Open-source Bionic Bipedal Robot Deep customization supported Apache-2.0 open source

OpenDuck Mini

v2 · An open-source bipedal robot replicating Disney BDX

A charming ~42 cm bipedal robot with 14+2 whole-body joints, walking via Raspberry Pi reinforcement learning. All 35+ 3D-printed parts are open source —Appearance, colors, and functions — all defined by you。

3D-printed appearance customization Fully open-source and secondary-development friendly Gamepad-controlled walking Facial expression interaction system
Customization Inquiry Open-source Repository
42cm
Overall height
14+2
Whole-body DOF
35+
3D-printed parts
100%
Open-source & customizable
WHY OPENDUCK MINI

More than just a robot,it is a customizable open-source platform.

From shell color to control algorithms, from teaching materials to research experiments — every layer of OpenDuck Mini is open to developers.

Fully customizable appearance

35+ 3D-printed structural parts (PLA/TPU) — colors, finishes, and shell shapes can all be customized to create a one-of-a-kind campus IP robot.

Fully open-source

Mechanical CAD, BOM, on-board software, simulation environment, and RL policies are all open source (Apache-2.0), safe for commercial use.

Free functional expansion

Reserved camera, microphone, LED expression, and projection interfaces let you add sensors and voice interaction for research and teaching.

Real-robot RL deployment

MuJoCo simulation + ready-to-use RL walking policies, with imitation learning for training your own gait — the full Sim2Real pipeline.

TECHNICAL SPECS

Specifications

OpenDuck Mini v2 core specifications at a glance

Overall height~42 cm (legs extended)
Robot configurationBipedal bionic robot (inspired by Disney BDX)
Drive system14× Feetech STS3215 bus servos + 2× SG90 micro servos (antennae)
Joint layout5 DOF per leg (hip yaw/roll/pitch, knee, ankle); head pitch/yaw/roll + neck pitch; 2 antennae
Main controllerRaspberry Pi Zero 2W（4×Cortex-A53 @1GHz，512MB RAM）
SensorsBNO055 9-axis IMU, foot contact switches, camera, microphone
Power2S 18650 Li-ion battery pack (~7.4V) with BMS protection
Audio / displayMAX98357A I2S amplifier + speaker; dual-eye LED + projected expressions
Mechanical structure35+ 3D-printed parts (PLA structural / TPU flexible), M3 heat-set inserts and bearings
Control methodXbox controller (Bluetooth) with start/stop, acceleration, antennae, and random sound effects
Open-source licenseApache-2.0 (commercial use allowed)
ARCHITECTURE

A five-in-one technical architecture

Mechanics · actuation · perception · compute · software — every layer is open and modifiable

Mechanical structure

PLA structural parts + TPU flexible soles, complete Onshape CAD models, one-click reusable print parameters

Drive system

14 serial bus servos + 2 PWM antennae, joint soft-offset calibration tool, identifiable motor parameters

Perception system

BNO055 9-axis IMU attitude sensing, foot contact switches, camera and microphone for expression expansion

On-board compute

Raspberry Pi Zero 2W running ONNX RL policies, GPIO directly driving LED / projection / antennae / foot switches

Software ecosystem

On-board Python runtime + MuJoCo simulation + RL training pipeline, flexibly configured via duck_config.json

SOFTWARE & RL

From simulation to real-robot walking,one seamless pipeline.

MuJoCo simulation validation

Built-in URDF model and Playground environment with complete Onshape-to-simulation documentation — zero-barrier simulation.

Pre-trained walking policies out of the box

The repository ships an ONNX RL policy (BEST_WALK_ONNX) — it walks right out of the box with no training required.

Imitation learning + Sim2Real

Reference motion generation and training docs, plus BAM motor identification to improve sim-to-real transfer.

Plug-and-play controller

Xbox Bluetooth control: start/stop, random sounds, antennae, and speed — ready for demos and interactive scenarios.

python v2_rl_walk_mujoco.py 
# OpenDuck Mini Runtime
from runtime import Duck

duck = Duck(config='duck_config.json')
duck.calibrate()        # Joint soft-offset calibration
duck.load_policy('BEST_WALK_ONNX.onnx')
duck.walk()              # Reinforcement-learning gait
duck.blink_eyes(2)  # Expression interaction ✨
Runtime Python ONNX inference MuJoCo simulation
CUSTOMIZATION

Three customization options,one is right for you.

WWNT Robotics offers a full customization service from turnkey delivery to joint R&D.

Standard Edition

Turnkey, ready out of the box

Fully assembled delivery with pre-trained walking policy — controllable right out of the box, ideal for course demos and hobby use.

  • Fully assembled unit + battery power kit
  • Preloaded runtime and walking policy
  • Xbox controller + user guide
  • Free open-source materials and community support
Most Popular
Custom Appearance Edition

Campus IP customization

Shell customization incorporating school logos, mascots, and anniversary elements to create a campus-exclusive AI ambassador.

  • Shell color / finish / school-badge customization
  • Custom mascot head design
  • Anniversary / event-themed skins
  • Batch customization (class / club kits)
Deep Development Edition

Joint R&D customization

Joint R&D for university labs and research teams, with deep customization from hardware to algorithms.

  • Sensor / end-effector add-ons
  • Custom gait and motion training
  • Teaching materials and lab manuals
  • Joint papers and grant applications
SCENARIOS

From classroom to lab,everywhere.

University course teaching

A full-chain training platform spanning robotics, motion control, embedded systems, and reinforcement learning — from theory to real robot.

Research validation

A low-cost bipedal platform for rapid iteration in gait control, Sim2Real, and imitation learning.

Maker education

From assembly to secondary development, with open BOM and assembly guides that build hands-on and engineering skills.

Competitions and events

The interactive star of campus tech festivals, open days, and welcome ceremonies — customized to become a campus IP.

CURRICULUM

Companion Curriculum

From embodied intelligence concepts to MuJoCo simulation, 16 systematic lessons covering the full bipedal robot pipeline

This is one of our curriculum modules · View the full curriculum

Module 1: Course Introduction

  • 01

    Course overview and schedule

    Overview of course objectives, knowledge map, learning path, and assessment.

  • 02

    Synchronization software used in class

    Configure the dev environment, SSH remote connection, and file sync toolchain.

Module 2: Embodied Intelligence Fundamentals

  • 03

    Defining embodied intelligence

    From traditional AI to embodied intelligence: the closed loop of perception, decision, and action.

  • 04

    Control systems for embodied intelligence

    Layered control architecture: high-level task planning, mid-level motion control, low-level servo actuation.

  • 05

    Two approaches to robot motion control

    Model-based vs. learning-based control — comparison, use cases, and evolution.

Module 3: Bipedal Motion Control

  • 06

    Center-of-mass projection

    The geometric relationship among center of mass, support polygon, and static stability.

  • 07

    Introduction to the Capture Point

    The linear inverted pendulum model and the physical meaning and calculation of the capture point.

  • 08

    Weight transfer between support foot and body

    Single/double support transitions, COM trajectory planning, and stability criteria.

  • 09

    Forward and inverse kinematics

    DH parameter modeling, forward kinematics solving, and numerical inverse kinematics.

  • 10

    Swing-foot end-effector motion control

    Foot trajectory planning: constraints and interpolation across lift-off, swing, and touch-down phases.

  • 11

    Computing the swing-foot Capture Point

    Predicting swing-foot touch-down using COM velocity and position.

  • 12

    Analyzing the bipedal walking process

    From state machine to gait cycle: decomposing the full bipedal walking control flow.

  • 13

    A complete walkthrough of bipedal gait

    Engineering practice in parameter tuning, online debugging, and real-robot gait validation.

Module 4: MuJoCo Simulation

  • 14

    Introduction to MuJoCo

    MuJoCo physics engine features, URDF/MJCF models, and contact models.

  • 15

    Command-line installation of MuJoCo

    Installing MuJoCo on Linux/Raspberry Pi and binding Python with dependencies.

  • 16

    MuJoCo Hello World

    Load the OpenDuck Mini model, run the first simulation step, and visualize it.

Get the full syllabus and sample lessons

Let this robotbecome your own.

Whether a campus IP mascot, a teaching platform, or a research testbed, OpenDuck Mini can be customized to your vision.

Book a Customization Consultation Explore the Open-source Community