What is Dobb·E?
Dobb·E is an affordable and versatile general-purpose system designed for learning robotic manipulation within household settings. It leverages a unique demonstration collection tool called "The Stick," built from inexpensive components, allowing users to quickly demonstrate tasks.
The system utilizes Home Pretrained Representations (HPR), trained on the Homes of New York (HoNY) dataset, to achieve a high success rate in various tasks. Dobb-E is open-source, providing access to software, models, data, and hardware designs to accelerate research in home robotics.
Features
- The Stick: A demonstration collection tool built from inexpensive components for easy task demonstration.
- Homes of New York (HoNY) Dataset: A dataset containing 13 hours of interactions at 22 different homes.
- Home Pretrained Representations (HPR): A model pre-trained on the HoNY dataset used to initialize a robot policy.
- Open-Source: Provides access to the software stack, models, data, and hardware designs.
Use Cases
- Learning new household tasks quickly.
- Adapting to novel home environments.
- Performing a variety of manipulation tasks.
- Collecting demonstration data for robots
