Dobb·E
About Dobb·E
Dobb·E is an innovative platform designed to teach robots new household tasks through imitation learning. With features like the Stick for easy demonstration collection and the Homes of New York dataset, Dobb·E allows users to effortlessly train robots, improving household automation efficiency and capabilities.
Dobb·E offers an open-source software framework available at no charge. Users can access the entire platform, including hardware designs and models. There are no subscription tiers, promoting accessibility and collaboration for developers and researchers interested in advancing home robotics.
Dobb·E's user interface is designed for seamless navigation, featuring an intuitive layout that allows users to easily access resources, tools, and documentation. This ensures a smooth browsing experience, making it user-friendly even for those unfamiliar with robotics and imitation learning.
How Dobb·E works
Users interact with Dobb·E by first gathering demonstration data using the Stick, which collects task-related videos in various homes. After collecting five minutes of demonstrations, users can adapt the Home Pretrained Representations (HPR) model to a new environment, enabling robots to learn and execute tasks efficiently within 20 minutes.
Key Features for Dobb·E
Imitation Learning Tool
Dobb·E's imitation learning tool enables household robots to learn new tasks in just 20 minutes. By utilizing user-friendly demonstration collection via the Stick, this innovative feature allows for versatile training and adaptability, showcasing Dobb·E's potential in advancing home robotics.
Home Pretrained Representations (HPR)
Home Pretrained Representations (HPR) in Dobb·E enhances robotic learning by providing pre-trained models on diverse household tasks. This feature simplifies the onboarding process, allowing robots to perform new tasks effectively with minimal user input, significantly improving their usability and efficiency.
Dataset - Homes of New York (HoNY)
The Homes of New York (HoNY) dataset is a key feature of Dobb·E, comprising 13 hours of interactive data from various homes. It includes RGB and depth videos, enabling precise training of robots in realistic settings, thus enhancing their performance in real-life environments.