Dobb-E

dobb·e is an innovative open-source framework that makes it easy to teach robots how to perform household tasks. Imagine a robot that can learn by watching you! That's what dobb·e enables. It simplifies the process of collecting data to train robots, making it more affordable and accessible than ever before.

Highlights

• Affordable and Ergonomic: dobb·e uses a simple and cost-effective "stick" tool made from a reacher grabber stick, 3D printed parts, and an iPhone, to collect demonstration data.
• Efficient Learning: dobb·e leverages a large dataset of real-world home interactions, called "Homes of New York" (HONY), to train its robots.
• Fast and Effective: dobb·e can train a robot to perform a new task in minutes, achieving high success rates with just a few minutes of data collection in a new environment.

Key Features

• Data Collection Tool ("Stick"): Provides an easy and affordable way to record demonstrations of household tasks.
• HONY Dataset: A comprehensive dataset of real-world home interactions, including videos and annotations, used to train robots.
• Home Pretrained Representations (HPR): A powerful representation learning model based on ResNet architecture that allows robots to quickly learn new tasks.
• Open Access: dobb·e offers access to pre-trained models, code, and documentation through GitHub, along with an open-access research paper detailing the framework's approach.