Dobb·E: Open-Source Framework for Household Robotics

Dobb·E is an open-source framework revolutionizing home robotics by enabling robots to learn household tasks through imitation learning. It overcomes the cost and complexity barriers of existing solutions by using a low-cost, ergonomic data collection tool.

Features

• Affordable Data Collection: Utilizes a "Stick" – a device built from readily available components (a $25 Reacher-grabber, 3D printed parts, and an iPhone) – for efficient and inexpensive data acquisition.
• Large-Scale Dataset: Leverages the Homes of New York (HoNY) dataset, comprising 13 hours of interactions across 22 diverse NYC homes. This dataset includes rich RGB and depth videos, along with precise action annotations.
• Powerful Representation Learning: Employs a Home Pretrained Representations (HPR) model, based on ResNet-34 architecture and trained via self-supervised learning, to enable fast adaptation to new environments.
• High Success Rate: Demonstrates an impressive 81% average success rate in tackling novel tasks within 15 minutes, using only five minutes of new home data.
• Open Access: Provides readily available pre-trained models, code, and documentation via GitHub, alongside an open-access research paper ("On Bringing Robots Home").

Benefits

• Reduced Development Costs: The low-cost hardware and open-source nature significantly lower the barrier to entry for developing home robots.
• Rapid Adaptation: The HPR model allows robots to quickly learn and perform new tasks in unfamiliar environments.
• Improved Ergonomics: The Stick's design simplifies the demonstration process, making data collection more user-friendly.
• Accessible Research: Open access to the framework, dataset, and research paper fosters community collaboration and advancement in the field.

Use Cases

• Developing household robots: Enables researchers and developers to build robots capable of performing a wide range of household chores.
• Improving robot learning algorithms: Provides a valuable resource for testing and benchmarking new imitation learning techniques.

Dobb·E empowers the development of affordable and adaptable home robots through its innovative open-source approach to imitation learning.