Robot training data startup XDOF is in late-stage talks to raise a Series B at a valuation of about $1.2 billion, less than three months after emerging from stealth. The round is expected to be led by 8VC, according to people familiar with the discussions, as XDOF’s annualized revenue approaches $50 million.
The total amount being raised remains unclear, as does whether the reported valuation includes the new capital. The terms are not final and could still change, while XDOF and 8VC did not respond to requests for comment.
XDOF was founded in 2024 by UC Berkeley researchers Philipp Wu, its CEO, and Fred Shentu, its CTO. The company formally introduced its robotics infrastructure business in an official announcement in June.
The startup raised $70 million in Series A funding that month, with Thrive Capital, Andreessen Horowitz, Lux and Spark Capital participating. XDOF had not planned to raise another round so quickly, but investors approached the company as its business grew, according to people familiar with the talks.
Building Training Data for General-Purpose Robots
XDOF develops data pipelines, collection tools and annotation systems for AI labs and robotics companies. Its business addresses a key difference between large language models and physical robots: robots do not have an internet-scale collection of real-world interactions available for training.
The company traces its origins to GELLO, a low-cost teleoperation system developed by Wu and Shentu while conducting robotics research. GELLO allows a person to remotely control a robotic arm while recording the interactions as training data.
XDOF now combines remote teleoperation with data collected by people wearing sensors while completing physical tasks. Examples include folding clothes and flattening boxes, with the resulting movements used to train robotic systems.
The company plans to hire and train data collectors worldwide. Its workforce will include teleoperators who remotely control robots and egocentric operators who wear sensors to record their movements.
XDOF Expands Its Robot Data Collection
XDOF previously said it was working with 20 customers, including several frontier AI labs. It is also working with UC Berkeley researchers on ABC-130K, an open-source collection containing more than 130,000 robot manipulation episodes across 195 tasks.
Other companies are also collecting real-world data for robotics training, including Mecka AI. Data platforms that initially focused on training large language models, including Scale AI and Micro1, have also expanded their work into physical-world data.
Featured image credits: XDOF
For more stories like it, click the +Follow button at the top of this page to follow us.