Connecting to sources…

How to build MiniLoong (Qinglong Mini, RL-Boy) small humanoid for reinforcement learning

28 steps to build MiniLoong (Qinglong Mini, RL-Boy) small humanoid for reinforcement learning, with the source documents.

Not sold and no kit. The README BOM's unit-price column, in CNY: 6006 motor CN¥799 (attribute column 4), 8006 motor CN¥1009 (attribute column 6), 3507 motor CN¥300 (attribute column 8, quote from Damiao), main board CN¥899, optional Jetson Nano CN¥1119, high-performance IMU CN¥900, ESP32 OmniTele remote CN¥600; carbon-fibre cutting, aluminium machining and 3D printing say “inquiry”, and there is no total. The links go to Taobao and Tmall (not read, 2026‑11‑01)

Buy parts

  • Prepare printing and carbon-fiber materials

    Prepare the 3D-printing materials and carbon fiber sheets listed in the BOM for DIY fabrication and processing.

  • Purchase the BOM components

    Purchase all components using the RMB prices in the Components and Prices table.

  • Purchase screws and nuts

    Purchase the required screws and nuts according to the Screws and Nuts table.

  • Purchase the required cables

    Purchase the robot cables listed in the Cables table.

  • Source BOM fabrication materials

    Prepare 3D-printing materials and carbon-fiber plates according to the BOM. The document says these materials support DIY and fabrication.

Assemble

  • Follow the hardware wiring diagram

    Connect the robot hardware according to the hardware wiring diagram in the document.

  • Wire the hardware by diagram

    Connect the robot's circuit boards, motors, and other hardware according to the hardware wiring diagram.

Software

  • Set the Tinker router credentials

    Use the router named Tinker-2.4G-ID with the password 11111111.

  • Use the matching training URDF

    Use TinkerV2_URDF.zip, which corresponds to the training model and is 2.32MB.

  • Flash the STM32 firmware files

    Update the STM32-Core firmware with core.hex, which is 230.33KB. The extcan.hex firmware file is 159.33KB.

  • Update the ODroid Tinker package

    Update the ODroid control software package firmware and parameters with Tinker.zip, which is 21.34MB.

  • Launch the hardware startup task

    Modify rc.local so the startup scripts run correctly. Wait 5 seconds, then launch hardware_task_tinker.

  • Launch the control startup task

    Wait 1 second in the startup script, then launch control_task_tinker.

  • Launch the navigation startup task

    Wait another 1 second in the startup script, then launch navigation_task_tinker.

  • Update the STM32 firmware

    Update the STM32 firmware using the provided extcan.hex dated 2025/7/14.

  • Update the ODroid control software

    Use WinSCP to update the control software package on the ODroid internal system, including its firmware and parameters.

  • Configure hardware-task autostart

    Modify rc.local to wait 5 seconds, start hardware_task_tinker2, and then wait 1 second.

  • Configure mission and navigation autostart

    Continue modifying rc.local to start mission_task, wait 1 second, and then start navigation_task_tinker2.

Calibrate and test

  • Configure the router network

    Set the router name to Tinker-2.4G-ID Number and the password to 11111111.

  • Connect and configure HMI boards

    Connect the Core and Extcan boards by USB, complete the motor reverse checkboxes and motor type selections from the configuration diagram, then use each motor's Dami host computer to configure its CAN_ID.

  • Configure left-leg CAN1 motors

    Use the Dami host computer to assign each motor the CAN_ID shown in the Left Leg STM32-Core CAN1 table.

  • Configure right-leg CAN2 motors

    Use the Dami host computer to assign each motor the CAN_ID shown in the Right Leg STM32-Core CAN2 table.

  • Configure left-arm CAN1 motors

    Use the Dami host computer to assign each motor the CAN_ID shown in the Left Arm STM32-Core CAN1 table.

  • Configure right-arm CAN2 motors

    Use the Dami host computer to assign each motor the CAN_ID shown in the Right Arm STM32-Core CAN2 table.

  • Configure the HMI core STM32

    Configure the HMI interface's core STM32 using the displayed configuration screen.

  • Configure the HMI ExtCan

    Configure the HMI interface's core ExtCan using the displayed configuration screen.

  • Set motor CAN IDs
  • Configure the RobHub-Core HMI

    Configure the RobHub-Core HMI according to the interface shown in the document. The image also labels extcan.

Parts

  • 6006, 8006 and 3507 brushless motors (the README sets CAN IDs with Damiao's tool): 5 per leg for yaw, roll, thigh, shank and foot, 1 each for waist and head yaw, and 4 per arm
  • Two STM32 boards, Core and ExtCan, each with 2 CAN buses for the legs and the arms; a power distribution board; a Linux main board (the README lists Raspberry Pi, Odroid, Coolpi and D-Robotics RDK; the autostart paths are for Odroid)
  • Optional Jetson Nano with up and down USB cameras; IMU and serial cable, 64 GB TF card
  • OLED face and ESP32 in the head; 2 brushless servos for the grippers (SCS40 optional), with an optional SCS125 head servo and servo driver board
  • 6S 24V 7Ah battery, 12V and 5V DC-DC converters, a buck module, a 32A breaker, a 2.4 GHz router, XT30 and XT-CAN cables
  • Structure: cut carbon-fibre plates, machined aluminium parts and 3D prints, M2.5–M5 screws and lock nuts, foot pads
  • Remote: a vibrating gamepad or a self-built ESP32 OmniTele remote

Tools

  • Carbon-fibre cutting, aluminium machining and 3D printing (quoted on request in the BOM, so outsourced or done on your own machines)
  • A Windows PC for the OmniBotHmi host program and Damiao's motor tool; WinSCP for updating the main board
  • An x64 cross-compiling setup (binaries run on the arm64 main board)

Skills

  • Wiring brushless motors on CAN, setting IDs and motor directions
  • Linux cross-compiling and the C++ control framework (the loong_ctrl state machine schedules each plan)
  • Reinforcement learning training in Isaac Gym, sim2sim checks and ONNX/TVM deployment
  • Imitation learning such as ACT and DP and VLA manipulation packages; collecting data with the dual-arm teleoperation rig

Software

  • Lower-level control_task2: standing, self-righting and several RL gait sources (rl_boy, rl_tinker, rl_lite), with Tinker14, Lite14 and Taitan14 parameter sets
  • loong_ctrl: a state-machine framework with ONNX and TVM inference, running in MuJoCo simulation (x64) and on the robot (arm64)
  • OmniBotHmi Windows host program; OmniRobLab V1.0 embedded software, with the STM32 firmware and URDF in the Feishu document

Documents