Open X-Embodiment: the first generalist cross-robot robotics dataset
In one sentence Google DeepMind and 33 labs collect 527k episodes from 22 different robots: the first unified dataset for training generalist policies that work across multiple platforms.
One of the biggest problems in robotics is that every robot needs its own training dataset: a robot trained on an industrial arm knows nothing about a different mobile robot. Open X-Embodiment breaks this barrier by aggregating data from 22 different robotic platforms.
The collaboration between Google DeepMind and 33 academic and industrial labs produces 527,000 robot-environment interaction episodes, with tasks ranging from object manipulation to navigation. The data is standardized in a common format and made publicly available.
With this data, RT-X is trained, a model showing emergent capabilities on robots it never saw during training, demonstrating significant cross-embodiment transfer for the first time.
Companies
Google DeepMind
Tools
RT-X, RT-2, Open X-Embodiment Dataset
Tags
Sources