Project files

README.md

Machine-translated by Robotheca from the original in Chinese (opens in a new tab). License: MIT (opens in a new tab).

Keywords: ROBOCON, quadruped, quadrupedal wheel, bionic legged, reinforcement learning, RL, Mjlab, Mujoco, Odin1, sim2sim, sim2real

Open-source repository of the HYNova quadrupedal wheeled robot project from the HYNova team of Shandong Huayu Institute of Technology.

We are a newly established team starting from scratch. From the initial establishment of the club to the formation of the team, from the first-ever National First Prize in the college's electronic competition to the National First Prize in RC, we have accumulated practical experience in mechanical design, hardware, control, simulation, and reinforcement learning. The current repository fully open-sources the mechanical, hardware, firmware, software, BOM, assembly materials, deployment documentation, training and competition version code, and various precautions of this project. We hope to provide a reference for teams that started late and are currently preparing for Robocon.

Acknowledgments

We would like to express our gratitude to the Robocon Youth Advisory Group for their assistance and support. We also extend our heartfelt thanks to Lain (opens in a new tab) for the LocoWiki (opens in a new tab) project, which provided early reference and inspiration for our solutions. Special thanks to the teams from Harbin Institute of Technology, South China University of Technology, Dongguan University of Technology, Weifang University of Science and Technology, Shandong University of Science and Technology, Taiyuan Institute of Technology, Jinzhong College of Information, and others for their strong support and warm help during the preparation and competition periods.

Finally, we would like to give special thanks to Liuxing Technology for their exclusive sponsorship and technical support for the 2026 season.

Teams and Projects

Competition project: Bionic legged obstacle race
Final results: National first prize, 7th place, top 5%
Main platform: 16DOF serial wheeled-legged robot
Current repository version: 16DOF serial wheeled-legged physical robot (upgraded to 16DOF)

Version Status

Stage: 16DOF Final Competition Version
Platform: Jetson Orin Nano 8G
Configuration: 4 wheel-legs, each with 3 leg joints and 1 drive wheel
Actuators: 16 actuators in total (RS02)
Control Pipeline: Reinforcement learning training, MuJoCo simulation, Sim2Sim validation, dual-path deployment on real hardware via Python and ROS 2/C++
Current Status: Project materials fully open-sourced, including mechanical, hardware, firmware, software, BOM, assembly, training, deployment, and final competition code
Current Version: v1.1.1 Final consolidated edition, with complete open-source documentation and code comments added
Repository: Final complete open-source consolidated version; subsequent updates will only include necessary bug fixes and errata corrections

Open-source Content

This project continues to iterate along the technical route of "8DOF Basic Version → 16DOF Upgraded Version," and the open-source content is gradually organized and released in stages by version.

8DOF Basic Version

8DOF is the serial robot version during the mid-term, focusing on mechanical structure design, motor control, CAN communication, leg inverse kinematics, PID tuning, and basic gait control. It accumulates foundational experience for subsequent mechanical upgrades, simulation training, and real-machine deployment. For complete 8DOF mechanical, hardware, and software documentation, please refer to the 8dof branch (opens in a new tab).

16DOF Current Version

The current main branch primarily maintains documentation for the 16DOF version, including mechanical design, simulation, control, and real-world deployment.

SolidWorks 2026 mechanical source files for the 16DOF version, with the assembly entry at 02_mechanical/CAD/solidworks/WEEKDOG.SLDASM.
Complete assembly, body, joints, legs, and wheel assembly STEP files, with the entry at 02_mechanical/STEP/WEEKDOG.STEP.
Next-generation mjlab training tasks and local dependencies located at 05_software/train/rc_mjlab/.
16DOF wheel-legged MuJoCo/MJCF model, meshes, Sim2Sim validation tools, and early policy weights.
Later Sim2Sim tools, navigation map and waypoint tools, route data, and sampled PCD located at 05_software/train/rc_mjlab/.
IK trajectory and early real-world control exploration located at 05_software/real/ik_real/.
Next-generation Python Sim2Real deployment stack, including policy runtime, motor mapping, IMU integration, standing initialization, safety checks, web debugging, and calibration tools.
Python Sim2Real v2 deployment stack, adding motor feedback freshness, Odin odom diagnostics, command smoothing, web runtime diagnostics, and safety monitoring tools, located at 05_software/real/sim2real_v2/.
ROS 2/C++ Sim2Real first-generation project, migrating the policy hot path to 50 Hz C++ inference and 200 Hz CAN motor loop, with added ROS 2 messages, command arbitration, Nav2, and unified launch structure, located at 05_software/real/sim2real_ros2/. sim2real_ros2_v2 adds navigation prototypes, route configurations, waypoint tools, remote control serial ports, and web/UDP debugging links, while sim2real_ros2_v3 organizes final competition deployment, Odin driver, TensorRT/ONNX inference, and odometry debugging code, located at 05_software/real/sim2real_ros2_v2/ and 05_software/real/sim2real_ros2_v3/.
Next-generation Sim2Real real-world validation videos and preview images.

Follow-up Organization Content

16DOF RL quadruped robot-related notes and various documents.

The current repository has been fully open-sourced and organized. Please refer to the directory descriptions and the versions corresponding to Git Tags.

Communication and Support

Students who are preparing for competitions or interested in legged-wheeled robots are welcome to join our discussions:

Legged-wheeled open-source discussion group: Click to join QQ group (opens in a new tab)
Group number: 767195310

Legged-wheeled open-source discussion group QR code

Feel free to discuss mechanical design, hardware setup, microcontroller control, simulation training, reinforcement learning, and real-machine deployment in the group. Suggestions for this project are also welcome.

Team Members

Member	Role	Main Responsibilities	Contact
晨小荼 (chentu12)	Team Leader / Algorithm	Building the initial platform; early-stage responsible for 8DOF platform construction and algorithm, participating in hardware and algorithm discussions; later-stage responsible for the 16DOF wheeled-legged robot Odin1 platform construction and hardware implementation, corporate and organizer coordination, and continuously advancing project implementation.	WX: `r15288746596`QQ: `2898468947`
时维 (opens in a new tab)	Technical Lead / Project Management	Leading the software and control architecture of the 16DOF wheeled-legged robot in the mid-to-late stages, participating in mechanical and hardware solution determination; responsible for reinforcement learning training, simulation verification, real-machine deployment, navigation integration, and project progress promotion and system construction.	WX: `zeitvex`QQ: `3156045992`
Dichen (opens in a new tab)	Electrical Control / Algorithm	Responsible for electrical control and algorithm software architecture design and system integration. Early-stage completed 8DOF motor control, CAN communication, PID, leg inverse kinematics and gait debugging; mid-to-late-stage responsible for motor control, sensor access, state estimation, and navigation link development for the 16DOF platform, assisting Odin integration, and participating in simulation verification, reinforcement learning training, and real-machine deployment.	WX: `Dichenccc`QQ: `3357573813`
Sulcunfu (opens in a new tab)	Electrical Control / Mechanical	Participated in early-stage 8DOF and mid-to-late-stage mechanical upgrade design from 8DOF to 16DOF; responsible for model assembly, URDF export, simulation and algorithm verification, participated in reinforcement learning model training optimization, configured Linux basic deployment environment, connected Odin1 visual perception, odometry and IMU data, and responsible for environmental mapping and export.	WX: `LcfNotFound`QQ: `3299459360`
王书琪 (opens in a new tab)	Mechanical / Hardware	Responsible for overall mechanical structure design and electrical hardware system construction; responsible for hardware adaptation of the 8DOF open-source mechanical structure, as well as modeling, optimization, and CNC supply chain coordination for the 16DOF overall mechanical structure; responsible for power management PCB design and CAN bus communication topology, providing high-rigidity and high-stability hardware foundation for algorithm and simulation deployment.	WX: `DouSfJade`QQ: `3070940094`

Real-world demonstration

The current repository contains Sim2Real real-machine test records and initial competition results, with videos and preview images located in 06_assets/:

Initial Wheel-Legged Real Machine Test Records

First-generation Sim2Real real machine test records, approximately 41 seconds long. Click the preview image to play or download the original video.

Final Competition Test Record

This video demonstrates the final competition results. The resolution is 1280×720 in landscape mode, with a total duration of approximately 1 minute and 20 seconds, closely matching the actual competition duration. Click the preview image to play or download the original video.

Directory Structure

RC_WheelLeg/
├─ 01_doc/          # Project technical documentation and usage instructions
├─ 02_mechanical/   # SolidWorks CAD, STEP, and mechanical source materials
├─ 03_hardware/     # Self-developed circuits, wiring, BOM, sensor, and computing platform documentation
├─ 04_firmware/     # MCU or other embedded controller firmware
├─ 05_software/     # Reinforcement learning training, MuJoCo, Sim2Sim, Sim2Real, and tools
├─ 06_assets/       # Images, videos, and test results
├─ .gitignore
└─ README.md

Project Entry Points

Mechanical documentation:

Software documentation:

Media assets:

Software

The current 05_software/ directory contains complete software-related materials:

MJCF + mjlab task
        |
        v
   PPO training strategy
        |
        +----> MuJoCo standalone model debugging
        |
        +----> Sim2Sim strategy validation
        |
        +----> Python Sim2Real / v2 ----> Motor / IMU
        |
        +----> ROS 2/C++ Sim2Real / v2 / v3 -> CAN / IMU / Navigation & competition deployment

IK real --------------------------------> Motor

Here, rc_mjlab is a self-contained training and simulation project, including src/robot training tasks, mjcf models, mujoco_sim standalone debugging tools, sim2sim validation tools, local mjlab dependencies, and early policy weights.

The sim2real deployment currently uses a 53D -> 16D policy model, with a control frequency of 50Hz. It includes standing initialization, standing balance, runtime safety checks, zero-command hold, and first-command smooth release logic. Before deployment, please read 05_software/real/sim2real/README.md, DEPLOYMENT.md, and ORIN_NANO_DEPLOYMENT.md.

sim2real_v2 corresponds to Python Sim2Real v2, retaining the 53D -> 16D policy interface while adding motor feedback freshness, Odin odom diagnostics, command smoothing, web runtime diagnostics, and safety monitoring tools. Before deployment, please read 05_software/real/sim2real_v2/README.md and DEPLOYMENT.md.

sim2real_ros2 is the first version of ROS 2/C++ Sim2Real; sim2real_ros2_v2 adds navigation prototypes, route configuration, waypoint tools, remote control serial ports, and web/UDP debugging links; sim2real_ros2_v3 organizes the final competition deployment, Odin driver, TensorRT/ONNX inference, and odometer debugging. For complete entry points, see 05_software/real/sim2real_ros2/README.md, 05_software/real/sim2real_ros2_v2/README.md, and 05_software/real/sim2real_ros2_v3/README.md.

Safety Instructions

The robot must be kept off the ground and the motor IDs, directions, zero positions, limits, and joint mappings must be confirmed before the first run.
During debugging, ensure that the emergency stop link in the remote controller, upper computer, or deployment script is available.
The robot should not be powered on and run directly without verifying the mechanical limits, simulation model, strategy output scale, and motor mapping.
Reinforcement learning strategies and Sim2Real deployment require offline checks, simulation verification, low-speed no-load testing, and small-motion testing to be completed first.
This project may generate large joint torques, and users must bear the hardware debugging risks themselves.
The zero position must be set while the robot is in a stable supported state, and the actual posture of the body, foot ends, knee joints, and wheel assemblies must be recorded to avoid inconsistencies between the subsequent model and the real machine coordinates.

License

This project is licensed under the MIT License.

Files by @zeitvex, at commit 1bc9aa5, read from GitHub in your browser. Hardware license: MIT; software license: MIT. Robotheca doesn't review code: read it before running it.