In-the-Wild Embodied AI Data in Real Working Scenarios.
TerraByte Intelligence develops real-world data products for perception, reasoning, and automation—built to be affordable, scalable, and measurable in robot training.
Focus areas
"We are all in the gutter, but some of us are looking at the stars."
Real Scenarios
100% In-the-wild Working/Living Scenarios.
Cost-Effective
No more need to purchase expensive data collecting equipments or hire data collecting specialists.
Scalable
Direct collaboration with factory/company owners. 100,000+ reachable workers.
Now building
Perhaps the most important infrastructure for Embodied Intelligence.
Contact
Tell us what you’re building. We’ll respond quickly.
Tip: Click on the Pixel Bear!
Scenarios
Action Labeling
We use deep learning algorithms to annotate actions in videos, capturing the shape of the human hand, the pose of the wrist, and the joint angles of the fingers. Adopting a human-centric learning paradigm, we believe that humans are the measure of machines. As a unified cross-embodiment action representation, these action annotations can be used to train human hand trajectory models and robotic foundation models.
Data-Collecting Hardware
We will utilize a split-type binocular data capture system, consisting of lightweight recording glasses and a neckband computing unit, to capture highly synchronized, wide-angle binocular video from a first-person perspective. The device is purpose-built for video capture, meticulously optimized to deliver high image quality, low cost, and extended battery life for data collection.
