Robotic teleoperation data startup XDOF launches with $70M in funding
The AMW Read
Robotics data infrastructure is already a recognized subsegment; XDOF is a new entrant that incrementally expands the player map but does not resolve any open debate about the viability of teleoperation data at scale.
Robotic teleoperation data startup XDOF launches with $70M in funding
XDOF, a robotics training infrastructure startup, emerged from stealth today with $70 million in funding from Thrive Capital, Spark Capital, Andreessen Horowitz, Lux Capital, and WndrCo. The company also released ABC-130K, the largest open-source bimanual robot manipulation dataset, containing 130,000 trajectories, 300 hours of simulation data, and 100 hours of evaluations. Co-founder and CEO Philipp Wu, a UC Berkeley Ph.D. alumnus, built the company on the GELLO teleoperation system he helped develop, which enables human operators to control robotic arms remotely to generate precise training data for physical AI.
Why it matters: XDOF is tackling the data scarcity bottleneck that has stalled progress in robotics foundation models, a problem the company frames as more acute than model architecture or compute constraints. The $70 million round signals venture capital conviction that "physical AI" — after years of being overshadowed by language models — is entering a new capital cycle, with data infrastructure as the rate-limiting layer. The launch aligns with OpenAI's recent decision to revive its robotics program, reinforcing the pattern that frontier labs lack proprietary training pipelines for real-world manipulation tasks and must rely on external specialists.
Grounded expert take: XDOF's model represents a bet on "data-as-a-service" for robotics, a structural mirror of how Scale AI and Labelbox carved out data-labeling empires for vision and NLP. Yet the unit economics of teleoperation — requiring physical warehouses, robotic arms, wearable sensors, and human operators at scale — introduce a labor-intensive variable that past data startups avoided. The company's three-tier strategy (bespoke teleoperation, generalized teleoperation, egocentric human data) attempts to build a self-reinforcing feedback loop where each tier improves the next. If XDOF can demonstrate that its curated data unlocks transferable robot skills cheaper than synthetic data or simulation alone, it will validate a new segment within AI infrastructure.
#Robotics #DataInfrastructure #PhysicalAI #Teleoperation #VentureCapital #FoundationModels