026

Aaed Musa joins us to talk about building robot dogs, custom actuators, and his YouTube channel. He shares how he moved from 3D-printing servo robots in his parents' garage to designing capstan drives, and how he uses reinforcement learning to train quadrupedal walking gaits. We also talk about his senior design project (CARA 2.0), his Spider-Man web-shooter build, lessons in storytelling and video editing, and why he's starting a company to bring friendly, expressive quadrupeds into the home.  Aaed Musa is a mechanical engineering graduate from Purdue University, a hardware creator on YouTube, and co-founder of the robotics startup Mecha.  DISCUSSION RESOURCES & LINKS AAED'S CHANNELS & PROJECTS: Aaed Musa on YouTube: Project videos covering quadrupedal robot dogs, Spider-Man web shooters, and custom actuators. Oh, and planetary gears. More detailed project write ups are under projects on his personal site, including CARA 2.0. Github repository Robots Everywhere with Aaed Musa | The Amp Hour ROBOTICS & ACTUATOR TECH: Dyneema DM20: synthetic rope with zero creep used in zero-backlash capstan drives NVIDIA Isaac Sim: simulation and reinforcement learning environment for robotics OpenMV: Very smart MicroPython-powered machine vision modules. Kwabena Agyeman has spoken with us about OpenMV before: 497: Everyone Likes Tiny  477: One Thousand New Instructions 212: You Are in Seaworld SOME YOUTUBE CHANNELS MENTIONED: Good 3d printing youtube channels: 3DPrintingNerd  CNCKitchen  MadeWithLayers  Jeremy_Fielding: great for explaining how things work and coming at engineering from a non engineering direction (Embedded 385: I Just Wanted an Industrial Arm) James Bruton: Robot dogs!  Mark Rober has fantastic storytelling and approachable engineering. Stuff Made Here has robots and gadgets, generally a broad scope of engineering.

Welcome to Embedded. I am Elysio White alongside Christopher White. Our guest this week is Spider-Man and his Rope Bot Dog. Hey guys. How's it going? I mean, Ait Musa is here to talk about his YouTube video channel. Hi, Ait, welcome. Hey, it's great to talk with you guys. Could you tell us about yourself as if we met at lunch at OpenSauce? Yeah. Sure. So yeah, my name is Ayd Musa. I am, most people know me from my YouTube channel. I like to make robots and a bunch of different gadgets. I've always been interested in engineering. As a kid, I was like really into 3D printing. And so in high school, I started a YouTube channel and I would just post a bunch of the stuff that I was building and started to get a bit of a following. And yeah, I just kept on making projects and posting them. I'm a recent mechanical engineering grad from,

Purdue University, and now I am co-founding a startup where we're making robot dogs. Awesome. And I'm disturbed to learn there are people who are adults now who enjoy 3D printing as kids. Yeah. I think there's a lot of us out there. Yes. We're going to do lightning round, where we ask you fast questions, and we want short answers, and if we're behaving ourselves, we will say, we won't say. I'm sure you don't want to consider that? Or tell us. us more. Tell us more. Are you ready?

Yeah, let's do it. Peter Parker or Tony Stark? Peter Parker. Peter Parker or Miles Morales. Ooh. Peter Parker. Miles Morales or Hero Hamada. I have no idea who that is.

That second person? Oh. Big Hero 6. Oh, oh, here. Okay. I'll still go Miles. I mean, Hero made some great robots. Yeah, Herod was pretty cool.

But I think I like Miles more. Ray Finn or Po? Ray Finn or Po? Finn. B-B-8, R2D-2 or C-3PO. R2D-2. He's too iconic. Jarvis or R2D-2?

Jarvis. Jarvis is really cool. What was the original planned name of Rom the Space Night? I said trivia. That's just pure trivia. I have no idea. What is this from? I have no idea.

Space Night, an ancient comic. It was their first robot comic. And the original name was Cobol. I should have limited to Marvel movies, not Marvel, all trivia. Fine. I am really an MCU fan. I don't know much about the comics. I just watch the movies

in the TV shows. Stepper Motor, Servo, Solenoid, or Hydraulics. Stepper. Complete one project or start a dozen? Complete one. For sure. Do you have a tip everyone should know? Oh, always seek knowledge.

Always learn. Every day is an opportunity to learn something new. What is your favorite fictional robot? Tars from interstellary. Oh, yeah. Those are cool. Well, he was cool. It was cool.

Okay, so not in Lightning Round, what is your favorite non-fictional robot? Non-fictional robot, I would have to go... You can choose your own. It's okay. Fought by Boston Dynamics. I mean, yeah, it would be easy to choose my own, but for me, Spot was like my gateway drug into robotics. Like, seeing Spot in high school, like, really got me excited about robotics, particularly quadrupedal robots.

It's just such a cool robot. It looks so, it's so cool. Okay. So you got into robotics in high school and you built your own robot dog. Yeah. Yeah, built several. So tell us about the first one. The first one was pretty bad looking back on it, but it was made completely from Servos.

It was called Ares, like the god of war. And yeah, it was just made from Servos. It was something that I whipped together in my parents' garage. and yeah, it was able to walk, just not very well. It was fully 3D printed, and I just made it for anything I could just find on Amazon, and I could find around the house. And, yeah, I think if I didn't build Ares, I probably wouldn't be building these cooler robots today. So if you compare Ares to Kara 2,

What is the biggest improvement? You can't choose everything. Yeah, that would be my answer, everything. But more specifically, the actuators, the actuators are like the heart of pretty much any robot. So if you have bad actuators, then you just have a bad robot. And so with Ares, I was using seromotors, which are cool for certain applications, but they have super high reductions and are not very compliant,

not very back-drivable. And so with Kara and pretty much all the other quadrupedal robots I've built since, they've used brushless motors with very low gear reductions, gear reduction gearboxes, which allow the actuators to have high compliance, high backdropability,

high torque transparency. And so it feels more like a actual biological limb than a server-im motor does. We're a firmware engineers. What does compliance mean? Compliance is basically... You know about motors? I don't know about motorists. She's looking at me like, you're dumb. I'm using my hands

to show him. It's not doing great. Yeah. So it's basically like the, I would say, the springingness, the softness of an actuator. So like your limbs are very compliant. Like if you're standing and someone, you know, pushes you, you're able to soften your pose to be able to like that rebalance yourself but you can also like stiffen your joints if you're i don't know punching something or lifting something heavy and so compliance is that ability to be able to um soften the the actuators uh joint and uh yeah it's it's very important in robotics

because you want your actuators to be able to to soften and to be able to stiffen depending on you know the application okay I'm glad you answered that because I actually did have it wrong. I thought it was mostly about, I had a really, really bad arm robot, which I kind of loved, and I was trying to use software to deal with some of its difficulties. And it was servo-driven. And it was servo-driven, but like if you sent it to the same place, if you went home and you sent it to a place, okay, fine, and you sent it to the same, you went home, you sent it to the same place again.

You didn't end up. you were you were a couple of centimeters out you were not you were not going to get to the same place each time which is why i was trying to figure out the software with the camera and and helping it and stuff there was just a lot of play in the whole thing that's so yeah that's backlash um okay if the joints like half slop if you like command it to the same position you know 20 different times and it ends up in 20 different places then probably have like a backlash issue and that's just uh something that's inherent with with gears it can removed a lot, but if you have gears, you technically will always have a small amount of backlash.

But some gears are more sloppy than others. But you've been really focusing on capstan system gear things that decrease the backlash quite a lot. Yeah. So capstan drives basically have like zero backlash, and they're rope-driven speed reducers. So they're basically like gear alternatives and they sort of work the same way gear does except like instead of teeth they have, they use rope. So imagine you just like take away the teeth from two gears. And so the way they interlock is they have rope wrapped around both. And so as one is turning, it's essentially using tension to pull the other in the other direction. And so they operate the same way as gears. It's just that they don't have teeth. And they are, they have

basically zero backlash, and they're super backdrivable, super compliant. But they are, they've only recently become more available in part because of the rope technology improvements? Yeah, so like the key thing with capstan drives is if you want to be able to have that zero backlash, your rope can't stretch. Oh, right. Okay. Rope loves to stretch when it's put under attention. And so pretty much all ropes will stretch except for this one rope called Dynema DM20, which basically has like a zero percent creep rate, which is amazing. And so that's like the only kind of, basically the only kind of rope that you can buy commercially that won't stretch. And it's a newer rope. I forget exactly which year, but it's, I think it's, you know, you.

younger than 20 years old. I forget all the details of how it was made, but yeah, it's a relatively new rope. And it's just, it's really cool because it's like the only way you can make a long, longstanding capstan drive. What about chain?

Well, I guess chains have the same problem. They're not under chains or metal cables. Like tracks or like bike chains. I guess. Yeah. So like steel cable doesn't stretch. So I guess I should qualify my statement and say it's the only synthetic rope that you can buy that won't stretch.

So yeah, steel cable won't stretch. The thing with steel cable, though, and I've tried this, is it snaps if it's wrapped around a small drum and a big drum. So there's something called a D-to-D ratio, and so with steel cable, it requires a certain D-to-D ratio. meaning that you can't wrap it around like a really small drum and then wrap it over a really big drum. Like that change in drum diameter like causes it to snap over time. And so if you're trying to make like a, you know, like gear reduction that's like eight to one, right? Meaning that your smaller drum is eight times smaller than your bigger jump. It just won't work.

So everything got better with as you made more robots. Yeah. Did the software improve a lot? Um, yeah, a bit. I'm not a programmer. So, like, all of my robots thus far, um, up until now that I'm doing the startup, I've basically just been like, uh, hard-coded step sequences, trotting sequences where like the diagonal legs, um, move in pairs.

And so I haven't proved that. Um, but, you know, now that I'm doing the startup, I'm taking bigger leaps because now I actually have someone who does software and now we're using reinforcement learning. Um, but with my own like, personal robots. The software has improved a bit, but now I'm taking an even bigger leap by having AI do the walking and not manually programming gate sequences. But with dogs, Abo was long before you were born, which kills me. But trotting is not the only thing dogs do. Yes. Yeah. I mean, I've seen you make it wave, which doesn't seem that.

truly difficult other than the balance problem. But you aren't having to do play bows and asking for pets and rolling over on its tummy or on its back to get Tommy rubs. You aren't acting. You don't have a dog agent in your robot. Yeah, that's correct. Yeah. So there's all of my personal quadrupedo robots have not had any sort of,

any sort of like intelligence. It's all like hard-coded commands. But yeah, what we're trying to do with the startup right now is the exact opposite. It's like we want it to feel like it's alive. We want it to be able to like emote. And it's funny you mentioned a bow and a wave. Like we actually just last week got our robot to do a bow and a wave. And that was all learned through a reinforcement learning policy.

And yeah, that's kind of where the technology is. going today, like five years ago that that wasn't possible, but because of AI now that that stuff is. And yeah, the goal is to like make a robot dog that feels like a real robot that can, you know, go and fetch something if you tell it to that can roll over and do cool tricks that can hop onto the couch with you and be expressive. And, you know, maybe your robot walks different than someone else's robot, someone, another person's robot. And, you know, it can show expressive, through even just its gate. It doesn't walk the same.

And so the world is, the world is its oyster. It can essentially just, you know, learn to do practically anything that an actual dog can do and more. We took our dog to the beach and there are dogs that prance and dogs that stomp and dogs walk differently. And it's not just about how tall they are. Yeah.

But, well, wait a say. I have a bunch of questions about the reinforcement learning, but I started to ask about the mechanical stuff. Okay. But what, but he was, you were talking about the software and how you didn't think it was great or whatever, because you do fix sequence. But you do have a control system in there that's, I've seen in your videos that you can push the dogs and they write themselves. Yeah. There's stability feedback control systems in there. So I think you're selling the software.

heard a little bit because it's doing it's doing some very complicated things just to be stable right yeah yeah yeah so there's a bit of stability embedded into the software it has an iMU an inertial measurement unit and you know it can sense its own uh roll pitch and yon how fast it's spinning and it can correct for that if you if you give the push so yeah that's another improvement that was made over the course of my robot making years. And yeah, something else, I was able to, like, program, I call it pose control where the robot's able to sort of like tilt in, you know, roll pitch and yaw and sort of demonstrate its kinematics. And that was another thing that has improved over the years. Yes. And if you ever put a head on your dog, you might want it to tilt its

head. Yeah, yeah, exactly. Okay, so the reinforcement learning, what are you reinforcing against? Are you showing it videos of dogs? You can, but for just getting it to like walk, we're essentially like giving it a bunch of targets to hit. And in simulation, it's running like many, many, many simulations, you know, hundreds of thousands of simulations. And it's trying to hit those targets. So, for example, we have a bunch of, they're called rewards. And, you know, maybe one reward is like, I want you to walk at two meters a second. Like, I want you to keep that, keep that velocity, keep that pace.

And, you know, if the robot in simulation is walking at 1.5 meters per second, well, there's definitely, there's clearly a gap there. And it's not doing something right. So then it'll do another simulation and try to get closer to that. Maybe in that next simulation, it walks that 2.5. five meters for second. Oh, too high. Okay, it's going to adjust its gate to try to achieve that. And so we have a bunch of these rewards of things that the robot is trying to achieve. And so over the course of every simulation is essentially trying to hit all of those rewards. And when it hits all of those rewards, then we have a functioning policy.

And the whole work with reinforcement learning is determining what those rewards should be and what the gains of those. reward should be. So like how much should it focus on hitting that two meters a second gate versus another reward? Maybe we want it to, I don't know. To be able to stop in a half a second. Yeah, or something like that. Because sometimes your rewards are like that where you have one where you want to go fast and one where you want to stop and you really have to do both. Yeah. So yeah, we have a bunch of rewards. I think there's around like 20 for what we're doing. doing right now and determining like what those rewards should be and like the gains for each of those rewards is is the whole work. And that's what a lot of reinforcement learning research

is going into. It's like how do we determine how a robot should walk? Like what are the parameters? What are the rewards that we need? Are you, this is on top of your control system. You're not just giving it like raw ADC values and raw DAC values or drag. values. You are giving it, here's how to do a step, or are you, where is your control system versus your reinforcement learning? I know it can be anywhere in there, but where's yours? So the, if you're talking about like the sort of classical controls, like a PID controller, that all just lives on board on the actuators. So, you know, we tell, like the reinforcement learning policy is determining how the robot should move. And then we just send those.

position commands to the actuators. And if we've tuned the actuators correctly, then, you know, we send the actuator to move to a certain position and then it's able to move to that position. And we don't really even touch the low-level control on the actuators after we've tuned them. Okay. Cool. I've seen people try to do reinforcement learning from the lower levels. And I never quite understand why. Because there is a point at which you just wanted to move to where it's supposed to go. And yet I had that terrible arm where... Yes, but the goal with the arm was to do stupid things.

To be clear, it wasn't a badly built arm. It was because of... It wasn't very cheap arm. It was you bought the cheapest arm you could find in hopes to fix it using tons of software. Yes. And instead, you just made comedy, which has its own value. For sure. Yeah, yeah. I mean, Iron Man had the dummy robot. Right. Right. Just was served as comedic relief for the Ironman movies.

So, Kara, you make videos and you post them on YouTube about your projects. Yeah. And you're very good at it. And you're very good. Thank you. Your most recent video was about making Spider-Man web shooters. Yeah. That was so good. You showed the slip-ups in the design process.

It could have been a video of here's how my perfect project worked, but you showed how it wasn't a perfect project to start with. And it got to a goal, but it wasn't a direct shot. 100%. Yeah. And I mean, that's something I try to show in all of my videos because it's true and it's real. Like, there's no projects that you'll work on for several months that won't have road bumps, that won't have, you know, changes. Like, you know, sometimes I even just have to change the direction of the project to, like, while I'm doing it. And so showing all of that, I think, is a pretty honest, but I think it makes for a better video.

Because, like, I think it would be very boring if I just showed you the cool highlights of everything working because you just intuitively know. that that's not actually what happened. Engineering, yeah, it's hard and difficult, and the whole goal of engineer is to, you know, fight through those challenges and problem solve and figure out what's wrong and how to fix it. And so I think if I can show that, then, A, I'm showing, you know, how I think,

but B, I'm giving you just a much cooler video, something that's more entertaining, Because like any movie, any story that you watch, there's always a problem and there's always a resolution. And so, you know, I try to model like my videos in the same sense of, in the same sense that a story is told where like, you know, you have characters or character, just me, a setting, you know, my bedroom and a problem. I'm trying to get this thing to work and it's not working. And then the solution, I eventually, you know, figure out some sort of hack to get it to work. And you do a lot of your engineering in what seems to be your room. How much is that, is that a studio or is it really like where you live?

Well, yeah, so it's just my bedroom. Right now I'm in a two-bed, two-bath. So I have a roommate. But like pretty much all the shots of my videos are like inside my room. Or, you know, if I'm, I have like two printers. my office space. So sometimes I'll record there. Just showing like the printing. But yeah, like all the assembly, all of the testing, that that's just in my really small San Francisco apartment bedroom. And that's really all you need. Like you don't need a fancy studio to

record YouTube videos. You just need a desk and a couple tools. A couple tools. I mean, yeah, maybe I'm selling it short. I've acquired a lot of tools over the years. Lighting seems to be the big difficulty. We had to record a video lately recently. Lighting was difficult. Yeah, I have like two ringlights and I have a pretty nice balcony out here. So natural lighting is good. But yeah, I have a ring light.

And then I have this like really cool overhead desk light that I think is just a lifesaver for me because my room actually doesn't have any lights. So if I didn't have it, I would just be in pure darkness at night time. But it also helps with recording. There are a lot of things that go into your videos. Yep. I made a list. So I'm just going to split out the list and hope you can tell me which one or two are the hardest parts. There's the physics and the planning and the equations.

We're all going to bunch that as one. The mechanical design and then there's mechanical limitation. The electronics, the software. those are all normal robotics things. But then you have a project write-up, the video editing, marketing the video, and getting the ad. Which of those, did I leave anything out, I guess is the first question. No, that pretty much covers everything. Which part is the most fun?

Oh, God. It's hard to pick. I mean, it's all fun to me. If I had to pick one, I'd say editing probably has a slight edge, just because, because that's like the culmination of the project. Like I get to, I'm sort of like doing everything at once when I'm editing. Like, you know,

I'm having to remember all the mechanical stuff and I'm having to present everything in a way that is understandable for people. I'm having to think about, you know, are people going to find this interesting, like the marketing aspect? And so, yeah, editing is like when everything comes together. And I'm sort of encapsulating like my four months or three months, of work into this one, like, 20-minute video. In some of your videos, I have noticed your T-shirt changes.

Does that mean you have, you're spending multiple days, or is that just simulating the passage of time? Yeah, definitely multiple days. I mean, I think any project I work on probably takes like a good three to four months. So, yeah, I'm definitely not. just recording everything in one day and just finishing it, yeah, in a weekend. It takes a while. So, yeah, the passage of time is very true.

So they're more project journals. You're recording as you go instead of, okay, I finish this project. Now I'm going to recap it. Yeah, I mean, it changes. Like, I try to record as I go, but inevitably, like, I always just hate what I recorded and want to record it again. So, yeah, there's a bit of that. But yeah, generally I just try to like record the, get the flow as I'm going.

And over the years that you've been doing this, has the video editing, the putting it all together, the rush of endorphins of finishing something been one of the highlights or has the highlight of it changed? Yeah, for sure. I think the highlight for me is definitely like finishing the video because it's, that's like I'm showcasing everything that I've done. it's always the most fun part. Even like before posting a video, like, I'm always just super happy to like have finished the project and to have showcased it in the video. Like even if no one saw it,

I'm just sort of just happy that I've done it. And I was able to accomplish the goal that I set out to do months ago. And yeah, that's always going to be the most fun part for me is like finishing editing a video. Which of the parts has gotten easier over the years? As anything, just like, it used to be hard and now it works on its own, or I went to college, and now I understand all the physics, so I don't have to think about it as hard.

No, the mechanical and the physics stuff, that's always, that always has its difficulty. Depends on the project. Like, if I'm making another robot dog, it's going to be a lot easier because I've just done that so many times. But if I'm doing like a completely new project and I'm not very familiar with the engineering, or the physics that goes into it, then it's a whole new set of learning that I have to do, which is fun because it's always fun to learn new stuff.

I'd say the part that just inherently gets easier with every video I make is probably just like the editing and the marketing, like figuring out, you know, how am I going to, like I've sort of developed my own sort of flow with the editing, and so you'll sort of see the similar beats with my videos and, you know, similar corny jokes and similar editing and similar transitions and whatnot. So that stuff sort of gets easier just because I've, like, done it so much. And I know how to create, like, I know how to flow with my own, like, personality.

And it's easier to, like, write scripts because I sort of know how I should go about writing it. And, you know, I know what parts to expand. expand more on and what parts, you know, maybe cut out because it might not have any relevance relevancy or interest for the audience and whatnot. Yeah, so the editing for sure gets easier. Do you ever go back and look at your earlier videos to see how much you've improved or? Or is that too difficult? How often do you go back and listen to do our old episodes? I don't. I don't do that at all. For very important reasons, I like myself and I don't want to be sad.

I'm the exact same way. It can hurt to watch sometimes, like watching an old video. For me, what I really don't like is like when someone is watching my video right in front of me. Oh, no, no, no, no. Yeah, that always, like, I don't know, it's like the sound of my hearing my voice with people around just, I don't know. It just doesn't sit well with me. So, yeah, I try not to, I try to leave the room if someone's, like, watching my video. I've had family members who will put my video on a TV with a bunch of other people watching.

And I'm just like, guys, what are we doing here? Yeah, it's not fun. But yeah, I really don't watch my old videos. It's, yeah, it's very painful to watch. But you were in high school when you started, and you started with like a Nerf gauntlet. Yeah. Did your channel help you get into Purdue? No.

I didn't actually mention my YouTube channel in like any of the essays that we had to write for college applications. I just sort of talked about like my love for 3D printing and my love for engineering, but I didn't specifically mention my YouTube channel just because my YouTube channel then was like nothing. Like I kind of just posted clips of things that I had built. Didn't really have much of a following, and so it wasn't really worth highlighting, I felt, at that point. Fair enough. When or how did your channel get more traction?

Was there a point where you said, wow, I'm a big deal in the rebellion now? Was there one video? Everybody was like, oh, that got noticed and then he picked up? It's slow growth. So I think the video that I think got me the really big boost was a video where I made a robotic actuator. And yeah, I made this robotic actuator with this off-the-shelf drone motor that I got. And I 3D printed this planetary gearbox to put with it. And yeah, I made a video on that.

And like it got like, you know, I think a couple, 100,000. views in like, I don't know, like a week or so. And that was like really huge for me. And at the time, and yeah, that was, I think, the video that sort of put my channel on the map. And yeah, ever since then, it's my channel is just kind of skyrocketed. Do you go out and get ads or do people just come to you and say, please use our 3D printer and here's a free one. And also can we give you money if you'll mention it. Yeah, yeah. So that's the cool part is like, yeah, I really don't have to like go out and seek, seek sponsorships. They all just sort of come to my inbox. And so I'm in a cool position, a very fortunate position where I can sort of just like sift through all of the sponsors

and pick one that I like. And it's not the opposite. I want to talk about your college experience because you obviously have a broad skill set watching your videos. You're doing all the mechanical stuff, all the electro-mechanical stuff, a lot of the electronics and software. How much of that was out of coursework at Purdue? How much of that stuff did you get exposure to there during your engineering degree, and how much of that did you either pick up on your own

or have since learned on your own? Yeah. So, yeah, this is an interesting question, because I get asked this sometimes. I'd probably say that, like, pretty much all the skills that I'm showcasing in like my videos and everything that I use to build my videos is self-taught. School is good at like a very specific thing. It's good at like teaching engineering school. That is it's good at teaching you the physics, the reasoning behind a lot

of the practical application stuff. And so you're learning your like coursework is mostly just like solving problems where you have to, you know, figure out the loading on this bridge or whatnot. And you're doing a lot of a lot of things like that. But like in terms of like actually learning how to build robots, like you can't really, it's hard to learn that by sitting in a lecture hall and taking notes on, you know, dynamics. Like you just have to go out and learn to build that stuff. Like equations can only take you so far. at the end of the day, if you are a mechanical engineer and you're building physical hardware,

you have to just learn to build physical hardware. And I really do think that you can be a self-taught mechanical engineer. You just, if you're building physical hardware, again, you just need experience. And college is needed for engineering, for sure, but it teaches you a specific thing. And practical application is just something you have to learn on the job or, you know, during internships. You have, I believe, credited YouTube to some of your knowledge about electronics and 3D printing. Do you have favorite channels? Oh, yeah.

Got a ton of favorite channels. God, how much time do you guys have? 3D printing nerd for 3D printing. I've been following him for a while. post a lot of great 3D printing content. C&C Kitchen also does a lot of 3D printing content made with layers. Again, a lot of 3D printing content. Jeremy Fielding, I think you guys have had him on the podcast.

I've learned a lot from him just about robotics and mechatronics and mechanical design. And he's a great example of someone who, you know, he didn't go to school for engineering. I think he went to school. I forget exactly what he went to school for, but it wasn't engineering. He became a self-taught engineer, and, you know, if you just met him, you know, you'd think that he went to school for this, but he didn't. James Bruton, another great roboticist. I've taken a lot of inspiration from his robot dogs when making mine.

Mark Rober, great inspiration for the storytelling aspect of making YouTube videos. He's a great engineer as well, but what I think most people gravitate to him, I think most people gravitate towards him because he's just so good at crafting stories and getting you excited about engineering, especially the younger audience. Stuff made here. He makes just these insane robots and these insane gadgets. And he's just, he's, he's really good at like a wide scope of the engineering. discipline. He just knows so much about like the software, the hardware, the machining, and it's very rare to find someone who can do such a broad scope of engineering. And yeah, he's really cool. But yeah, I won't ramble off too much. But yeah, those are just a

couple of the guys that I think are really cool. Looking at Kara and one of the recent videos, actually, I think it was your Capstone project for school. You did a video about Did you turn in that homework twice? Yeah, so Caro 2.0 was my senior design project. And, yeah, after making Car 1, I was like, cool. I think there's a lot to improve here. And senior design was coming up, and I'm like, you know what, what if I just use senior design as an excuse to keep working on my YouTube project?

And, yeah, it got approved, and I was able to. to do that with my friends. And so that was pretty fun. And the video you made for that Kara project for Kara 2.0, it was a little strange because it had a lot of cat meows. What were you trying to say? I mean, it wasn't just like one cat meow at the beginning. There were a lot of cameos.

Yeah. Yeah, well, so the whole, the reason I believe that I'm really interested in robot dogs is because I myself wanted to have an actual dog, and my parents were like, no, we're not going to do that. And so I think I just developed, like, this obsession with, yeah, exactly. Like, I just wanted to build one myself because, you know, my parents said, said no. And so they eventually got me cats. And so I'm like, all right, this isn't really what I asked for, but okay. And so, but I love my cats.

They're cool. They grew on me. And, yeah, that's why I had the cat me out. just because, like, I have cats now. So, yeah, that's the reason. Would it be easier to build a robot cat or a robot dog? Cats sleep more. Yeah.

I mean, in terms of, like, like, mechanically, I think it'd be the exact, like, it'd be the exact same. I mean, I call it or a robot dog, but it's just a robot, quadripet. Yeah, it's just a quadripetal robot. Like, it just has. four legs. And if you wanted to, yeah, if you wanted to make like a robot cat, maybe you would just program it to not be able to jump

and not be able to, or no, actually you would want it to jump because cats can jump very high. How do you program something to be sullen and haughty and off-putting? That's, I guess, where you have to deploy some reinforcement learning policies to give it a bit of emotion. Yeah, I guess you just got to program it to like

not to hop off your bed when you bring it to your bed or just walk away from you when you want to pet it. Roll over on its back, except three or four pets and then hit you. Yeah, exactly. Yeah, yeah. There you go. Robot cat.

All right. All right. Okay, so that actually answers that question, kind of. Why do you keep doing your channel? I mean, it's really cool. I'm glad you do. I don't want to discourage you. Yeah.

But it's not easy. Yeah. It would be easier to just do the projects. I like the showcasing of it. Like, you know, whenever I post the video and, like, I see that people are gaining value from it and people are like, oh, wow, this is really entertaining. This is really, this is really cool. I learned something new. Like, that just motivates me to keep doing it because, yeah, I could just build the project in my bedroom and show it to no one.

but I think it's a fun end goal to be like, okay, I'm not only going to just build this thing, but also create a story around it, create a video around it, and show it to other people because I think people can get value from it. And it's just fun for me. Like I actually like, again, the editing process.

Like, I like crafting it and creating something from the, a video from the project and not just doing the project. So a little bit of just, it's fun for me, but also, like, I think a lot of people get value from it. Does knowing you're going to present a project in video form or any other form change how you approach the project itself? Yeah, yeah, a bit. I'd say the presentation definitely I feel,

I feel like it has to be more polished if I'm going to show it to people. So, you know, even just, like, designing a lot of curves into the, into the part or making it look presentable or adding some color like a duotone style or whatnot like all of that all of the aesthetics is very much because I know people are going to watch this and I know people are going to judge if I if it looks really

really crappy so yeah I think a lot of the aesthetics is is driven by the fact that I know people are going to watch and I have to make this a video about it. But yeah, the engineering part, not so much. It's mostly just the presentation. And that's industrial design, right? That's a totally different discipline. I mean, you could make your robots act the same way

and as functional as they are while at the same time being made out of, I don't know, popsicle sticks or something, you know, not aesthetic. So is that something that you had to, is that something you just absorbed or something you consciously went out and spent time learning about? I think that's just something that I've just sort of picked up upon, like watching other people's stuff and just sort of using my own creativity.

I'm definitely not an artist. I can't draw to save my life, but I think I can make something that may look, that may otherwise look super mechanical and super complex. I think I have the ability of making it look a little more soft, a little more presentable, something that people can look at and not think is a super industrial, super high-techy thing with like a bunch of wires poking out of everywhere. Yeah, so I think that's just something I've sort of picked up upon while doing a lot of projects. It's funny. You say you're not an artist, and I understand because I always say that myself.

and I think it makes me think about identities and I definitely I'm happy to say I'm an engineer but I won't say I'm an author unless somebody really tries to get me to say it do you you're an author under duress I'm an author I guess okay

but you you're an engineer okay definitely I think you're going to own that one that one's not going to cause problem. There is YouTube creator, storyteller, and educator. How do those fit with you? You're like, yeah, yeah, that's all me? Or you're like, no, I'm just an engineer who sometimes has stories to tell. Yeah. Yeah, it's, that's a good question. I don't know. I don't, it's hard to put myself in like one specific box or even like multiple boxes. But I think in general, I just say I'm a creative. I like to make things and I like to share that knowledge with with other people. So yeah, I think, you know, YouTube creator, educator, I think, yeah,

that definitely all falls under me or I fall under it. Yeah, I think, yeah, I wear multiple hats. And I think you just kind of have to you if you make YouTube videos. I get that. I never introduced to myself as a podcaster. So as far as long as we're talking about identities. But do you, okay, graduate high school, you went to Purdue. You graduated Purdue last winter quarter, winter semester. And now how many people have asked you, what do you want to be when you grow up? And am I going to be the next person to ask you that?

Because I don't really want to, but I also want to know. So what do you want to be when you grow up? Yeah, yeah, I got asked that a lot. It's a good question. No, it's not. Yeah, I mean, you know, everyone should be future-driven. And, yeah, I mean, I understand at the very least why people ask it. Like, they're like, oh, you're doing these cool stuff online, you know, where do you think it's going to take you?

Is this the thing that you want to do? Or, you know, is this sort of like a pit stop towards? something else. So I understand the question. Yeah, after graduating or like before graduating, I thought I was going to just go to grad school. And so I kind of had my mind set on that. But I realized that I didn't really want to go to grad school.

I kind of just wanted to said I was going to go to grad school just because, I don't know, that seemed like the next logical step. Like, all right, get a master's now after you've done your bachelor's. But, you know, when I look back at my entire undergrad experience, I realized that like, everything that I learned wasn't actually helping me do the thing that I was passionate about. Gave me some, you know, some of the behind the scenes knowledge, some of the first principles knowledge. But, you know, if I'm making a robot, I'm using the skills that I used to make the previous robot. I wasn't using the learnings of my, you know, dynamics class. Like, it was all, everything that I was using to build robots was everything that I was

self-learning and everything that I was learning from watching YouTube videos and doing research. And so, yeah, after thinking more about that and, you know, I was already talking with, who's my, Aaron, who's my co-founder now, and we were kind of talking about just the robotic space. in general. I figure that, you know, I think the best next step for me is to just, like, go out and work on a startup because, you know, why not? Like, you know, I think this is really the only time in my life that I will be able to, like, just work super long hours and to just, like, completely take that risk of setting out to do a startup that, you know, may or may not, like, work out. And so, yeah, I figured that that was the best next step for me.

So CEO and co-founder of the startup is going in the identity column. Is CEO or are you? Yeah, so I'm not CEO. My co-founder is CEO. We don't really, I don't really have a title at this point. I guess I'm just the mechatronics. Well, then your title is CTO. You get a C title if you're the first two.

It's just those are the rules. Yeah, it's still very early on, but yeah, I guess I'll take the CTO title. I think where do you want to go with your career is a question that's asked because it seems like you have so many possibilities in front of you. Everybody who asks it is just a little bit jealous. Yeah, maybe, or if you're my parents, then concerned. So it sounded like the separation between the theory that you get in classes and the practical applied knowledge you've gotten through just doing things and learning from other roboticists. It didn't get any closer in college. It actually got further apart.

Did I understand that part right? I wouldn't even say it got farther apart. It was just sort of indifferent. Sure, I picked up one or two things here and there in college that maybe helped my understanding of physics or of just mechanical design. But, you know, if I didn't go to college, like I think I would still be doing what I'm doing right now.

I'd still be able to make the robots that I'm building. So it didn't really. hinder or like enhance my ability to to make robots. It was just sort of indifferent. And did Purdue have a robotics club you were part of? Yeah, Purdue has several robotics clubs. And I think robotics clubs are great ways to get like practical hands-on knowledge. I was part of the, the VEX robotics team for like a semester.

And then I also did Purdue prosthetics, which isn't, I don't know if I'd classified as robotics, robotics adjacent maybe. But yeah, I did that for like a semester. And then I just, I felt like I just wanted to sort of do my own thing in my dorm room and, yeah, just devote all my time to working on personal projects. And so, yeah, I quit those two clubs and just did my own thing. It's interesting whether you decide to follow your own vision or whether you decide to have a larger team. Robotics is often considered a very team-oriented thing because of these specialties of mechanical engineering and electrical and software.

And yet you are doing most of it yourself or with a small team. Do you think there are benefits with that or do you think it's mostly you can only make a robot this big? Yeah, there's definitely benefits and drawbacks for sure. I mean, I was on a robotics team in high school. I did first robotics FRC. And, yeah, with the team, it's easier to divide labor. You know, you don't have to own.

Yeah, exactly. You can specialize and get just really good in one discipline, right? So, you know, I was doing all the CAD design for my robotics team, didn't touch any of the programming, didn't need to. And so someone else could take care of that. And yeah, that's the main benefit of working with the team is that if you have someone that you trust and that's good at what they do, then they can fully own that stack. You can fully own another stack and you guys can work in parallel and get things done much, much faster than if you were just doing it by yourself. Working solo, you don't have that. But what you do have is freedom to sort of make your own decisions about everything.

You know, you don't really have to be beholden to someone else's creative vision. You can sort of just go at your own pace, make the robot exactly how you'd want to. It's going to be a lot more work because you have to do everything and you might have to learn skills that you don't know. But you have a bit more creative freedom. So, yeah, there's good and bad. There's benefits in drobex. Are you familiar with robot operating system, Ross? A little bit.

What are you using for simulation for reinforcement learning? Oh, we're using Isaac Sim. That's Invidias thing. Yeah, yeah, the NVIDIA software. I'm not familiar with it, so I'm just trying to figure out. Yeah, so yeah, it's NVIDIA's simulation software. So, yeah, you can essentially load in your robot. and train it to do what you want.

And all of those rewards that I talked about earlier, that's where you do it in Isaac Sim. Brief pause. Well, you know, looks this up, figures out if she should be using it for work. Yeah, I think it's, yeah, Isaac Sim, I think, especially because, you know, Navidia is, like, leading the semiconductor industry right now.

Their software is very important. So Mecca is making quadrupedal robots, and you're looking at reinforcement learning for behaviors. Is this going to be another one of those military dogs? Definitely not. So we're doing the exact opposite of that. The goal is to make home robots in general. And we're starting with quadrupedal robots because we think that is the easiest form. to integrate into homes right now.

And so, yeah, like you mentioned, pretty much everyone who's making quadrupedo robots right now is targeting more industrial applications, more military applications. Spot, you know, from Boston Dynamics, that's an industrial robot. I think the only other company that's making quadrupedo robots in the U.S. is ghost robotics, and they're a military,

that's a military robot. And so, yeah, our goal is to make a friendly robot, something that, you know, is not going to be shift off to war, something that can be in your house and be sort of this all-in-one smart device for your home, something that, you know, you can go on walks with, something that you can ask questions with, that can be a personal AI assistant, something that, you know, can help carry in your groceries, something that you can play with with your kids. We want something that feels real, something that is expressive, something that is emotive. And yeah, that's the goal. We want a cool home robot because we really don't have any robots out in the world that are doing that, that can be in the home right now. A lot of people are focusing on humanoid. And that's been the big buzz in the robotics world. but I don't know, humanoid's just aren't quite ready for the home yet.

They're a huge safety hazard. These things are like 150 plus pounds. Yeah. And, you know, what people are promising that they're going to do is just realistically, like another five to 10 years away. There's no humanoid right now today that can go into any home and, you know, do a task like, you know, wash your dishes or do your laundry non-autonomously. or autonomously. There's no humanoid that can do that today. And so we think we'll eventually get to that point,

but there has to be something before that, something before the humanoid. And we think the quadruped is the perfect form factor for that. I like the things you listed, carry your groceries, play with your kids. Those are two different weight classes, though. Carry in your groceries, it has to be massive enough to withstand my groceries. But playing with kids, I don't want it to be more than 20 pounds at that.

Yeah. Yeah. Great point. And so that's sort of the goal weight we're targeting as well, around like 20-ish pounds. And yeah, with a 20-pound robot, you can carry some groceries. Maybe not like all of the, like probably is not going to be able to carry like 20 pounds of groceries, like double its body weight or the same amount as his body.

body weight, but you could probably carry like 10 pounds of groceries, especially if you're not doing it for super long, super long distances. You can carry some amount. But yeah, that's very important for us. Like safety, we don't want this thing to be something that can harm you if it falls over on you. It's low to the ground, which is good. But yeah, we want it to be low weight, something that you could pick up or a kid could pick up if you had to. I spend quite a lot of my time trying to figure out how to make sure my robot will not fall and kill somebody. It's very distressing. So it's very top of mind for me. Do you, is it a toy or a tool? I'm just going to add that to all of my questions now.

Is it a toy or is it a tool? But it came up recently on a different show. But I think it's relevant for you. Is it for fun or for usefulness? Yeah, that's a great question. The goal is both. We want this to be something that can be used in research labs for, for, you know, AI and reinforcement learning, but we also want this to be something that can live in the home. So we're sort of targeting that gap because right now there's just a huge divide.

Like you either have, like I think you mentioned the Sony Ibo, earlier. So that's like a pure toy, something that can't really do anything other than like sit on your desk and do a couple of cool pre-programmed actions. And then you have something like Spot or like the unit you go-to, which is like just an industry tool. And I think the intersection between those is something really cool, something that can serve very useful, practical applications like being an AI assistant, being something that could even be like a security device to monitor your house,

but could also be something that you can go out and play with and go to the field with and run with you on runs and play with your kids. And so we think that we can target both of those, both ends of that spectrum, because the goal is to essentially make a platform that developers can create apps for the robot. So we want this thing to be something that you can just use for so many applications. And all those applications will just live in an app store.

And so maybe you want an app that can have the robot, you know, go and check up on the grandma and see how she's doing upstairs. Or maybe you have an app that can follow you around and record, you know, your time. TikTok or whatnot. And maybe there's an app that can, you know, go on and run with you. And so we want to target all of those applications through an app store. I like that idea. But from an entrepreneurial perspective, I don't think you can keep it that open forever and you will have to make decisions. And making it a platform, the research, there are a couple other robotics platforms for research institutions. And so there is a market there, not a huge one, but at least it gets other people to have time on your robot.

I like the idea of a pet most because that's what I would like most. Yeah. But it can't, like, I'm thinking like, okay, what would I pay $2,000 for? Not that I have any idea how much you're aiming for, but that seems like a not impossible price. Yeah. Like, as a pet sitter for our dog to be gone all day, but then it has to open the door and that's kind of hard, so maybe not that.

Wouldn't take it to the beach. Probably wouldn't take it on walks because we have an actual dog, and she would get jealous. I don't know. I mean, the things he's mentioned, like security, having it wandered to the house while you're gone and be aware of things or to be able to be able to. Yeah. So I think

Fetch me water. Yeah. So like the Sony Iibo, I think sold like 150,000 units like the first two years. Yeah. It's so expensive, but it was so cute. That's a lot of units. Well, and it was programmable. Yeah.

Yeah, it was like, yeah, it did extremely well. And again, this was like in the early 2000s. Right. So, you know, like accounting for inflation, I don't even know how much that would cost today. But like people were. were super excited about it. The reason it failed is because it just wasn't feature rich. So, like, people would play with it for five seconds and then they'd exhaust all the features. And, you know, there's only so many times that you would want your Sony Ivo to, like, bow or, I don't even know

if it could roll over, but like bow or, like, do whatever pre-programmed sequence it could do. And so our goal is like, okay, you know, there's interest. People will pay, you know, a pretty penny for a cool robot. But how do we get people to keep that robot and stay interested in what we're doing? And it just has to be feature rich enough. So that's why the App Store idea came about. It's like, we want a place where you can just do just a bunch of different things. And we want you to be able to use it for very practical things, but also just use it for fun, playful things. And so I think that's really the only way right now that you can get people to want to adopt robots into their homes is if it does just a whole lot of things. And with the platform, you give people the opportunity to make their own applications.

Exactly. Although there's only a small percentage of the population who is able to do that. Yes. And so it's funny you mention that because we are trying to close that gap. Like, we want this to be easy enough where just anyone can, you know, create a cool application for whatever their specific use case is. And so the goal is to, you know, make an SDK with high-level APIs where it's just as simple as, like, you know, typing in natural language for the robot to go do something and having it go do that. So we want it to be very simple to make apps for developers and make commands.

for end users. Okay, I have an app idea. Christopher, you have to think of an app idea while I explain this one. All right. Great. Let's hear it. Our dog sometimes does hysterically cute things. It's just who she is. And I have wondered, could we outfit the house with cameras so that we could have her own YouTube

channel? This is never actually going to happen. It's just, you know, fantasy. But if I had a robot that could just follow her around. around and tag whenever she is moving her toys from one place to another because they're in the sun or whatever. Yeah. She could, yeah.

I could. And then building on that, instead of just follow my dog around, you could have follow my toddler around. Yeah. Alert me if she leaves this room. So the follow somebody around would be pretty cool. Yeah. Okay, Chris, now your app idea.

Well, I like the idea of having something carry a bunch of tools and stuff, from wandering around the house doing stuff. So it would be a... Oh, but he already has a mobile toolbox dog. You saw that in the video, right? I don't know, but I did see a caro that was a toolbox, which was hilarious. It was so cute. Did that toolbox actually carry anything, or was it all full of batteries?

Wait, which robot are you talking about? In one of the Kara videos, it looked like it was built into like a, I want to say a Home Depot toolbox, but a standard toolbox shape. Yeah, I think that's just how the robot looks. It has a handle and it looks at its toolbox like. Yeah, it's definitely not a toolbox, but I'm sure it can carry like 10 pounds. I tried it with, or I think it was 15 actually. No, yeah, it was like 15 pounds.

I don't know how long it could carry 15 pounds for. Didn't test it for long durations. But yeah, it could definitely carry, you know, a couple tools for you and follow you. fall year round. It just, Carr didn't have any vision, so the fall year around part would probably be a little, a little hard. Are you familiar with OpenMV from Kwabana Adjerman?

I'm not. No. So he has this amazing board that's a python, micropython. And it does all kinds of intelligence with recognizing faces. recognizing specific faces.

Objects, people. And so it's a whole camera intelligence system built into one thing. It's really cool. Yeah, we're sort of trying to figure out the right perception stack right now. So, you know, you have a lot of different options. The robots, LIDAR is, of course, like the main one, the go-to for, like, like being able to figure out, or to be able to, like, sense, you know, maybe obstacles on the ground and step over them, sense the terrain. But it's so expensive.

Yeah, exactly. So that's the thing. Cameras are so cheap. Yeah. Yeah. You can get, like, four cameras for the price of a LIDAR. They have small, small LIDARs now. But yeah, they are expensive.

Yeah. So cameras are great with, you know, maybe recognizing phases and stuff. but like using a camera for depth perception and obstacle avoidance is harder, but I think if you can do it, it's a lot more rewarding because that's what, you know, the Tesla's do. And, you know, as you see with the cyber cab, it's like a fraction of the cost of like a Waymo. You had one camera on each ear or eye. I guess you don't have a head.

So on each shoulder joint area. It might have a head eventually. Yeah, we will have a head. for what we're doing with the startup. Oh, good. I'm glad. I don't want to say that Kara is creepy, but... No, you can say it.

Yeah, I think there's definitely an uncanniness with quadrupedal robots that don't have heads. Yeah, and I'm definitely aware of that. Like, yeah, I'm a very harsh critic of my work. And Kara, for example, like, even the legs, they're five-bar linkages, which aren't biological at all. But, yeah, there's... There's definitely sort of this uncanniness with something that with a quadrupedal robot that doesn't have a head.

And so that's something we want to do is like we want this to look like a dog, like really look like a dog. Like a lot of people say robot dogs and they have just this quadruped with no head and the same front and hind legs. But we want something that actually looks like a dog. So like proper front and hind legs that like actually look like a. real dogs, front and hind legs and an actual head, you know, a curved body, not something that is like a toolbox or, you know, that's just very boxy and very simple. We want something that looks intricate and biological. What about a tail? And would it be just a tail that goes back and forth maybe, or would you go ahead and make it like a full tentacle that it could grip things?

Prehensal tail or regular tail? So yeah, the tail is something we're also sort of figuring out now. I think it'd be cool if we had a tail that can actuate. I don't know if we'd be able to have it be like a full manipulator to like pick up stuff. But I think a tail that could I just at the very least wag and then move a little bit would be really, really cool. It'd be like a small detail that I think would add to like the user experience. Between the tail and a heartbeat, if you had some small vibra motor that gave a heartburn. breed indicator that you could feel or a per indicator if you decide a cat you might not need a head I think we'll have a head no matter what just because I think it'll just look really cool with the

head because no one else adds heads to their quadrupedal robots so I think that would be really cool but yeah all this other stuff you mentioned I think would also be really cool like a heartbeat that's not something I've even thought about but you know if you can like touch the robot and like actually feel its heartbeat, like, you know, a thud. Like, I think that would be really cool. And it's one of those things that, like, you know, that you would, you would assume that, like, Apple would do if they made, like, a robot. Like, these small, like, intricate details that you were like, wow, that's really

not necessary, but I'm glad it's there, you know? Yeah, I've put heartbeats in a few small, noncommercial things. and it's really nice. It takes it from something that is clearly inanimate to something that borders uncuddleable. Oh, wow. I have so many ideas, but we're pretty much at a time. Chris, do you have any other questions you want to?

No, no. All right. It has been great talking to you. Aide. Do you have any thoughts you'd like to leave us with? Yeah, yeah. My one thought, is that real? robots are the future. I think we're going to see a very interesting wave of physical AI coming up next.

We've seen the AI wave and we're still in it. But the next step is physical embodiment. And so that's kind of why I'm doing the startup now. And I think the future is going to be very, very interesting. I think we're at a time for robotics similar to the early days of computers. And we're going to just see robots everywhere in the next five to ten. years. So brace yourselves. And yeah, it's going to be a fun time. I think, I think robots are really going to change the world. You're telling me, I really should read that AI and embedded from Open Starge book and be prepared to start writing that. Yes. Yeah, watch all the sci-fi movies that you can. And it won't be that terrifying. But I think robots will help with our day-to-day lines. We didn't ask one of the lightning round questions. Murder bot or Iron Giant?

Iron Giant. 100%. Yeah. No murder bots. Our guest has been Aide Moussa, co-founder of Maka and YouTube creator. Check out his YouTube channel at A-E-D-M-U-S-A or check the link in the show notes. Thanks, A-E-D. It was really great to talk to you. Thanks. Yeah, it's great talking to you guys as well. Thank you to Christopher for producing and co-hosting. Thank you to our Patreon listener Slack group for all the weird things that have happened in the last two weeks. And thank you for listening.

You can always contact us at show atembedded.fm. Or hit the contact link on the Embedded FM website where you can also find show notes, transcripts, and assorted other things. And now a quote to leave you with. This one is from Miles Morales, one of the Spider-Man. Everyone keeps telling me how my story is supposed to go. Nah, I'm going to do my own thing.