Dobb-E

Overview of Dobb-E

Dobb-E is an innovative framework in the realm of household robotics, specifically designed to teach robots how to perform various household tasks through imitation learning. This approach allows robots to learn from human demonstrations, making it easier to adapt to the complexities of home environments.

Core Technology and Methodology

At the heart of Dobb-E is a unique tool known as the Stick, which combines a $25 Reacher-grabber stick with 3D printed components and an iPhone. This setup enables the collection of demonstration data in a cost-effective and ergonomic manner. The data is sourced from a comprehensive dataset called Homes of New York (HoNY), which features 13 hours of interactions across 22 different households in New York City. This dataset includes RGB and depth videos, along with detailed action annotations that capture the gripper's 6D pose and opening angle.

Training and Learning Capabilities

Dobb-E utilizes the collected data to train a representation learning model named Home Pretrained Representations (HPR). This model is based on the ResNet-34 architecture and employs self-supervised learning techniques to initialize a robot policy capable of executing tasks in new environments. The framework has shown promising results, achieving an average success rate of 81% in solving novel tasks within just 15 minutes, using only five minutes of demonstration data.

Accessibility and Resources

Dobb-E is designed to be accessible for developers and researchers interested in household robotics. The framework is open-source, providing users with access to pre-trained models, source code, and comprehensive documentation through GitHub. Additionally, an open-access research paper titled "On Bringing Robots Home" offers deeper insights into the framework's methodology and findings.

Potential Users and Applications

This tool is particularly beneficial for researchers, developers, and hobbyists in the field of robotics who are looking to explore household automation. Its focus on imitation learning and the use of affordable tools make it an attractive option for those aiming to develop or enhance robotic capabilities in domestic settings.