Vision
Primary View + Dual Wrist Views: 1080p @ 60 FPS
Hand Kinematics
Hand Joint Angles: 25 DoF per hand Fingertip Pose: 6 DoF per finger
Tactile
Touch points (per hand): 400+, minimum actuation force: 30 g
Spatial Pose
Wrist + Head Relative Pose
Real Human Operation Multimodal Data Acquisition Platform
X-Ego-Pro
Our platform captures high-quality multimodal human operation data by synchronizing first-person visuals, dual wrist-mounted camera feeds, hand kinematics, and tactile signals with precise spatiotemporal alignment.
Multimodal Data Streams: Visual, Tactile, Kinematic, and Pose
Millimeter-Level Precision in Hand Joints and Spatial Localization
High-precision hand skeletal tracking and head/wrist pose estimation enable accurate reconstruction of fine-grained hand manipulation.
Loading hand demo...
Precise 5‑Stream Sync
Real Human Operation Data Collection Platform
X-Ego
Our platform captures high-fidelity first-person vision to generate realistic, continuous, and reproducible human operation data, fully equipped with automated semantic labeling.
Annual Capacity
500+
Scenarios
5,000+
Weekly Capacity
30,000+
Accumulated Hours
4K High-Quality Scene Data
High-Spec Data Capture
3840×2880 @ 30 FPS monocular vision, with full camera intrinsics (enabling distortion correction)
Multi-Dimensional Perception
frame-synchronized IMU motion pose data and natural audio
Data Ready for Direct Use in Training and Iteration
Annotation Pipeline
VLM-powered automated data preprocessing pipeline — semantic annotation + hand pose (camera coordinate frame)
Long-Horizon Logic
complete causal relationships preserved in temporal sequences
Diverse Scenarios
Task List
52 task categories
Multimodal Synthetic Data Platform
X-Sim
Leveraging generative physical simulation technology, our platform automatically generates high-diversity synthetic data for arbitrary dual-arm manipulation tasks — including grasping, assembly, folding, and articulated object manipulation.
Large-Scale Synthetic Data
Our generative simulation engine enables fast synthetic manipulation data generation with broad scenario coverage — allowing flexible changes to objects, robot embodiments, and environments to produce more realistic and complex simulation data.
World's Largest Rigid-Body Asset Library with Manipulation Annotations
100,000+ rigid-body assets with manipulation annotations — including grasp points, manipulable parts, and physical properties — ready for task planning and policy training.
Large-Scale Scene Assets
Multi-domain scene assets covering industrial manufacturing, modern logistics, and service applications, with on-demand customization to align with real-world business environments.
Support for Mainstream Robot Embodiments
30+ mainstream robot embodiments — including tabletop dual-arm, mobile manipulation, and humanoid robots — with cross-embodiment data synthesis and evaluation, and freely swappable configurations on demand.
High-Fidelity Physical Simulation
High-fidelity physical simulation is the core capability of X-Sim. By accurately reproducing key physical parameters such as material properties, lighting conditions, and friction, X-Sim ensures that simulation data can effectively transfer to real-world deployment. Based on this capability, X-Sim can infinitely generate diverse, fully annotated training data for challenging scenarios such as deformable object manipulation and fluid dynamics interactions, where traditional data collection is difficult, supporting policy transfer for complex tasks including grasping, assembly, folding, and articulation.
Provide Multimodal Perception Data
X-Sim supports the simulation and collection of multimodal perception data, including vision, tactile sensing, force sensing, and language. It covers the training input formats required by mainstream embodied AI models, providing comprehensive information for policy learning.
Scenario Customization Services
For personalized requirements beyond the coverage of standardized data products, we provide deep customization of high-fidelity scenarios.
- Customizable Scenarios and Assets: Fully controllable workstation layouts and operation SOPs
- Targeted Error Data Injection: Addressing edge cases in policy networks
- Joint Simulation and Real-World Data Development: Enabling closed-loop data iteration and optimization
AIGC-Powered Data Platform
X-Gen
Driven by AIGC-based video generation with an Agentic closed-loop quality assurance mechanism, our platform provides high-fidelity, high-diversity synthetic video data on demand for various physical manipulation tasks — covering both human and robotic operation modes.
Explore the X-Gen video generation platform now
Try It Now