XDOF Launches to Provide Essential Data for Robotics Training Amid AI Labs' Push

OpenAI is relaunching its robotics program, highlighting the need for training data to develop capable robots.

XDOF, a startup founded by Philippe Wu, Fred Shentu, and Nemo Jin, aims to fill this gap by building data pipelines and collection tools for robotics companies. The company has raised $70 million from investors including Thrive Capital and a16z.

XDOF is collaborating with UC Berkeley’s AI Research lab to release a large dataset called ABC, which includes 130,000 robot manipulation trajectories and extensive simulation data.

The startup plans to create a data ecosystem that includes teleoperation data, general data from teleoperated robots, and egocentric data from human activities. This approach is designed to enhance the training of robots and improve their performance in real-world tasks.

XDOF's business model focuses on outsourcing data collection and processing, which requires significant resources and expertise that many AI labs prefer not to manage themselves. The name XDOF reflects the company's ambition to enable robots with a wide range of motion capabilities.