No longer up to three months after emerging from stealth, XDOF, a startup that collects valid-world teleoperation data for practicing fashioned-motive robots, is in unhurried-stage talks to know a Sequence B at a valuation of about $1.2 billion valuation led by 8VC, several folk with data of the deal mentioned.
XDOF become co-based 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 know again so quickly after that round. However the company’s hasty enhance — with annualized earnings drawing near $50 million — introduced about VCs to come it a pair of recent round, the folk mentioned.
TechCrunch become unable to learn the complete capital being raised or whether the valuation contains the recent funding. The terms of the deal are not closing and would possibly well per chance mute commerce.
XDOF and 8VC didn’t answer to our query for comment.
The startup aims to accumulate the info pipelines, sequence tools, and annotation systems that frontier AI labs and robotics companies can’t easily accumulate themselves, truly acting as an outsourced data-provide chain for the robotics industry.
As a PhD pupil, Wu become discovering out how robots learn from massive datasets. One massive impediment to his examine become the dearth of “massive-scale data to work with,” he told TechCrunch in June.
So he teamed up with Shentu on a project known as GELLO, a low-set aside teleoperation design that permits a human operator to manipulate a robotic arm remotely in voice to generate practicing data. Their work ended in an influential paper in robotics.
That examine fashioned the muse for XDOF, which merchants now portray because the Scale AI or Mercor for bodily robotics, a reference to the info-labeling giants that helped fuel the AI boom. No longer like LLMs, which originally trained on the whole lot of the on-line, bodily robots don’t have an equal valid-world dataset to plan from, making data sequence a valuable bottleneck to building fashioned-motive machines.
XDOF is partnering with UC Berkeley’s AI Research lab to open what it believes is the splendid sequence of excessive-quality robotic practicing data ever assembled, dubbed ABC.
To grab this data, XDOF combines faraway robotic teleoperation with human collectors who set aside on sensors to file day to day obligations esteem folding apparel and flattening bins.
The startup plans to hire and prepare teams of data collectors worldwide, together with teleoperators who steer robots remotely and egocentric operators who set aside on physique sensors to grab stream data.
XDOF previously told TechCrunch that it is a long way already working with 20 prospects, together with several frontier AI labs.
Varied startups attempting to fetch valid-world data for robotic practicing embody Mecka AI, to boot to human-data platforms rising previous LLMs, such as Scale AI and Micro1.
For fogeys that aquire thru links in our articles, we can also develop a diminutive rate. This doesn’t have an effect on our editorial independence.
Marina Temkin is a enterprise capital and startups reporter at TechCrunch. Earlier than joining TechCrunch, she wrote about VC for PitchBook and Endeavor Capital Journal. Earlier in her occupation, Marina become a monetary analyst and earned a CFA charterholder designation.
You would possibly well per chance also contact or test outreach from Marina by emailing marina.temkin@techcrunch.com or by capability of encrypted message at +1 347-683-3909 on Label.
Leer Bio
