Most of the robots that matter to me did not come from a billion-dollar lab. They came from a handful of obsessive engineers, a video pitch shot in someone's living room, and a few thousand strangers on the internet deciding — collectively — that the thing was worth existing.
Kickstarter changed robotics in a way that almost nobody talks about. The big-name humanoids we cover on the rest of this site — Optimus, Iron, Atlas, Figure 03 — get the headlines and the venture capital. But the robots that have actually made it into hundreds of thousands of homes over the last decade, the ones that defined the entire desktop and companion robot category, almost all started life as a crowdfunding pitch. And that is not a coincidence. It is the only path that works for a category the big labs have, until very recently, refused to take seriously.
"Crowdfunding is not a financing mechanism. It is permission. The first thousand backers are not customers — they are the proof you bring to your first real investor that the product should exist." — every successful indie robotics founder, eventually
Why Kickstarter Works for Robots in a Way It Does Not for Most Hardware
Hardware on Kickstarter has a brutal reputation. Most physical-product campaigns either ship late, ship broken, or do not ship at all. Robotics, on paper, should be the worst category of all — it combines hardware complexity, software complexity, mechanical engineering, AI, and the unforgiving expectation that the thing has to feel alive when you take it out of the box.
And yet robotics is one of the few hardware categories where Kickstarter has produced genuine, lasting, category-defining products. Why? Because building a charming desktop robot is exactly the kind of project a small team can finish. The bill of materials is small. The component supply chain is mature — ESP32s, servos, OLED displays, IMUs, lithium cells are all commodity. The hard parts are the personality, the expressions, the firmware polish — and those are writing problems, not manufacturing problems.
A team of four people can ship a delightful $300 desktop robot. The same team cannot ship a delightful $30,000 humanoid. The economics of crowdfunding map perfectly onto exactly the shape of project that the indie robotics scene is best at producing.
The Indie Robotics Hall of Fame
Here are the campaigns I think defined what crowdfunded robotics could be — and why each one matters for the next wave that is just about to arrive.
The first credible attempt at a fully autonomous robotic cat that did not need a phone. Elephant Robotics used the campaign as launch capital and is now one of the most prolific desktop-robot manufacturers in the world.
When Anki collapsed, Vector died with it. Digital Dream Labs ran a Kickstarter to bring him back, raised nearly two million dollars in 30 days, and proved that community can resurrect a robot the original company gave up on.
Open-source quadrupeds you assemble yourself. Hundreds of universities and hobbyists adopted them as the cheapest possible platform for learning real-world locomotion algorithms. The hardware is genuinely hackable.
A small, fully open-source humanoid head designed for HuggingFace integration. The first crowdfunded robot whose pitch was, explicitly, "we ship the hardware and you write the AI."
A walking, dancing Arduino-based robot you 3D-print at home. It became the de facto educational platform for an entire generation of kids who wanted to build a robot before they could legally drive.
A super kawaii desktop AI companion built around the M5Stack CoreS3 (ESP32-S3, camera, mic, touchscreen) with a fully open hardware and software stack. The campaign we backed — and the reason this article exists.
Why We Backed Stack-chan
Out of every indie robot pre-order open right now, the one we put our own money on is Stack-chan. It is not the most ambitious project on this list. It does not walk. It does not roll. It is a tiny desktop companion with a face on a screen, two servos, and a head that turns to look at you. And we ordered one immediately.
Three reasons.
First, the hardware is honest. Stack-chan is built around the M5Stack CoreS3 — an ESP32-S3 development kit with a 240 MHz dual-core processor, 16 MB of Flash, 8 MB of PSRAM, Wi-Fi, BLE, a 2-inch touchscreen, a 0.3 MP camera, dual microphones, a 9-axis IMU, and a 1 W speaker. Add a 700 mAh battery, two feedback servos, twelve RGB LEDs, an IR transmitter/receiver and an NFC module, and you have a complete AI-companion platform that costs roughly the same as a mid-range mechanical keyboard. Nothing about the spec sheet is hidden behind marketing. You can read the schematic. You can buy the parts individually. You can replace anything you break.
Second, it is genuinely open source. The factory firmware ships with expressive faces, motions, the XiaoZhi AI agent, iOS app integration, video calls and device discovery — but every single layer is hackable. You can flash your own firmware via Arduino or UiFlow2. You can write your own personality. You can route the microphone to whatever speech-to-text service you trust, the camera to whatever vision model you want, and the speaker to whatever TTS voice feels right. The robot is, in the most literal sense, a body waiting for the brain you choose to give it.
Third — and this is the part that matters most for what we are doing here at RobotTesters — Stack-chan is the cheapest possible way to test the question that obsesses us: can a tiny, cheap, open-source robot deliver more genuine companionship than an expensive, closed-source one?
| Specification | Stack-chan (M5Stack) |
|---|---|
| Main controller | M5Stack CoreS3 (ESP32-S3, 240 MHz dual-core) |
| Memory | 16 MB Flash + 8 MB PSRAM |
| Display | 2.0" capacitive touchscreen, glass cover |
| Camera | 0.3 MP front-facing |
| Audio | Dual microphones + 1 W speaker |
| Motion | 2 feedback servos (360° horizontal, 90° vertical) |
| Sensors | 9-axis IMU, proximity sensor, NFC, IR Tx/Rx |
| LEDs | 12 × RGB (two rows) |
| Connectivity | Wi-Fi + BLE |
| Battery | 700 mAh, USB-C charging |
| Programmable via | Arduino, UiFlow2, native ESP-IDF |
| Software | Open-source firmware + factory XiaoZhi AI agent |
| Status | Pre-order — shipping in the coming weeks |
The Plan: We Will Write Our Own Code
Our Stack-chan unit is on its way. When it arrives — in the coming days, according to the M5Stack shipping queue — we are going to do something the factory firmware does not invite us to do: strip it down to bare metal and write our own software stack from scratch.
The factory firmware is genuinely good. The XiaoZhi agent is competent. The motions are cute. The iOS app works. But the whole point of buying an open-source robot is that you do not have to settle for the personality somebody else picked for you. So we are going to build:
Our own conversational backbone — routed through a modern foundation model of our choice, with memory of past conversations, awareness of the time of day, and an actual sense of who is talking to it. The factory agent is reactive; we want one that remembers.
A custom expressive engine — tying facial expressions and head motion directly to the emotional valence of what the model is saying, not to scripted triggers. A robot that looks down when it is unsure, tilts its head when it is curious, and brightens its LEDs when it is happy to see you.
Local-first wake word and speech-to-text — so the microphone audio never leaves the device unless the user explicitly asks something that needs the cloud. Privacy is a feature, not an afterthought.
A small ecosystem of MCP-style tools the robot can call on its own: reading our calendar, summarising the news in the morning, controlling smart-home devices, reminding us to drink water. The hardware can do all of this. The firmware just has to be written.
We will publish the code. We will document the build. We will say what works, what does not, and what we wish we had known before we started. If you backed Stack-chan too — or if you are sitting on the fence about it — we want our build to be useful to you.
The Bigger Argument: Open Source Is the Only Way This Industry Stays Honest
I have written before about why open-source software in robotics is not optional. Let me say it one more time, because Stack-chan is the perfect example of why it matters.
Every closed-source consumer robot of the last fifteen years has had the same end-of-life story. Anki Cozmo — bricked when the company folded, until the community reverse-engineered it. Jibo — sold for $899, killed by a server shutdown that turned a beloved family member into a paperweight. Kuri — discontinued mid-shipment, never updated again. Even Aibo, Sony's flagship, has spent decades oscillating between full support and total abandonment depending on whether the current Sony leadership feels like it.
Closed-source robotics treats your robot the way SaaS treats your data: it is yours until the company decides it is not. The hardware sits on your desk. The intelligence lives on a server you do not control. Pull the plug on the server and the robot dies, regardless of how many years of memories it has built up with you.
Open source breaks that contract in the buyer's favour. Stack-chan can outlive M5Stack as a company, because the firmware is on the device, the code is on GitHub, and the protocols are documented. If M5Stack vanished tomorrow — they will not, but if they did — the community would keep Stack-chan alive the way it kept Cozmo and Vector alive. That is not a hypothetical. That is what already happened the last two times the industry tried to sell us closed companion robots.
You should not buy a robot that can be killed by a press release. That is the single most important consumer-protection rule in this category, and almost nobody talks about it. Open hardware and open software are not a hobbyist preference. They are the only guarantee that the robot you brought home is still going to be there in five years.
What to Watch on Kickstarter Right Now
If you are convinced — or even half-convinced — that the indie crowdfunded route is where the most interesting robots are coming from, here is what we are tracking right now and will likely cover in the next few months:
The next generation of open-source humanoid heads — Pollen Robotics has already proved Reachy Mini works as a category. Two more campaigns are rumoured for the second half of 2026, both targeting a sub-$1,000 price point.
Educational quadrupeds under $500 — Petoi has competition coming. At least three independent teams are working on Bittle-class robot dogs with better servos and improved gait libraries. One of them is expected to launch on Kickstarter before summer.
Stack-chan accessories and forks — the M5Stack ecosystem has a long history of community-driven add-ons. Expect Stack-chan to spawn its own peripheral economy within months of shipping: docking stations, additional sensors, modular limbs, even mobile bases.
The Verdict (For Now)
The headlines belong to Optimus and Figure and Atlas. The Reddit threads belong to Unitree's latest backflip. The conference keynotes belong to whoever can demonstrate the most fluid manipulation in front of a live audience.
But the robots that are actually going to live in your house this year, the ones built by people you can email directly, the ones whose firmware you can read, the ones that will not die when their parent company pivots to enterprise — those are still coming out of Kickstarter. Stack-chan is the one we picked. Our unit ships any day now. The build log starts the moment it lands on the desk.
If you are tired of waiting for the billion-dollar humanoids to become affordable, and you want a robot that is honest about what it is and what it can do, the indie scene is where you should be looking. And if you decide to back something yourself, do me one favour: pick the one whose source code you can read. Everything important about robotics for the next decade is going to be decided by who controls the brain. Make sure the answer is you.