XDOF Raises $70M for Physical AI Data Infrastructure

A significant capital injection is set to accelerate the development of crucial data infrastructure for physical artificial intelligence. XDOF, a startup focused on robotic teleoperation data, has successfully closed a $70 million funding round. This substantial investment was led by prominent venture capital firms including Thrive Capital, Spark Capital, Andreessen Horowitz, Lux Capital, and WndrCo, signaling strong market confidence in the company's mission.

The core challenge XDOF aims to address is the scarcity of high-quality data required to train robots for real-world tasks. Unlike large language models that thrive on vast internet datasets, physical AI demands nuanced, action-specific information. This funding will empower XDOF to build out the specialized data pipelines, collection tools, and annotation systems essential for bridging this critical gap. The company's efforts are particularly timely, coinciding with renewed industry interest in "physical AI" and the complex data requirements that accompany it.

To underscore its capabilities and provide a foundational resource for the research community, XDOF simultaneously unveiled ABC-130K. This initiative represents what the company claims is the world's most extensive open-source dataset for bimanual robot manipulation. The dataset offers researchers unprecedented access to high-fidelity training data, a vital component for advancing robotics. This move directly tackles the "chicken-and-egg" problem that has historically hindered progress, where the lack of data prevented the development of models capable of collecting more data.

The genesis of XDOF lies in the founders' firsthand experience with data limitations during academic research. Co-founder and CEO Philipp Wu highlighted the difficulties encountered while pursuing his Ph.D. at the University of California at Berkeley, where the absence of large-scale datasets impeded progress. This realization fueled the development of XDOF's innovative approach, which moves beyond simply generating data to offering comprehensive data cleaning, annotation services, and specialized tooling, creating a self-sustaining ecosystem.

XDOF's strategic vision for scaling involves a three-tiered data pyramid. This structure encompasses bespoke teleoperation data derived from direct remote control of specific robots, generalized teleoperational data, and "egocentric" data captured by humans performing everyday tasks. This comprehensive data collection strategy requires significant operational capacity, including a global network of teleoperators and data gatherers, and potentially proprietary wearable sensors to ensure precise hand-tracking algorithm alignment. The company estimates that establishing the necessary infrastructure, such as large warehouses filled with robots and trained personnel, is a substantial undertaking that many AI labs would prefer to outsource.

With a team of over 60 employees and approximately 20 existing customers, including leading frontier AI labs, XDOF is already demonstrating traction. The company's focus on creating this specialized, labor-intensive data infrastructure positions it as a key enabler for the broader physical AI sector, which is projected for significant growth as robots become more integrated into various industries, from manufacturing and logistics to healthcare and domestic assistance.