[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fbww8g9ri7pcc":3},{"code":4,"message":5,"data":6,"timestamp":34},10000,"success",{"full_name":7,"position":8,"description":9,"language":10,"stars":11,"forks":12,"stars_gained":13,"age_days":14,"hn_threads":15,"one_liner":22,"summary_md":23,"topic":24,"github_topics":27,"license":28,"homepage":29,"updated_at":30,"history":31},"apirrone\u002FOpen_Duck_Mini",6,"Making a mini version of the BDX droid.https:\u002F\u002Fdiscord.gg\u002FUtJZsgfQGe","Python",4486,568,44,892,[16],{"id":17,"title":18,"points":19,"comments":20,"posted_at":21},43334591,"Open Duck Mini",2,0,"2025-03-11","Open hardware plans, 3D files and reinforcement-learning code for building and training a small bipedal duck robot at home.","This is a build-and-research hub, not a kit. If you want a roughly 42 cm bipedal robot that walks, expect to print parts, source motors and boards, then spend most of your effort on simulation and tuning rather than unpacking a product.\n\n## The problem\n\nA hobbyist who wants a legged robot usually starts from a commercial quadruped or a pile of loose parts with no shared reference design. Here the intended end state is a working duck: mechanical design, bill of materials, assembly guidance and walking policies that can run on an onboard computer. The friction is that no single step is turnkey — you move from CAD to printed parts to a MuJoCo model to a controller. The project changes that by keeping those artifacts in one place and linking out to companion repositories for the runtime and training, so the result is a reproducible starting point rather than a one-off build log.\n\n## What's different\n\nThe mechanical design is published in Onshape, and the README says it is \"pretty much finalized\" apart from planned expression features such as eye LEDs, camera, speaker and microphone. The BOM targets under $400 and is available in English and Chinese. Assembly and print guidance are separate documents, though the assembly guide is explicitly marked incomplete and the CAD-to-simulation preparation document is marked outdated. That matters: the design intent is shared, but the documentation is uneven and the reader must judge which pages still match v2.\n\nThe walking side is reinforcement learning. The repo ships two ONNX walking policies you can load into the MuJoCo script, and training has moved to the separate Open_Duck_Playground project, with reference motion generation in another companion repo and actuator identification done with Rhoban's BAM. So the interesting mechanism is a sim2real loop: train or load a policy, run it in simulation, then deploy it to the onboard Raspberry Pi Zero 2W runtime. Be aware that the README calls this a working repo with \"a lot of undocumented scripts,\" so expect to read code and ask in the Discord rather than follow a polished pipeline.\n\n## Who should try it\n\nThis fits someone with access to a 3D printer, comfort sourcing motors and electronics, and an interest in RL or legged locomotion who wants a project to iterate on. It also suits a small group or lab that can divide printing, assembly and simulation work.\n\nIt is a poor fit if you want a finished walking toy or a beginner soldering project with a guaranteed first-walk path. The incomplete assembly guide, outdated simulation-prep document, and external training repository mean the project is best treated as a base to customize. People who only want to run inference can try the provided ONNX policies in MuJoCo without building hardware, but that is simulation, not a robot.\n\n## Requirements and limitations\n\nBudget and fabrication are the first constraints. The under-$400 BOM is a target, not a quote; sourcing, shipping and failed prints can change the real cost. You need a printer and the ability to assemble a small legged robot, and the assembly guide may not cover every step. The onboard runtime assumes a Raspberry Pi Zero 2W, and training lives in a separate repository, so the project is split across several codebases with different levels of documentation.\n\nOngoing upkeep is real. The design is described as still being fixed \"here and there,\" documentation is written as the project goes, and sim2real is presented as active work. Expect to track changes and rely on the community channel for help. Verify that the CAD revision, printed parts and the policy you intend to run correspond to v2 before committing to a full build.\n\n## HotRepo's take\n\nHotRepo has only one dated observation: 4,475 stars and +44 stars on 2026-10-25, a single appearance on the daily board. That is enough to show a snapshot of interest, but not to judge whether it persists; the repository is about 891 days old and remains a working project. Before investing in a build, check that the assembly guide and sim2real document match the v2 design and the runtime repository you plan to use.\n\n[Project source and documentation](https:\u002F\u002Fgithub.com\u002Fapirrone\u002FOpen_Duck_Mini)",{"slug":25,"name":26},"hardware","Hardware & IoT",[],"Apache-2.0",null,"2026-10-25T11:03:29Z",[32],{"date":33,"stars":11,"stars_gained":13},"2026-10-25",1792968015]