Northstar Robotics provides large‑scale, real‑world data collection services for training physical AI systems. By deploying capture teams equipped with head‑mounted RGB‑depth cameras and instrumented gloves, they record egocentric video and tactile signals from skilled workers across diverse production environments, enabling developers to train manipulation and navigation policies with authentic human‑level context.
Funding
Funding not disclosed
Founders
Product
Problem
Robotic learning systems require large amounts of high-quality, task-relevant data that captures both visual context and physical interaction. Existing datasets often lack egocentric perspective, depth information, or tactile signals, limiting the ability to train manipulation and navigation policies for real-world industrial and service robots.
Solution
Embodied collects and provides high-resolution egocentric video, synchronized RGB‑depth streams, and force/pressure data from instrumented gloves in active production environments such as factories, kitchens, and warehouses. Their capture teams record first‑person footage of skilled operators performing real tasks, delivering datasets that reflect authentic spatial context and contact dynamics. The data is organized and made available through a cloud infrastructure, enabling researchers and developers to train and validate manipulation policies, navigation systems, and general‑purpose robot controllers on realistic scenarios. By supplying both visual and tactile signals, Embodied bridges the gap between simulation and deployment, accelerating the development of physical AI for industrial and service applications.
Target Audience
Primary customers are robotics research labs, AI developers, and industrial automation companies building manipulation or navigation capabilities for robots operating in manufacturing, logistics, food service, and other service environments.
Features
- First‑person (egocentric) video captured with head‑mounted cameras, providing synchronized RGB and depth streams for full spatial awareness
- Instrumented gloves that record force and pressure signals, delivering tactile data unavailable from vision‑only sources
- Data collection across a wide range of real production settings, including construction, food processing, textile manufacturing, electronics assembly, logistics, healthcare, and more
- Curated datasets that pair visual and tactile modalities, supporting training of manipulation policies and navigation algorithms
- Cloud‑based data delivery platform that enables scalable access for research and development teams