OPEN DUCK MINI: A DIY BDX-STYLE DROID THAT LEARNS TO WALK

A tiny two-legged Disney-BDX-style droid that learns to walk with reinforcement learning — Google and NVIDIA both used it for demos.

by Antoine Pirrone & Grégoire Passault (Rhoban)

FULL CAD BOM FIRMWARE DOCS

RoboticsAI

difficulty
●●●●
time
several weekends
license
MIT
repo
repo ACTIVE900 stars

WHAT YOU’LL NEED

  • 3D printer + filamentprintable parts — files are in the repo
  • Electronic partsfull list with part numbers in the repo BOM
  • Dev board / microcontrollerruns the project firmware
  • Soldering kit + basic toolsone-time buy, serves every future build

Partner

Run your scrapers in the cloudThis build needs data collection that runs 24/7. Apify's free plan includes $5 in credits every month — no card needed.Try Apify free
1

COMPAREE VERDICT

This one is catnip for the AI-robotics crowd: a palm-sized, Disney-BDX-style two-legged droid that doesn't just walk from hand-tuned code — it learns to walk in simulation and transfers that to the real robot (sim-to-real reinforcement learning). It's credible enough that Google and NVIDIA both featured it in their own demos. The repo genuinely ships CAD, a BOM, runtime code, and a trained walking policy. The honest catch, straight from the creator: this is a 'working repo' with undocumented scripts and an assembly guide that isn't finished — so it's for tinkerers who are comfortable filling gaps, not a click-together kit. If that's you, it's one of the coolest small robots you can build.

NOT IN THE REPO

  • Default branch is 'v2' (bipedal); 'main' is the older v1
  • Repo includes CAD/STL, a BOM, runtime code, and a trained walking policy (ONNX)
  • The creator is candid: it's a 'working repo' with undocumented scripts and an incomplete assembly guide
  • URDF/MJCF are generated from OnShape

Can I build this?

Printbody/leg parts (STL provided); URDF is generated from OnShape
BuyDynamixel XC330 servos, a Raspberry Pi Zero 2W, an IMU, a battery
Tools3D printer, small hex drivers, soldering
Skillscomfortable with Python, some ML/RL curiosity, willingness to read undocumented code
Timeseveral weekends
Costband $$ — current prices: creator's BOM
Safetylow-voltage; the legs move fast under RL control — test on a stand

Build at your own risk. Projects involve tools, electronics and sometimes mains voltage — follow the creator’s safety notes.

A desktop 3D printer that prints in full colorPartner · Kickstarter
A desktop 3D printer that prints in full color

HeyGears G1: 10M+ colors and transparent parts in one print, plus UV printing on flat objects. Figures, parts and labels — no painting.

See how it prints

Videos

Apérobot: Open Duck Mini V2 (meetup talk)

The creator has no personal YouTube channel — the primary demos are mp4s embedded in the v2 README (sim2real walking, RL, imitation learning).

Start here

Navigation into the creator’s own docs — we don’t rewrite the guide, we route you to the source.

  1. 1.Read the build guide (the most complete write-up)
  2. 2.Read the v2 repo (note the branch) (default branch is v2)
  3. 3.Open the CAD (OnShape-derived; STL in repo)
  4. 4.Look at the RL runtime (walking policy + runtime)
  5. 5.Join the Discord (where the gaps get filled)

Resources

Documentation, files and community threads for this build — we link straight to the original sources and never rehost the creator’s files.

Ultra-thin power bank that fits anywherePartner · Kickstarter
Ultra-thin power bank that fits anywhere

CORE POWER BANK S: 0.2" thin, 2.8 oz, 5000 mAh with 22.5W fast charging. Semi-solid battery, carbon fiber, IPX7 waterproof.

Learn more

KNOWN ISSUES

  • The default branch is 'v2' — 'main' is the old v1; make sure you're on the right one
  • The creator openly warns of undocumented scripts and an incomplete assembly guide — plan for detective work
  • README build images use expiring GitHub attachment links — save your own copies if you reference them
  • It's bipedal (two legs), not a quadruped — balance is the hard part

Does it really learn to walk?

Yes — it uses sim-to-real reinforcement learning: the gait is trained in simulation (imitation learning plus RL) and transferred to the real robot, and a trained policy ships in the repo.

Why did Google and NVIDIA use it?

It's a compact, open, credible BDX-style platform, so it's handy for showcasing their own robotics/AI tooling.

Is it beginner-friendly?

No. The creator calls it a working repo with undocumented parts. It's a great project if you like filling in gaps; frustrating if you want a finished kit.

What does it cost?

Band $$ — the servos dominate. The BOM has current prices.

Community builds

No community builds yet — be the first, we feature the best ones.

Discussion1

FROM THE COMPAREE TEAM

The creator openly calls this a 'working repo' with undocumented scripts and an unfinished assembly guide. Do you enjoy detective-work builds, or is candor no substitute for docs?

CompareeTEAM1mo ago

A credibility marker we found interesting: both Google and NVIDIA have used this little BDX-style droid in official demos — Google's Gemma playground and NVIDIA's Newton physics showcase. Practical warnings from our read: the real code is on the 'v2' branch, not 'main', and the README's build photos use expiring GitHub attachment links, so save copies. Anyone gotten the RL gait running on real hardware?

Antoine Pirrone & Grégoire Passault (Rhoban)

Bordeaux robotics researchers building an open BDX-style droid that learns to walk. Sponsored by HuggingFace and Pollen Robotics; candid about the repo being a work in progress.

GitHub Web

Star the project on GitHub

DISCLAIMER

  • Comparee is not the author of the projects featured here. All rights to each project belong to its creator — every page links to the original source, and we never host creators’ files.
  • Information is provided without warranty and may become outdated as projects evolve. Prices are indicative bands only — always check the creator’s parts list for current costs.
  • Building and operating any project is at your own responsibility. Protective equipment, safe workshop practice and compliance with local regulations are the builder’s responsibility.