Hi Hacksters,
I'm RzLi. I started the journey with OpenCat here on Hackster.io, and the Nybble cat and the Bittle dog grew out of it. My third product, Quaddle, is now on Kickstarter from $99. So I'm posting again to close the trilogy.
10 years ago I built that 14-DoF (degrees of freedom) OpenCat prototype in my university dorm room. Since the Nybble campaign in 2018, we've shipped more than 30, 000 quadrupeds.
I want to walk through:
- What the difference really is between a demo and a delivered product
- How those units are actually used, and what they did to our support desk
- A simple formula I ended up deriving, plotting a product's feasibility against its active DoF
- How that formula turned into a robot with only 4 servos
Running through all of it is the one question I have not been able to close in 10 years — I'll call it the question from here on:
Once you can build a robot, how do you make the work meaningful for others, so that you can get the support to keep creating? Is it easier or harder to start a robotics career as a maker in 2026?
2016: the dorm roomMaking a quadruped animal — not a spider-like crawler — has always been hard. The first OpenCat prototype had 14 DoF, bionic tendon-linked legs with 3 DoF each, a pan-tilt cat head carrying cameras, three Time-of-Flight rangers, a microphone and a speaker, a skeleton with 8 touchpads along the ribs, paw buttons, an Arduino and a Raspberry Pi, AI conversation, and — the most challenging part — lifelike walking with auto-balancing and self-righting.
The BOM was cheap, around $300. Time was not. I taught part-time at Wake Forest University, so I could work on OpenCat full-time. That's when the question first came up, and I took a few boot camps in 2017 to learn how to translate the technology into business words. Petoi LLC was founded.
2018: the video worked, maybe too well.I posted the first OpenCat video after attending CES. It went viral and brought a wave of opportunities, and later that year the Nybble cat (11 DoF) was successfully crowdfunded. The Bittle dog (9 DoF) followed in 2020. COVID landed right on top of that second campaign and squeezed production, cost, and the classroom scenarios those robots were built for. Still, as one of the earliest robot dog makers, I was lucky: I could get attention simply by posting a robot that walked and did tricks.
Attention was never the goal, though. Making the work meaningful was. So the question became harder: how do you even define meaningful?
Look at robotics in the past few years, and the most-delivered products in the field have been videos. People have started complaining that robots only dance or run fast and aren't doing their jobs. I don't think that's entirely fair. If you see many robots as a modern form of puppetry, it makes more sense — ritual and spectacle have long been part of the job. Under pressure to survive, each embodiment eventually finds the path closest to what it does best. And the robot sports we watch serve the same purpose as the Olympics: showing what a body can achieve within fixed physical limits. The world records don't necessarily improve the general public's health. More makers than ever can build a complete robot and demonstrate full-stack skill. They don't owe anyone a physical product that works in every household. Inspiration alone is worth a lot.
But inspiration may not be sustainable. A high-DoF robot creates an emotional connection: more joints, more possibilities, more impressive — and a higher bar for the next person, which in turn lowers the odds that any of it ships as a reliable product. Audiences and influencers get tired of endless demo reels, and they are far less likely to share an ordinary-looking robot whose improvements are real but subtle. For new creators, that leaves diluted attention in a short-video feed—far from enough to push a prototype into a product. AI-generated video can now produce visual miracles in parallel, at zero marginal cost. If robotics wants to be something other than a branch of the video industry, it has to deliver real things. Funding is also getting impatient, so robots increasingly have to earn their living from customers paying out of their own pockets.
For commercial robots, another, quieter signal appears a few months after delivery: compare the second-hand price of an impressive robot with its BOM cost. Neither number captures the product's full value, but the relationship between them can reveal how much practical value the market still sees after the novelty wears off.
The moment you start talking about delivering hardware products and customer service, all the stories and promises slow down.
I've tried to keep our business boring and ordinary: BOM- and channel-based pricing, real applications instead of teasers, community building, prompt customer service, loose replacement and return policies (30-day return). What I didn't expect is that keeping the system running costs roughly 10 times what it cost me as a solo maker to build a prototype and post a video.
Here are the numbers that reshaped my thinking. Bittle has 31 screws — 3×9 for the joints, 4 for the board — and we can assemble one in 30 minutes. We made hours of tutorial videos and a detailed doc, improving them after every failed edge case. However, users, myself included, rarely open a manual until something goes wrong. With a complex, feature-rich robot, simply getting it running and trying every feature can become a long test of the user's confidence and patience.
Many users shared videos of its default motions on social media; most teachers reached out to improve their curricula; a few contributed to the repository and the forum; and users published about two dozen papers. A few came to us a few years after they bought the robots because they finally got the time to build. For most users, no support request usually means the basic experience went smoothly. But it also means we rarely learn whether they went on to explore it in depth.
When users are dissatisfied or run into trouble, however, they usually contact us. Problems can surface at almost any stage of assembly, and sometimes during play with the pre-assembled version. Each case takes us 15 to 60 minutes to resolve over email or a remote session, and some also require international shipping for replacements or returns. I read 100% of the support emails, and handle 1/4 of them directly, whether it's from an 8- or 92-year-old. We do it out of responsibility, out of a genuine wish to fix the problem, and yes, partly out of fear of returns. We still serve customers who bought our oldest kit in 2018.
Plenty of our users do exploit every single DoF, for research, teaching, and their own projects. They're wonderful. They're also a minority of 30, 000. A considerable share are novices who buy because these robots are less intimidating than traditional ones. That also explains why support gets harder once you try to enter the household. You have to be clever enough to impress the smartest people; you must be humble enough to serve the slowest ones.
I have a lot of respect for the more complex open-source robots out there, and this isn't a complaint. I just know from our own numbers what it takes to get even a 31-screw robot working in 30, 000 pairs of hands. If a cool project can only be reproduced by a few hundred people out of 8 billion, its value comes back to inspiration. And from the outside you can't tell the difference: a well-edited demo of a robust design and a well-edited demo of a fragile one look the same. That isn't cheating. It's just that video carries no reliability information.
This isn't a gap in users' knowledge either. It's the oldest question, wearing a robot costume: product–market fit. We're blessed with advanced chips and AI, and we have very few affordable chances to validate which combinations actually work. Filtering costs money and time, and above all, calling things as they are.
A formula: feasibility vs. active DoFOur quadrupeds never scaled the way the team's effort deserved — I still sleep at the office — and I wanted to know why. Similar products from other companies hit the same problem once they actually have to deliver. So I worked out a simple feasibility formula, with active DoF as the main variable. The data points come from my own data and from public sources. I'll state the assumptions openly for your feedback.
The curve decays clearly as DoF rises. When I plot modern products across fields on it, their relative popularity lines up reasonably well with what actually happened in the market. The formula also says something less obvious: getting more functions out of the same DoF improves feasibility far more than adding DoF does. Seen through that lens, the fast rise of Bambu-style 3D printers compared with traditional ones makes sense, and so does the strange quiet that followed the drone boom.
I'd rather you poke holes in this than take it on faith. The full write-up and an interactive calculator are here. I'm not a mathematician. You can participate and improve the source code on GitHub with your own version of the formula. Tell me where it breaks.
I keep comparing introducing a new robot product to life crawling out of the sea. It has to develop a locomotion pattern before it can do anything else, and then it has to earn energy in an ecosystem that was not designed for it. It should also start small, fighting for its living. Selection, not ambition, decides the direction of evolution. Selection by users or by capital will shape different monsters.
That comparison turned out to be literal. The core of Quaddle is a bionic structure that reproduces life's first legs on land — primitive and tendon-linked, not well-muscled. I call it the MinDoF (Minimal DoF) leg. I also added wheels at the toe tips, allowing it to glide. It took me two years to study the mechanism and make injection molds to test its consistency.
The idea came straight from the formula. People love the lifelike motion of a quadruped, but few ever directly manipulate how joints collaborate. Most tinkering happens a level up: calling motion APIs as if the robot were a wheeled vehicle, or going through black-box algorithms—which pushes users further from the hardware, into what we've decided to call physical AI. Abundant DoF are the source of that emotional connection, and also of the cost, the cognitive load, and what eventually breaks.
It's like a vase that's too deep and too heavy to lift: the water is right there, but you can't just drink from it. Inside the vase neck, the path is a maze. We spent years designing straws that fit the maze, and each straw became one more thing for users to learn. A few hundred people may get very good at it. However, robotics doesn't enter households at that scale.
Quaddle is designed as a cup: you can pick it up and pour, and the deeper contents are still there after you get a taste. The MinDoF leg halves the number of servos and the power budget, while preserving — and in some motions exceeding — what our own 8-DoF Bittle can do.
What Quaddle can doJust as Mecanum wheels enable full omnidirectional motion with 4 motors, Quaddle proves just 4 servos can unlock complete quadruped agility.
You can feed the specifications below to an AI and compare them with traditional quadrupeds (all measurements were taken on pre-production units on a hard tabletop):
Petoi Quaddle
- 11 cm long, 170 g, 4 servos, 4 screws, 15 minutes from parts to walking
- Walking at 1 BL(BodyLength)/sec, with blended gaits filling the range up to gliding at 3 BL/s on the toe wheels, on a smooth surface
- Turning a full 90° within a single gait cycle
- Walking sideways at 1/6 BL/s, with varying performance on different surfaces
- Walking on three legs at 2/3 BL/s
- Continuous back flips and front flips
- Walking while carrying its own body weight
- Climbing onto a step as tall as its own knee
- With magnets on the toes, walking upside down under a steel plate
- A swappable Nokia-format battery: it takes a standard BL-5C, and we upgraded it to a BL-10C (1800 mAh / 3.7 V, 6.7 Wh) — good for about 2 hours of load-free continuous stepping.
Walking used to account for most of a robot's complexity and cost. With that part cheap and simple, we can focus on what the robot is for and how it feels to use — which, in the end, is the only reason anyone pays out of pocket.
Everything that defined OpenCat — Arduino, Raspberry Pi, Python, 3D-printable parts — is here at roughly half the price of our previous models, plus a few things that weren't. You can move the legs and drag them to record a motion; you can teach a cute trick or a fast gait that way in about 10 seconds. It has a color display for system info and vivid expressions, is LEGO-compatible, and supports LLM conversation, interactive behaviors, and smart-home control.
Our own APIs let you vibe-code a new behavior without ever opening the firmware. It goes further: I recently added mood expressions straight into the color-display firmware that way in about 15 minutes, no deep embedded background needed even at that level.
A more complex demo using the web camera to project posture to the robot took only two hours.
Realistically, Quaddle is a tinkerable toy, a teaching tool, and a research platform, and a natural place to start before stepping up to Bittle's 8 servos. I know "toy" sounds like a smaller claim than "real work," and I'd rather make the honest claim than the impressive one. Robotics has attracted billions in investment, with household applications remaining one of its most common — and compelling — visions. Yet families keep being told to wait another three to five years. I'd rather shorten that wait than keep describing it. A cheap robot that actually sits in your house does more for that future than another viral video.
Someone replied on X calling Quaddle "the Arduino moment for robotics" — the missing middle between a toy that does a few canned tricks and a research platform that assumes you already know mechanics, control systems, and ROS. I didn't ask for that comparison, but I think about it often: Arduino didn't matter because blinking an LED was impressive. It mattered because the first success came fast, then invited people into harder projects.
The question is still openThat's what 10 years looked like from where I stand, and the question I started with hasn't moved:
To make our intelligence and hard work meaningful — valuable enough that somebody funds the next round of it — do we keep building inspirational visual miracles, or do we build simple, affordable robots and get them onto as many desks as possible?
I still don't have a clean answer to that. I feel it's harder to start a robotics career today as a maker who builds a robot and wants to improve it simply by selling it. The outcome of this campaign will show whether there is enough support for the harder path: turning prototypes into products, then delivering and supporting them for years. With the OpenCat framework, I could choose another path — quickly building fresh, interesting prototypes, attracting attention through demos, and sustaining the work through sponsorships, partnerships, or other forms of support.
There's a related debate going on right now about which frontier model or world model will produce robotics' "GPT moment." I'll leave that race to people better positioned to call it. What I know is a narrower, separate problem: almost nobody can actually touch a real robot outside a toy aisle, no matter which model eventually wins. That's an accessibility problem, not a capability one, and it's the one I've spent these 10 years trying to solve. I'd rather there already be something on a desk people can pick up and try for themselves today, so they can find out for themselves what physical AI is actually good for.
If you're a maker or engineer who can already build a robot, standing at that same crossroads：
- Chasing the next viral video
- Getting hired by a robotics giant
- Delivering a real product
Where would you put your next 1,000 working hours? I'll be in the comments.
~
The story text and figures in this post are licensed CC BY-NC-SA 4.0. Quaddle's firmware, API, and 3D-printable parts will stay open source at github.com/PetoiCamp under their own repository licenses.