Description
Dobb·E is an open source framework for teaching household robots everyday tasks by imitation learning. It has three parts: a low-cost demonstration tool called the Stick, built from cheap parts and an iPhone, which records how a person does a task; a dataset of homes in New York City, called Homes of New York, collected with the Stick; and a representation model, Home Pretrained Representations, trained on that data. The team reports that a new task can be learned from about five minutes of demonstrations in a new home, with around fifteen minutes of adapting the model, and that the system succeeded on 81 percent of tasks across 109 tasks in 10 homes on the Stretch mobile robot. The project's paper, videos, hardware designs and code are published, and the code is on GitHub. It is a research project aimed at roboticists and machine learning researchers.…
Highlights
- Teach a robot a task in minutes
- The Stick demonstration tool
- Homes of New York dataset
- Home Pretrained Representations model
- Open source code and hardware
Ratings & Reviews
No ratings yet
Be the first to rate Dobb.
People also visit
Dobb FAQ
What is Dobb·E?
It is an open source research framework for teaching robots household tasks by showing them how, using imitation learning.
What do I need to use it?
The project includes the Stick demonstration tool built from cheap parts and an iPhone, plus the code and models. The site also links the hardware and data collection code on GitHub.
How well does it work?
The site reports an 81 percent success rate across 109 tasks in 10 homes, with about 20 minutes to learn a new task.
Where is the code?
The code is on GitHub at notmahi/dobb-e, with folders for the hardware and the Stick data collection.
What are the best Dobb alternatives?
Popular alternatives to Dobb include Docs.nav2.org, Statbotics and Home-Assistant. See all Dobb alternatives






