Generalist’s GEN-1 foundation model now supports a range of robot end effectors

A foundation model for robotics has been updated to work with many different types of robot hands and tools, from five-fingered hands to specialized equipment with various actuation methods. The model was trained on over half a million hours of real interaction data across approximately 9,000 variations of end effectors, exposing it to a broad range of contact physics and different ways of interacting with the physical world. By learning across many different hands and tools, the model develops universal physical understanding that transfers to new end effectors, allowing it to recognize which tool is attached and adapt its behavior mid-task when end effectors are switched. This approach mirrors how training language models on multiple languages produces more capable systems, with the model learning what is universal about physics versus what is specific to each tool.