XDOF in Talks for $1.2B Series B After Three Months

XDOF is in late-stage talks to raise a Series B funding round at a valuation of about $1.2 billion. The startup, which collects real-world teleoperation data for training general-purpose robots, emerged from stealth less than three months ago, according to several people with knowledge of the deal who spoke to TechCrunch. The round is reportedly being led by venture capital firm 8VC.

Founded in 2024 by UC Berkeley researchers Philipp Wu and Fred Shentu, XDOF wasn't planning to raise again so soon after a $70 million Series A in June. That earlier round saw participation from Thrive Capital, Andreessen Horowitz, Lux, and Spark Capital. The company's rapid growth, with annualized revenue approaching $50 million, prompted venture capitalists to approach it about a new investment. The total capital being raised and whether the valuation includes the new funding are not yet known, and the terms are not final.

XDOF and 8VC did not respond to requests for comment from the source.

The Data Bottleneck for Robotics

The startup's core mission is to build the data pipelines, collection tools, and annotation systems that frontier AI labs and robotics companies struggle to create internally. It aims to act as an outsourced data-supply chain for the industry. CEO Philipp Wu, as a PhD student, identified a major impediment to his research on how robots learn: the lack of large-scale data to work with.

This challenge led to the creation of a project called GELLO, a low-cost teleoperation system developed with CTO Fred Shentu. GELLO allows a human operator to control a robotic arm remotely to generate training data. Their work resulted in an influential robotics paper and formed the foundation for XDOF.

Investors now describe the company as the Scale AI or Mercor for physical robotics. Unlike large language models, which initially trained on vast internet data, physical robots lack an equivalent real-world dataset. This makes data collection a critical bottleneck for developing general-purpose machines.

Building the ABC Dataset

XDOF is partnering with UC Berkeley's AI Research lab to release what it believes is the largest collection of high-quality robot training data ever assembled. This dataset is dubbed ABC. To capture the data, the company combines remote robot teleoperation with human collectors who wear sensors to record everyday tasks. These tasks include activities like folding clothes and flattening boxes.

The startup plans to hire and train teams of data collectors worldwide. These teams will include teleoperators who steer robots remotely and egocentric operators who wear body sensors to capture movement data.

Market Context and Competition

XDOF has previously stated it is already working with 20 customers, including several frontier AI labs. The market for robot training data is attracting other players. Other startups attempting to collect real-world data for this purpose include Mecka AI. Also, established human-data platforms that initially focused on large language models, such as Scale AI and Micro1, are also expanding into this physical data domain.