High-Precision Hand Tracking
Tracks more than 42 hand-joint keypoints across both hands, with per-frame confidence scores for stable fine-manipulation labels.
Lumos Ego STD & Lite
Head-Mounted Egocentric Multimodal Data-Collection Device
Purchase & PartnershipsLUMOS EGO
Turn Natural Human View, Hand Motion, and Spatial Context into Synchronized Training Data
A head-mounted egocentric capture device integrating a wide-angle fisheye camera, high-precision SLAM, hand-joint tracking, and multimodal synchronization. It is available in both professional and lightweight versions for different deployment scales.
First-Person ViewMultimodal SynchronizationSTD / Lite
Core Capabilities for Manipulation Learning
Multimodal Synchronization
Microsecond-level hardware timestamp alignment across RGB, ToF*, SLAM pose, and IMU streams keeps multi-source training data consistent.
*ToF is included only in Lumos Ego STD.Lightweight Wearable Design
At only 235 g, the head-worn structure supports a wide 20°~180° viewing adjustment range for comfortable long-duration real-world capture sessions.
Ready Out of the Box
Natively supports Ubuntu 20.04 and ROS / ROS2, with automation scripts that lower the barrier for egocentric capture deployment.
Versions
STD / Lite Versions Comparison
Professional Version
Lumos Ego STD
Integrates a wide-angle fisheye RGB camera, high-precision SLAM, ToF depth sensing, and other sensors to synchronously capture visual, depth, pose, and hand-joint data.
Includes ToF depth sensing
Multimodal synchronized capture
Built for high-quality training data
Sensors
Key Perception Modules
From fisheye vision and SLAM to depth and IMU, the system covers the full sensing chain needed for egocentric capture.
Fisheye Camera (RGB)
FOV: DFOV=200°, HFOV=172.52°, VFOV=172.52°, resolution 1280 × 1280, 60±1 fps
SLAM Cameras
4 units, 640 × 480, HFOV 130°, 30 fps
ToF Camera
320 × 240, 30±1 fps, depth error < 1.5% within 2 m, not included in the Lite version
IMU & System Interfaces
6-axis ICM42688 at 500 Hz; USB 3.2 Type-C 10 Gbps; aligned timestamps across all video streams
Combinations
Typical System Configurations
Configuration
Ego STD / Lite + PC Host
Focused on egocentric visual demonstration, using an ultra-wide field of view to record visual input, object features, and scene structure.
Use Case
Bench-top tasks such as object recognition, grasp demonstrations, and desktop assembly learning
Product Specifications
Key metrics, sensing modules, and version differences at a glance.
SENSOR MODULE
- Fisheye camera (RGB)
- FOV: DHOV=200°, HOV=VFOV=172.52°, resolution 1280×1280, 60±1fps
- SLAM cameras
- 4 cameras, 640×480 resolution, HFOV 130°, 30 fps
- ToF camera
- 320×240, 30±1fps, FOV: D78° H64° V50°, depth error <1.5% within 2m (Only STD version)
- IMU
- 6-axis, ICM42688, 500Hz
POWER & INTERFACE
- Voltage & power
- 5 V, max power <4 W
- Data interface
- USB 3.2 Type-C 10Gbps
- Time sync
- Timestamp alignment across all camera streams
- SDK
- Ubuntu 20.04
ERGONOMICS
- Total weight
- 235 g
- Headband
- High-elastic knitted headband; housing temperature <40°C
- Viewing-angle adjustment
- 20°–180°
- Hand-joint tracking
- 42 keypoints (both hands)
Typical Applications
Built around real egocentric workflows, capturing operator view, hand motion, and task context for more human-like training data.
Dexterous Manipulation
Captures grasping, assembly, and tool use to restore fine interactions between the hand and objects.
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.