Inside XRZero-G0, a new 2,000-hour open dataset for robotics research

To break the data bottleneck slowing down embodied AI, X Square Robot said it has made XRZero-G0 open-source. The company said its new hardware-software framework reduces real-robot training data requirements by up to 20× under experimental conditions.

Released alongside the G0-Dataset, a 2,000-hour multimodal repository, the system bridges the gap between human and machine perception by standardizing robot-free data collection, said X Square Robot. It said this allows human-demonstrated tasks to be reliably checked for quality and transferred to entirely unseen robotic platforms.

The company described XRZero-G0 as a comprehensive hardware-software framework designed to enhance scalable, high-quality, robot-free data collection and cross-embodiment policy transfer for dexterous robotic manipulation.

XRZero-G0 collects robot training data

XRZero-G0 features an ergonomic, wearable virtual reality interface with multi-view cameras and specialized dual grippers to decouple human mobility from robot kinematics. X Square Robot said the system:

• Uses a high-precision PICO 4 VR headset with inside-out spatial tracking
• Equipped with dual physical grippers: an H-shaped press-actuated and a G-shaped finger-driven gripper
• Supports millimeter-accurate 6-DoF pose estimation
• Incorporates edge-side spatiotemporal parsing for synchronization of visual, language, and trajectory data
• Ensures high collection throughput and stability, enabling sustained data capture without structural constraints

Data quality has been a critical barrier in robot-free learning, noted X Square Robot. It said XRZero-G0 formalizes trainability governance via a closed-loop “collection–inspection–training–evaluation” pipeline:

• Observation level: Multi-view geometric consistency suppresses visual-kinematic misalignment.
• Kinematic level: Full-body inverse kinematics with collision and joint-limit constraints filter invalid trajectories.
• Policy level: Real-robot playback serves as the final validation criterion.

---

---

X Square Robot validates

X Square Robot said it has completed controlled experiments to prove that combining approximately 10 robot-free episodes with one real-robot episode can achieve performance comparable with purely real-robot datasets in evaluated tasks.

The company has also scaled the G0-Dataset XRZero-G0 into a 2,000-hour dataset and open-sourced the result. The dataset integrates robot-free collection, automated quality inspection, mixed-data training, and real-robot evaluation for research purposes.

G0-Dataset supports large-scale pretraining and cross-embodiment transfer experiments, providing a reproducible open resource for robotics research, explained X Square Robot. By open-sourcing XRZero-G0 and releasing G0-Dataset, the company said it provides hardware designs, automated inspection pipelines, training methodologies, and high-quality datasets to the research community.

These resources are intended to accelerate the development of general-purpose robots and scalable embodied AI, supporting a transition toward more systematic and large-scale data generation approaches.

The full research paper is available for download. The code is available on GitHub, and the Open Dataset is available on HuggingFace.