US robotics infrastructure startup providing large-scale robot training data (teleoperation, egocentric data) for physical/embodied AI models.
XDOF is a US robotics infrastructure startup headquartered in San Mateo, California, founded in October 2024 by researchers with roots in UC Berkeley. The company builds data pipelines, teleoperation collection tools and annotation systems to produce large-scale, high-quality robot-manipulation training data for robot foundation models and physical (embodied) AI. XDOF acts as an outsourced data supply chain for frontier AI labs and robotics companies, combining remote robot teleoperation with data collected by human operators. In June 2026 it emerged from stealth with a $70M Series A (Thrive Capital, Spark Capital, a16z, Lux Capital, WndrCo). In September 2026 a Series B was reported in negotiation at a ~$1.2B valuation. Products include the open-source ABC-130K dataset (with UC Berkeley BAIR) and the GELLO teleoperation system.
Founders
Former UC Berkeley PhD researcher; creator of the GELLO teleoperation project.
UC Berkeley researcher.
Classification
External links
