XDOF, a robotics data infrastructure startup emerging from stealth in October 2024, has secured $70 million in funding to address a critical bottleneck in physical artificial intelligence. As major technology companies accelerate their robotics initiatives following OpenAI’s recent program relaunch, a severe shortage of high-quality training data comparable to the datasets that powered large language models has emerged. XDOF was founded to build the data pipelines, collection hardware, and annotation systems required to train robots for real-world physical interaction. Co-founded by UC Berkeley alumni Philipp Wu, Fred Shentu, and Nemo Jin, the company operates on the premise that frontier AI laboratories lack the operational scale and capital to independently manage massive robotics data operations. Building a data feedback loop requires extensive warehouse infrastructure, continuous robot calibration, and a globally distributed workforce of teleoperators and egocentric data collectors. XDOF plans to standardize this labor-intensive process, allowing AI labs to focus on model development while outsourcing the physical data infrastructure. The company will deploy a three-tier data acquisition strategy, prioritizing teleoperation data captured directly on target robots, followed by broader teleoperated datasets derived from their GELLO open-source system, and finally egocentric data gathered via custom wearable sensors. To demonstrate its capabilities, XDOF is partnering with UC Berkeley’s AI Research lab to release the ABC dataset, comprising 130,000 robot manipulation trajectories, 300 hours of simulation, and 100 hours of evaluation metrics. This release represents the largest curated collection of high-quality robotics training data available to the research community. Already serving 20 unnamed customers, including several leading AI research groups, XDOF has attracted investment from Thrive Capital, Spark Capital, a16z, Lux, and WndrCo. The company name references the robotics concept of degrees of freedom, with its leadership emphasizing a commitment to enabling arbitrary motion capabilities across physical AI systems. By standardizing data collection and cleaning, XDOF aims to eliminate the historical lag between robot hardware advancements and software training requirements, positioning itself as an essential infrastructure provider in the race toward general-purpose robotics.