Collecting robot training data is dirty, unglamorous work. Some AI labs are already paying XDOF to do it.
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- XDOF emerged from stealth with $70 million from Thrive Capital, Spark Capital, a16z, Lux, and WndrCo, and is already working with 20 customers including several frontier AI labs, CEO Philipp Wu told TechCrunch
- XDOF was founded in October 2024 by Wu, CTO Fred Shentu, and COO Nemo Jin — all three previously built the GELLO teleoperation system at UC Berkeley to address what Wu called a chicken-and-egg data shortage in robotics
- XDOF is partnering with UC Berkeley's AI Research lab to release the ABC dataset, which it calls the largest collection of high-quality robot training data ever assembled: 130,000 trajectories, 300 hours of simulation, and 100 hours of evaluations
- OpenAI said two weeks ago it would relaunch the robotics program it shuttered in 2021, and Wu told TechCrunch that 'all of the top labs are trying to pursue robotics' because no one wants to fall behind as physical AI becomes the next frontier
- XDOF plans to work across three tiers of data — teleoperation on the deployed robot, teleoperated general data, and egocentric data from humans wearing its own custom-built sensors, since camera choice affects hand-tracking algorithm quality
- Wu said labs outsource the work because it requires warehouses of hundreds of thousands of square feet, hundreds of robots, calibration, and trained operators — an operational scale most AI labs would rather not build in-house
Why it matters: Frontier AI labs are now publicly racing into physical-world robotics, but none want to build the teleoperation warehouses and human labor pools required to feed the models — XDOF has positioned itself as the dedicated infrastructure vendor for that data supply chain, with $70M, 20 signed customers, and 60 employees already in motion.
