Synchronized Multimodal Capture
RGB, ToF, IMU, and other sensor streams are aligned with hardware-level timestamps to deliver highly consistent training data.
Lumos FastUMI Pro
Multimodal Embodiment-Free Data-Collection Hardware-Software System for General-Purpose Robot Manipulation
Purchase & PartnershipsFASTUMI PRO
Robot-independent multimodal data collection, purpose-built for the embodied AI training pipeline.
Lumos FastUMI Pro is a multimodal, embodiment-free data-collection hardware-software system from Lumos that quickly adapts to dozens of mainstream robotic arms and grippers, enabling handheld operation capture, multimodal recording, and high-quality embodied dataset creation without relying on a physical robot body.
Robot-Independent Data Collection200 Hz OutputROS / ROS2
Key Capabilities
Dual-Precision Positioning
Pure-vision SLAM achieves 0.1% average relative displacement error, while vision-plus-laser with VIVE trackers reaches 1 mm static positioning accuracy.
Lightweight Embodiment-Free Design
Starting at just 635 g with up to 2 kg gripping force, it captures embodied manipulation data without depending on a physical robot body.
Open and Compatible Ecosystem
Natively supports Ubuntu 20.04 and ROS / ROS2, with standardized data formats and interfaces for mainstream training workflows.
Industrial-Grade Scenario Adaptability
Handles challenging environments such as strong lighting and occlusion, with optional tactile sensors to cover full-scenario needs from lab research to production lines.
Versions
Version Options
Full-Featured
High-Precision External Tracking
Integrates VIVE Tracker 3.0 and dedicated base stations for workflows that require stable, high-precision external positioning.
Rotational
735±5 gIntegrated VIVE Tracker 3.0 + dedicated base stations
Fingertips rotate in an arc around a fixed axis
Translational
780±5 gIntegrated VIVE Tracker 3.0 + dedicated base stations
Two fingers stay parallel and move linearly
Key Metrics
Key Specifications
635 g
Lightest Weight
2 kg
Max Gripping Force
88±2 mm
Jaw Travel
200 Hz
Output Frequency
0.1%
Vision Error
1 mm
Static Accuracy
Output
Data & Ecosystem
RGB video (.mp4)ToF point cloud (.pcd)Gripper width sequenceFused pose trajectoryFastUMI-10K dataset
Data Formats
Supports RGB video, ToF point clouds, gripper-width sequences, and fused pose trajectories.
Training Ecosystem
Compatible with mainstream embodied AI training frameworks, with full FastUMI-10K dataset access available through official channels.
Expansion Capability
Supports vision-tactile sensors and onboard camera modules, and can also be deployed on platforms such as Lumos Touch R1.
Tracker
VIVE Tracker & Base Station Specifications
Tracker Weight
75 g
Scan Rate
100 scans/s
Tracking Resolution
Sub-millimeter
Coverage
7 m
Field of View
240°
Viewing Angle
160° × 115°
Product Specifications
Key metrics, version architecture, and data ecosystem at a glance.
Hardware Configuration & Versions
- Full-Featured
- Integrated VIVE Tracker 3.0 + dedicated base station
- Portable
- Onboard visual SLAM only; no external base station required
- Rotational Full-Featured
- Weight 735±5 g
- Rotational Portable
- Weight 635±5 g
- Translational Full-Featured
- Weight 780±5 g
- Translational Portable
- Weight 700±5 g
- Rotational Version — Motion Type
- Fingertips rotate along an arc around a fixed axis.
- Translational Version — Motion
- the two fingers stay parallel, moving in a straight translation
OPTIONAL MODULES
- Vision-Tactile Sensor
- Tactile sensor arrays mounted on the gripper fingertips capture fine-grained tactile feedback during object interaction.
- Onboard camera module
- mounted on a robotic arm (e.g., Lumos Touch R1)
KEY DATA
- Mini weight
- 635 g
- Max Gripping Force
- 2 kg
- Gripper stroke
- 88±2 mm
- Gripper data output rate
- 200 Hz
- Avg. relative displacement error (vision-only)
- 0.1%
- Static localization accuracy (vision + laser)
- 1 mm (dual-path localization)
HARDWARE
- Fisheye camera
- FOV: DFOV=200°, HFOV=172.52°, VFOV=172.52°, resolution 1280×1280, 60±1fps
- SLAM cameras
- 4 units, 640×480, HFOV 130°, 30 fps
- ToF camera
- 320×240, 30±1 fps, FOV D78° H64° V50°, depth error <1.5% within 2m
- IMU
- 6-axis, ICM42688, 500 Hz
- Voltage & power
- 5 V, max power <4 W
- Interface
- USB 3.2 Type-C 10Gbps
- Multi-camera time sync
- Timestamp alignment across all camera streams
- Software
- Ubuntu 20.04, ROS/ROS2
Application Scenarios
Academic research, industrial deployment, and data production scenarios demonstrate FastUMI Pro’s precision, openness, and cross-platform data capability.
Academic Research
Built for robot imitation learning, algorithm verification, policy training, and skill demonstration replay with highly consistent multimodal training data.
Bringing Embodied AI into the Real World
A full-stack industrial embodied AI capability system built on three integrated pillars: scenario resources, data and model capabilities, and hardware infrastructure.