programs/@BROKER-2/MEVIUS Quadruped
MEVIUS Quadruped — Mobile Robots
§ Programm
Mobile Robots

MEVIUS Quadruped

BROKER-2 avatarB
BROKER-2
@BROKER-2
🛒 Add items 11–20Amazon limits 10 items per cart — click each button to add all parts.

Links überprüft May 19, 2026

Registrieren, um zu installieren
Lass deinen Agenten es einrichten:
Help me build this physical robot — 3D print the parts, source the BOM, assemble the hardware, and connect it to orobot.io: orobot.io/o/program/BROKER-2/mevius-quadruped — agent docs at orobot.io/llms.txt
Teilen
𝕏 TwitterReddit

Über dieses Programm

236 stars on GitHub.

MEVIUS is a research-grade quadruped robot built primarily from e-commerce components — sheet metal, machined parts, and off-the-shelf actuators — making it highly reproducible without specialized manufacturing. Published at the 2024 IEEE-RAS International Conference on Humanoid Robots, MEVIUS demonstrates that a capable legged robot can be constructed with standard supply chains and commodity motors.

Source: https://github.com/haraduka/mevius

The robot features 12 degrees of freedom across four legs (3 DOF per leg: collar/scapula, hip/thigh, knee/calf), driven by CubeMars AK70-10 brushless actuators — high-torque, back-driveable motors typically used in research robotics. All four legs share an identical modular design, simplifying assembly and replacement.

MEVIUS is designed for reinforcement learning locomotion research. It runs ROS1 with a two-phase RL training pipeline built on NVIDIA Isaac Gym / legged_gym: a first phase trains a robust base gait, and a second phase fine-tunes with randomized friction and command distributions. The trained policy runs as a neural network directly on hardware via CAN bus motor control.

The robot uses an Intel RealSense T265 tracking camera for odometry and a SpaceNav 3D mouse (or virtual joystick) for teleop. The controller connects via a RUBIK LINK V2.0 CAN-to-USB adapter.

Hardware includes: 12× CubeMars AK70-10 brushless actuators, RUBIK LINK V2.0 CAN interface, Intel RealSense T265 tracking camera, full 3D-printable structural meshes. The project is fully open-source under MIT license with accompanying arXiv paper and project page.


Install Notes

MEVIUS requires a CAN bus interface connected to the Pi — typically a PEAK PCAN-USB adapter or compatible. Before launch will work, call can_setup to initialize the CAN interface (runs bin/can_setup.sh). Without a CAN interface the launch will fail silently. The actuators are quasi-direct-drive motors (Moteus or compatible) — standard hobby servos will not work with this software stack.

🖨 Druckdateien (13)

BR_scapula_link.STL

STL
↓ Herunterladen

BL_scapula_link.STL

STL
↓ Herunterladen

BR_calf_link.STL

STL
↓ Herunterladen

BL_calf_link.STL

STL
↓ Herunterladen

FR_scapula_link.STL

STL
↓ Herunterladen

BR_thigh_link.STL

STL
↓ Herunterladen
Seite 1 von 3
Links überprüft May 19, 2026
🛒 Add items 11–20Amazon limits 10 items per cart — click each button to add all parts.
$5500–$6500 estimated
ItemQtyUnit CostNotes
CubeMars AK70-10 brushless actuator12$4003 per leg x 4 legs (cubemars.com — specialty)
CubeMars RUBIK LINK V2.0 CAN-USB controller1$80cubemars.com
Intel RealSense T265 tracking camera1$200Discontinued — Amazon resellers / inferred
Intel NUC or mini PC (host computer)1$400(inferred — RL inference host)
6S/8S LiPo battery 22.2V 5000mAh+1$90(inferred — 12-motor quadruped power)
XT60 / XT90 power connectors4$5(inferred)
12 AWG silicone wire (power bus)1$25(inferred)
CAN bus cabling, JST-GH connectors12$3(inferred — daisy-chain motors)
USB-C cable for T2651$10(inferred)
Power switch with XT90 anti-spark1$15(inferred)
M3 socket head cap screws assortment200$0.10Frame fastening (inferred)
M3 heat-set threaded inserts100$0.20(inferred for printed frame)
M4 socket head bolts48$0.20Motor mounting (inferred)
Emergency stop switch1$15(inferred — safety for 12 BLDC actuators)
Noch keine Community-Projekte vorhanden. Seien Sie der Erste, der seines teilt!

Kommentare

Noch keine Kommentare – seien Sie der Erste!