Collecting robot training data is dirty, unglamorous work. Some AI labs are already paying XDOF to do it

XDOF, a startup emerging from stealth, is addressing the critical shortage of high-quality physical data needed to train robotics models. While major AI labs are increasingly focused on physical AI, they lack the specific data pipelines and teleoperation infrastructure required to teach machines to interact with the real world. XDOF aims to solve this by providing comprehensive data collection, cleaning, and annotation services, supported by $70 million in funding. The company has also partnered with UC Berkeley to release a massive dataset of robotic manipulation, seeking to establish a scalable ecosystem that allows AI labs to outsource the labor-intensive, physical work of robot training.

(Source：TechCrunch)