Intellectually Curious
Intellectually Curious is a podcast by Mike Breault featuring AI-powered explorations across science, mathematics, philosophy, and personal growth. Each short-form episode is generated, refined, and published with the help of large language models—turning curiosity into an ongoing audio encyclopedia. Designed for anyone who loves learning, it offers quick dives into everything from combinatorics and cryptography to systems thinking and psychology.
Inspiration for this podcast:
"Muad'Dib learned rapidly because his first training was in how to learn. And the first lesson of all was the basic trust that he could learn. It's shocking to find how many people do not believe they can learn, and how many more believe learning to be difficult. Muad'Dib knew that every experience carries its lesson."
― Frank Herbert, Dune
Note: These podcasts were made with NotebookLM. AI can make mistakes. Please double-check any critical information.
Intellectually Curious
Microduck: A tiny biped robot you can teach new tricks
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Microduck is a 25 cm tall bipedal robot developed by Pollen Robotics, the robotics team at Hugging Face. Designed for both play and education, the robot features a 15-motor system, a grasping beak, and advanced sensors like LiDAR and a camera. A major focus is its open-source software stack, which utilizes reinforcement learning and a sim-to-real workflow to allow users to train and deploy new behaviors. This software architecture manages everything from Bluetooth connectivity and system updates to autonomous movements like walking, kicking, and skating. Available in four colors for an introductory price of $399, Microduck is scheduled for its first deliveries in late 2026.
Note: This podcast was AI-generated, and sometimes AI can make mistakes. Please double-check any critical information.
Sponsored by Embersilk LLC
I spent uh well, three entire weeks trying to teach my childhood dog, Buster, how to roll over. Three weeks of treats and coaxing and sheer frustration, only for him to just, you know, stare at me and drool.
SPEAKER_01Yeah, that sounds about right for a dog.
SPEAKER_00Right. And it's funny to think about that struggle when today we're looking at a companion you can literally program to do backflips in an afternoon.
SPEAKER_01And it uh it never even demands a treat. But actually, before we dive into how you teach this particular companion new tricks, we should probably mention that this deep dive is sponsored by Embersilk.
SPEAKER_00Oh, absolutely.
SPEAKER_01Yeah. So if you're trying to uncover where AI agents could make the most impact for your business or you know, your personal life, or if you need help with AI training, automation, integration, or software development, you should definitely check out Embersilk.com.
SPEAKER_00That is a great call because today we are exploring the Microduck. It's uh created by Pollen Robotics, which is the robotics team at Hugging Face.
SPEAKER_01Right, the open source folks.
SPEAKER_00Exactly. And it's this 25 centimeter tall, 800 gram open source bipedal robot. It costs about $399, and it waddles, it roller skates, and it learns. So the mission for you, our intellectually curious listener, is to figure out how this tiny device bridges that massive gap between virtual AI simulation and, well, messy physical robotics.
SPEAKER_01Yeah, and let's start with the physical unit itself. Out of the box, it packs 15 different motors, a lighter depth sensor, an onboard camera, and uh an articulated grasping beak.
SPEAKER_00Okay, I have to jump in here and ask, why engineer a waddling biped with a beak? I mean, when it walks, it genuinely looks like a toddler learning to take its first steps.
SPEAKER_01It really does.
SPEAKER_00Right. And I get that acute form factor lowers intimidation, but surely that deliberate inefficiency costs you battery life and processing power just to keep the thing upright, right? Like why not build a highly efficient rover with wheels and claws?
SPEAKER_01Well, because a rover with wheels is treated like a simple appliance, like a forklift, basically. The microduct's design is fundamentally about testing human AI collaboration.
SPEAKER_00Oh, so it's intentional.
SPEAKER_01Completely. The constant micro adjustments required to keep a biped upright aren't wasted processing power. They're the exact complex control challenges researchers actually want to solve. Plus, the fact that it occasionally falls over and pushes itself back up makes us way more forgiving and engaged with its learning process.
SPEAKER_00Aaron Powell That is fascinating. So the hardware's vulnerability is essentially a physical interface for the software.
SPEAKER_01Exactly. Which brings us to the brain. This runs on what's called a sim to reel workflow.
SPEAKER_00Okay, what does that look like?
SPEAKER_01Aaron Powell So anyone can load up MooJoko, which is this advanced virtual physics simulation, and use reinforcement learning to train a digital version of the duck to do a completely new movement.
SPEAKER_00So you just do it all on a computer first.
SPEAKER_01Right. And once the virtual duck figures it out, you deploy that exact software policy, which is basically its newly wired neural pathways straight to the physical duck's onboard AI accelerator.
SPEAKER_00I have to challenge the reality of that though. I mean, transferring a skill from a sterile, perfectly Newtonian simulation into the messy real world isn't as seamless as downloading a file, is it? Well, no, it's because, you know, my living room has slippery floors, uneven carpets, and gravity isn't perfectly uniform when a motor starts heating up.
SPEAKER_01Yeah, you're hitting on the biggest hurdle in robotics right now. But modern reinforcement learning solves this by applying random physical noise to the virtual environment.
SPEAKER_00Wait, what do you mean by random noise?
SPEAKER_01It's called domain randomization. Think of it like training an athlete by making them run on a treadmill that randomly changes speed, it randomly tilts, and occasionally shoots tennis balls at them.
SPEAKER_00That sounds terrifying.
SPEAKER_01It is for the virtual duck. But by the time they run on a normal track in the real world, it feels like easy mode. The duck's 50 hertz control loop expects imperfection. So a wrinkled rug is just another minor variable to correct for.
SPEAKER_00Wow. So you're essentially weaponizing failure in the simulation to build resilience in the real world. That makes perfect sense.
SPEAKER_01Exactly. But it's not just adapting to physical noise, it's also dynamically responding to us. The way it processes its surroundings is just fascinating.
SPEAKER_00Like the Laraman mode, right?
SPEAKER_01Yes. So it isn't just playing a pre-recorded sound file. It uses its built-in LIDAR to shoot out laser pulses, calculating the exact millimeter distance of your hand from its beak.
SPEAKER_00And then it changes the pitch based on that distance.
SPEAKER_01Exactly. It continuously maps that spatial data to a pitch frequency, essentially turning its environment into a dynamic synthesizer.
SPEAKER_00That is so cool. And it also runs a tiny audio classifier, constantly listening through its microphone. Like if it detects the specific acoustic signature of a finger scratching its plastic head, it actively coos in response.
SPEAKER_01It does. But what happens when you introduce another unpredictable element, like a second AI robot?
SPEAKER_00No, the chorale mode.
SPEAKER_01Right. If you have multiple ducks in a room, they detect each other's Bluetooth beacons, they sync their internal clocks through that signal, and we'll literally sing a synchronized musical chorale together.
SPEAKER_00They're basically forming a localized swarm to turn isolated learning into a shared tangible experience.
SPEAKER_01It represents a massive shift. I mean, we are moving robotics away from being a million-dollar hardware problem locked away in corporate labs to a highly accessible software problem.
SPEAKER_00Because if you can simulate a physical behavior, you can teach it.
SPEAKER_01Precisely. Yeah. And suddenly the barrier to entry is just a $399 open source toy sitting on your desk.
SPEAKER_00It's just an incredibly optimistic glimpse into the future. Everyday people can now learn, tinker, and physically solve problems with AI right in their homes.
SPEAKER_01It really is a bright new era for human AI collaboration.
SPEAKER_00Truly. Well, if you enjoyed this deep dive, please subscribe to the show. Hey, leave us a five-star review if you can. It really does help get the word out. Thanks for tuning in.
SPEAKER_01Thanks for listening, everyone.
SPEAKER_00But before we go, I want to leave you with one final thought. Imagine a near future where sharing a new physical behavior for your robot online is as common and easy as sharing a photo. If you could download any physical skill instantly, what unique trick would you teach your own robot? Hopefully it takes less than the three weeks it took me with Buster.