Less than three months after exiting stealth mode, XDOF, a robotics data startup founded by UC Berkeley researchers, is in advanced discussions to raise a Series B round valuing the company at approximately $1.2 billion. The funding effort is being led by venture firm 8VC, according to individuals familiar with the negotiations. The rapid pace of fundraising follows the company’s recent public emergence after operating in secrecy. XDOF specializes in gathering real-world teleoperation data to train general-purpose robots, a niche area critical for advancing robotic learning and adaptability. The startup’s technology captures human-guided robot movements in physical environments, creating datasets that help machines learn complex tasks more efficiently.
This approach addresses a key bottleneck in robotics: the need for diverse, high-quality real-world data to train AI models that can generalize across different settings and functions. How XFOF’s Data Collection Approach Differs from Simulation-Based Training Unlike many robotics firms that rely heavily on simulated environments for training, XDOF focuses on collecting data from actual physical interactions, which reduces the reality gap often seen when transferring skills from simulation to real robots. The company’s founders, including PhD researchers from UC Berkeley’s robotics lab, argue that real-world teleoperation captures subtle nuances in force, friction, and environmental variability that simulations struggle to replicate.
This data is then used to train foundation models capable of powering a wide range of robotic applications, from manufacturing to logistics. What Challenges Could Arise from Scaling Real-World Data Collection Scaling teleoperation-based data collection presents logistical and financial challenges, including the need for human operators, specialized hardware, and safe testing environments. Critics point out that this method may be slower and more expensive than simulation-heavy alternatives, potentially limiting the volume of data that can be generated. However, XDOF’s investors appear to believe that the quality and applicability of its datasets justify the higher costs, especially as demand grows for robots that can operate reliably in unstructured, dynamic environments. Frequently Asked Questions What is teleoperation data and why is it valuable for robotics? Teleoperation data consists of recordings of human-controlled robot movements in real-world settings, capturing detailed sensory and motion information.
This data helps train AI models to understand and replicate complex physical tasks, improving robot performance in unpredictable environments. How does XDOF plan to use the funds from its Series B round? While specific allocations have not been disclosed, the company is expected to use the funding to expand its data collection operations, hire additional engineering and research staff, and further develop its machine learning infrastructure for training general-purpose robotics models. Is XDOF planning to sell its data or offer it as a service to other robotics companies? XDOF has not publicly confirmed its business model, but industry observers suggest it may either license its datasets or provide access to trained models through partnerships, enabling other firms to leverage its real-world data without building similar collection systems from scratch.