XDOF, simply three months out of stealth, is in talks for a Sequence B at a $1.2B valuation

Lower than three months after rising from stealth, XDOF, a startup that collects real-world teleoperation information for coaching general-purpose robots, is in late-stage talks to lift a Sequence B at a valuation of about $1.2 billion valuation led by 8VC, a number of folks with information of the deal stated.

XDOF was co-founded by UC Berkeley researchers Philipp Wu (CEO) and Fred Shentu (CTO) in 2024. TechCrunch reported on the startup’s $70 million Sequence A in June, with participation from Thrive Capital, Andreessen Horowitz, Lux, and Spark Capital. XDOF wasn’t planning to lift once more so quickly after that spherical. However the firm’s speedy development — with annualized income approaching $50 million — prompted VCs to method it a few new spherical, the folks stated.

TechCrunch was unable to be taught the entire capital being raised or whether or not the valuation consists of the brand new funding. The phrases of the deal usually are not ultimate and will nonetheless change.

XDOF and 8VC didn’t reply to our request for remark.

The startup goals to construct the information pipelines, assortment instruments, and annotation techniques that frontier AI labs and robotics firms can’t simply construct themselves, basically performing as an outsourced data-supply chain for the robotics business.

As a PhD scholar, Wu was finding out how robots be taught from massive datasets. One large obstacle to his analysis was the dearth of “large-scale information to work with,” he instructed TechCrunch in June.

So he teamed up with Shentu on a venture known as GELLO, a low-cost teleoperation system that enables a human operator to regulate a robotic arm remotely with a view to generate coaching information. Their work led to an influential paper in robotics.

That analysis shaped the inspiration for XDOF, which buyers now describe because the Scale AI or Mercor for bodily robotics, a reference to the data-labeling giants that helped gasoline the AI growth. In contrast to LLMs, which initially educated on everything of the web, bodily robots don’t have an equal real-world dataset to attract from, making information assortment a vital bottleneck to constructing general-purpose machines.

XDOF is partnering with UC Berkeley’s AI Analysis lab to launch what it believes is the most important assortment of high-quality robotic coaching information ever assembled, dubbed ABC. 

To seize this information, XDOF combines distant robotic teleoperation with human collectors who put on sensors to file on a regular basis duties like folding garments and flattening containers.

The startup plans to rent and prepare groups of knowledge collectors worldwide, together with teleoperators who steer robots remotely and selfish operators who put on physique sensors to seize motion information.  

XDOF beforehand instructed TechCrunch that it’s already working with 20 clients, together with a number of frontier AI labs.

Different startups making an attempt to gather real-world information for robotic coaching embrace Mecka AI, in addition to human-data platforms increasing past LLMs, similar to Scale AI and Micro1.