Connecting to sources…

How to build MEVIUS sheet-metal welded quadruped: the whole process

27 steps to build MEVIUS sheet-metal welded quadruped, from buying parts, printing and assembly to software, calibration and training, with the bill of materials, shop l…

The creator's parts list (a Google Sheet, yen with US-dollar conversions): the standard build totals $12,390.09, of which the 12 AK70-10s are JP¥69,846 ($498.90) each, $5,986.80 in all, and the meviy sheet-metal and machined parts $3,179.81 (optional limit parts another $320.25); the low-price build totals $5,304.11 (cheaper motors, computer, CAN-USB and IMU, and no wireless E-stop). Neither includes screws, printing filament or shipping. RT's made-to-order version is priced on request (2026‑11‑04)

Buy parts

  • Order MEVIUS Materials

    Obtain machined-metal parts, sheet-metal welding parts, and off-the-shelf components through e-commerce; all hardware and software components are available from GitHub.

  • Obtain all hardware components

    Obtain all MEVIUS hardware components from the Goodle Drive link in the document.

  • Order the Custom MEVIUS Robot

    MEVIUS is a custom-order research quadruped robot assembled with the required parts. Its exterior can be customized, and specifications and appearance may change while development continues.

Assemble

  • Prepare Ten Basic Structural Components

    Prepare the 10 component types that form MEVIUS's basic structure as shown in the document. Mirror parts and feet are excluded from these 10 component types.

  • Machine and Weld Base and Hip Links

    Machine the metal components and fabricate the Base-Link and Hip-Link through sheet metal welding.

  • Build the MEVIUS Circuit

    Connect the electrical components according to the shown MEVIUS circuit configuration.

  • Arrange Circuits Inside the Base-Link

    Arrange the MEVIUS circuits inside the Base-Link at the positions shown in the image.

  • Install MEVIUS Circuit Components

    Prepare and install the MEVIUS circuit-related components shown in the document.

  • Weld Base-Link and Hip-Link

    The basic structure consists of 10 types of components excluding mirror parts and feet. Construct the Base-Link and Hip-Link through sheet metal welding.

  • Arrange Base-Link Circuits

    Arrange the circuits inside the Base-Link according to MEVIUS's circuit configuration, and complete the simple circuit with circuit-related components.

Software

  • Download the upper computer installation program

    Download the upper computer installation program from the RUBIK LINK V2.0 link.

  • Install the software dependencies

    Install ROS, realsense-ros, urdfdom-py, and legged_gym; spacenav_node and virtual_joystick are optional dependencies.

  • Install the Python packages

    Install bitstring, pytorch, scipy, mujoco, and mujoco-python-viewer.

  • Simulate the robot in MuJoCo

    Run mevius_main.py with the --sim option, then press D in the MuJoCo viewer to turn off every-step rendering.

  • Configure the MEVIUS Software

    Configure MEVIUS using the software components supplied through GitHub, following the simple software configuration described.

  • Prepare a Development Computer

    Development computers are not included and must be prepared separately.

Calibrate and test

  • Set all motors to MIT mode

    Use the upper computer installation program to set all motors to MIT mode.

  • Configure motors in the specified order

    Configure the motors in the order FR(1-3), FL(4-6), BR(7-9), and BL(10-12), arranged from proximal to distal.

  • Set the link names

    Name the links FR_scapula_link, FR_thigh_link, and FR_calf_link.

  • Set the joint names

    Name the joints FR_collar, FR_hip, and FR_knee.

  • Configure Circuit and Software

    Complete the simple circuit and software configuration for MEVIUS.

Data and training

  • Train Walking Policy in IsaacGym

    Use the IsaacGym simulation environment to perform reinforcement learning for MEVIUS walking.

  • Train Walking in IsaacGym

    Use the IsaacGym simulation environment for reinforcement learning of MEVIUS walking.

Use it

  • Conduct Indoor Terrain Traversal

    Run traversal experiments with MEVIUS over changing indoor terrain.

  • Conduct the Outdoor Robot Experiment

    Test MEVIUS outdoors using reinforcement learning and Sim2Real.

  • Test Changing Indoor Terrain

    Conduct traversing experiments with MEVIUS on changing indoor terrain.

  • Conduct Outdoor Experiment

    Conduct MEVIUS experiments in an outdoor environment.

Parts

  • 12 T-Motor (CubeMars) AK70-10 joint motors, set to MIT mode with RUBIK LINK V2.0
  • meviy parts: a welded A5052 sheet-metal Base-Link and 4 Hip-Links, a sheet-metal Base-Cover, and 4 each of machined A7075 Thigh-Base, Thigh-Link, Thigh-Cover, Calf-Base, Calf-Base-Cover and Calf-Rod; optional 8 Hip-Limits and 4 Thigh-Limits
  • Printed parts: 4 POTICON NTL34M calf links, 4 TPU 95A feet and a PLA T265 mount
  • 8 RNAF6138N needle bearings, 8 C-B628ZZ ball bearings, hinge pins, dowel pins and washers; M3 (140 of M3×8) and M5 screws
  • Electronics: an Intel NUC12WSKi7, an ESD CAN-USB/2, 2 KOOKYE boost modules, a Harmony ZBRRA wireless E-stop, an OMRON G9EN power relay, a GeneSiC MBRT40040 diode (for motor regeneration) and an Intel RealSense T265
  • Batteries: one HRB 3S 6000 mAh 11.1V logic pack and two HRB 6S 3300 mAh 22.2V power packs (in series)

Tools

  • A MISUMI meviy account (upload the CAD for automatic quotes and order the welded and machined parts)
  • A 3D printer that can handle POTICON, a high-strength nylon-based compound, and TPU
  • CubeMars' RUBIK LINK V2.0 and its PC software; a computer with Isaac Gym and legged_gym for training the policy

Skills

  • Ordering the welded and machined parts from the STEP files and assembling them; setting up CAN (bin/can_setup.sh) and testing each motor
  • ROS 1 and catkin, realsense-ros (ros1-legacy) and spacenav_node; simulation in MuJoCo (--sim)
  • Training the walking policy in two phases with the creator's legged_gym fork and deploying the exported policy.pt on the robot

Software

  • The mevius ROS package (MIT, Python): mevius_main.py for the robot and MuJoCo simulation, tmotor_test.py for motor tests, URDF and MuJoCo models and a trained policy.pt
  • The mevius branch of haraduka/legged_gym: the Isaac Gym reinforcement-learning environment

Documents