Dobb-E

Dobb-E is an open-source framework designed to enable learning household robotic manipulation through imitation learning. It provides tools, including a novel demonstration collection hardware ('the Stick'), a large real-world interaction dataset (Homes of New York), and a pretrained model for initializing and training robotic policies. Dobb-E is aimed at robotics researchers, developers, and engineers focused on real-world household robotic tasks, data collection, and policy learning.