In a significant development, XDOF, a startup focused on gathering real-world teleoperation data for training versatile robots, is in advanced negotiations to secure a Series B funding round that could value the company at approximately $1.2 billion. The round is reportedly being led by 8VC, as detailed by sources familiar with the progress of the deal.
XDOF was established in 2024 by researchers from UC Berkeley, Philipp Wu (CEO) and Fred Shentu (CTO). The company previously attracted attention with a $70 million Series A funding round in June, involving contributions from investors like Thrive Capital, Andreessen Horowitz, Lux, and Spark Capital. Although XDOF did not plan to seek additional funding so soon, rapid growth—reportedly nearing $50 million in annualized revenue—has prompted interest from venture capitalists.
While specific details regarding the capital being raised or whether the valuation incorporates the new funding remain undisclosed, sources indicate that the terms are still under negotiation.
Neither XDOF nor 8VC has responded to requests for comments on this matter.
The startup’s mission revolves around developing data pipelines, collection tools, and annotation systems that are often beyond the capabilities of top AI labs and robotics firms. Basically, XDOF aims to serve as an outsourced data supply chain for the burgeoning robotics sector.
During his time as a PhD student, Wu encountered a significant limitation in his research: a scarcity of “large-scale data.” To address this, he collaborated with Shentu on a project called GELLO, a cost-effective teleoperation system enabling human operators to remotely control robotic arms for data generation—research that laid the groundwork for XDOF.
Investors have likened XDOF to Scale AI or Mercor for physical robotics, referring to pivotal data-labeling firms that have fueled the AI revolution. Unlike language models, which rely on vast internet datasets, physical robotics face a critical challenge in acquiring real-world data, making XDOF‘s service invaluable.
In partnership with UC Berkeley’s AI Research Lab, XDOF is preparing to launch an ambitious project that boasts the largest collection of high-quality robot training data, dubbed ABC.
To compile this data, the startup utilizes remote teleoperation together with human collectors who wear sensors to document everyday activities, such as folding clothes and flattening boxes.
Looking ahead, XDOF plans to recruit and train global teams of data collectors, encompassing teleoperators who control robots from afar and egocentric operators who collect movement data.
Currently, XDOF reports collaboration with 20 various clients, including several prominent AI labs.
Other players in the market include Mecka AI and human-data platforms extending their services beyond language models, such as Scale AI and Micro1.
Editor’s Take
This development underscores a growing recognition of the necessity for robust real-world training data in robotics. As companies like XDOF attract significant investment, it highlights an increasing focus on building effective data supply chains, which could streamline the robotics R&D process. Enhanced training capabilities will not only benefit developers but could lead to more sophisticated, reliable robotics solutions across various industries.
Source: techcrunch.com
