Insider Brief
- XDOF has emerged from stealth with $70 million in funding to build infrastructure for robot foundation models, including datasets, robotic systems and software tools designed to support physical AI development.
- The company said it works with robotics labs and commercial developers on hardware, operations and model training, focusing on the underlying infrastructure needed to build more capable general-purpose robots rather than developing robots for a single task.
- As part of its launch, XDOF released ABC-130K, which it describes as the largest open-source teleoperation dataset, containing more than 130,000 demonstrations across 195 bimanual manipulation tasks developed in collaboration with researchers from UC Berkeley, Carnegie Mellon, MIT and Amazon.
XDOF emerged from stealth with $70 million in funding to develop the infrastructure for robot foundation models.
The company is developing datasets, robotic systems and software tools designed to help robotics companies and research labs build more capable physical AI systems. Backers include Thrive Capital, Andreessen Horowitz, Spark Capital, Lux and WnderCo, according to XDOF.
“My co-founders Yide Shentu, Nemo Jin, and I have been pursuing the dream of general purpose robots for our entire lives,” co-founder Philipp Wu noted in a post on LinkedIn. “It’s clear to us that general purpose robots are coming, and we are building XDOF to help make them a reality.”
The company said it has already been working with robotics developers behind the scenes, providing support across hardware, operations and model training as interest grows in general-purpose robots.
Rather than building robots for a specific task, XDOF said it is focused on the infrastructure layer that underpins robot development. The company stressed that advances in AI have transformed digital work but that many everyday physical tasks remain difficult to automate, creating an opportunity for more capable robotic systems.
As its first public release, XDOF announced the open-source launch of ABC-130K, which it describes as the largest open-source teleoperation dataset. The project was developed in collaboration with researchers from the UC Berkeley, Carnegie Mellon, MIT and Amazon’s Frontier AI & Robotics group.
