UMI-3D Hardware Building Guide

🌐 UMI-3D Project Homepage

🔧 UMI-3D Hardware	🛰️ UMI-3D SLAM Pipeline	🤖 UMI-3D Policy
Hardware design, BOM, CAD, 3D-print parts	SLAM, synchronization, calibration, and data processing	Policy training, deployment, inference 📦 Dataset & Models

---

1. Bill of Materials (BOM)

Target total cost: ≈ ¥5,000 (≈ $700, subject to actual purchase)

UMI-3D Handheld Gripper

_{Left: real UMI-3D handheld gripper. Right: CAD overview (view in Onshape).}

Mechanical Components

Component	Description	STL Files	Price	Preview	Assembled View
Gripper Mount	3D Print, PLA Infill: 20% Gyroid	📁 Link	—
Soft Finger	3D Print, TPU 95A Infill: 100% Lines	📁 Link	—
150mm MG9NC Linear Module	Aluminum, CNC Bracket	—	—

Sensors

Component	Model / Name	Price	Image
LiDAR	Livox MID-360/MID-360S	¥3999/¥3599 ($560/$510)
Camera	Hikrobot MV-CB013-A0UC-S	¥700 (≈ $100)
Lens	ZLKC MTV185IR12MP	¥139 (≈ $20)

Electrical & Wiring

Component	Model / Name	Price	Image
Synchronizer	LiDAR–Camera Hardware Synchronizer	¥288 (≈ $40)
Battery	12V DC Battery	¥79 (≈ $10)

---

UMI-3D Ego (Under Development):

Mechanical Components

Component	Description	STL Files	Price	Preview	Assembled View
Sensor Mount	3D Print, PLA Infill: 20% Gyroid	📁 Link	—
Headband	TELESIN GoPro Headband	—	¥35.9 (≈ $5)

Sensors (Same as Above)

Electrical & Wiring (Same as Above)

---

UMI-3D Robot Arms (Manipulator-Side Mounts)

ARX L5/R5/X5

Component	Description	STL Files	Price	Preview	Assembled View
Finger Mount	3D Print, PLA Infill: 40% Gyroid	📁 Link	—		ARX-mounted UMI-3D setup
Camera Mount	3D Print, PLA Infill: 40% Gyroid	📁 Link	—

Agilex Piper (Under Development)

Component	Description	STL Files	Price	Preview	Assembled View
Finger Mount	3D Print, PLA Infill: 40% Gyroid	📁 Link	—
Camera Mount	3D Print, PLA Infill: 40% Gyroid	📁 Link	—

---

2. Electrical & Wiring

_{UMI-3D wiring overview. The 12V battery powers the LiDAR and the hardware synchronizer. The synchronizer provides trigger/sync signals between the LiDAR and camera. LiDAR data and camera images are transmitted to the host computer, recorded in rosbag format, and subsequently used for SLAM, sensor calibration, data processing, and policy training.}

Connection Overview

Module	Connects To	Purpose
12V Battery	LiDAR	Main power supply for Livox MID-360
12V Battery	Hardware Synchronizer	Power supply for synchronization board
Hardware Synchronizer	Camera	Trigger / sync signal for image capture
Hardware Synchronizer	LiDAR	Shared timing reference
LiDAR	Host Computer	Point cloud data transmission
Camera	Host Computer	Power supply and Image data transmission

Note

• Please make sure the battery polarity is correct before powering on.
• Power on the synchronizer and LiDAR first, then check whether the camera trigger is working properly.
• Secure all cables to avoid motion-induced looseness during handheld or robot-arm operation.

---

3. UMI-3D ROS Driver

This repository provides the ROS drivers and helper scripts for the UMI-3D sensing setup:

• Livox MID-360 / MID-360S LiDAR driver
• Hikrobot industrial camera ROS driver
• MVS launcher (for camera configuration)
• Rosbag recording tools

---

Repository Structure

umi_3d_ros_driver/
├── camera.sh
├── lidar.sh
├── mvs.sh
├── record.sh
└── src/
    ├── livox_ros_driver2/
    └── mvs_ros_driver/

---

Environment

• Ubuntu 20.04
• ROS Noetic
• System Python (required)

Note

• Always run conda deactivate before building or launching ROS drivers. Using Conda may cause ROS dependency conflicts.

---

Dependencies

• MVS (camera SDK)
  https://github.com/bitcat-tech/MVS_V3.0.1

• Livox-SDK2
  https://github.com/Livox-SDK/Livox-SDK2

• ROS Noetic

---

Quick Start

cd ~/umi_3d_ros_driver
conda deactivate

# 1. Build Livox driver
cd src/livox_ros_driver2
./build.sh ROS1
cd ../..

# 2. Build full workspace
catkin_make
source devel/setup.bash

# 3. Configure camera if needed
bash mvs.sh
# → set trigger mode, then CLOSE MVS

# 4. Launch drivers
bash lidar.sh
bash camera.sh

# 5. Record data
bash record.sh

Note

• livox_ros_driver2 must be built with: ./build.sh ROS1

---

Manual Launch (Optional)

LiDAR

conda deactivate
source devel/setup.bash
roslaunch livox_ros_driver2 msg_MID360.launch

Camera

conda deactivate
export LD_LIBRARY_PATH=/opt/MVS/lib/64:/opt/MVS/lib/32:/opt/ros/noetic/lib:$LD_LIBRARY_PATH
source devel/setup.bash
roslaunch mvs_ros_driver mvs_camera_trigger.launch

Record

rosbag record /livox/lidar /livox/imu /left_camera/image

Note

• Before launching the camera ROS driver, make sure:
  • MVS has been closed
  • the camera has already been set to hardware trigger mode
  • LD_LIBRARY_PATH includes the MVS runtime libraries

---

Verification

rostopic hz /livox/lidar
rostopic hz /left_camera/image

Expected:

• LiDAR publishes normally
• Camera ≈ 20 Hz

Note

• Do not run MVS and mvs_ros_driver at the same time. Use MVS only for camera configuration, then close it before launching ROS.
• If the image publishing rate is lower than 20 Hz, check the USB 3.0 port and USB 3.0 cable quality.
• If build issues occur, re-run:

  cd src/livox_ros_driver2
  ./build.sh ROS1
  cd ../..
  catkin_make

• Data is recorded in rosbag format and can be used for:
  • SLAM
  • Sensor calibration
  • Data processing for Policy learning

---

Citation

If you find this work useful for your research, please consider citing:

@misc{wang2026umi3dextendinguniversalmanipulation,
  title={UMI-3D: Extending Universal Manipulation Interface from Vision-Limited to 3D Spatial Perception},
  author={Ziming Wang},
  year={2026},
  eprint={2604.14089},
  archivePrefix={arXiv},
  primaryClass={cs.RO},
  url={https://arxiv.org/abs/2604.14089}
}

Acknowledgements

This work builds upon several outstanding open-source projects, including UMI (Universal Manipulation Interface), UMI-on-Legs, and LIV-Eye. We gratefully acknowledge the authors for their pioneering contributions to embodied intelligence, perception, and robot learning.