Dobb-E
Dobb·E is an open-source framework designed for teaching robots household tasks through a process known as imitation learning. The framewor...
Last verified:
What is Dobb-E?
Dobb·E is an open-source robotic imitation learning framework for household manipulation. It is designed to help robots learn new tasks from a small amount of human demonstration, with the project claiming that new household tasks can be learned in about five minutes of training data. The system combines a low-cost demonstration collection tool called the Stick, a large household dataset called Homes of New York, and a pretrained vision representation model called HPR. The overall goal is to make home-robot learning practical, affordable, and fast for research and real-world household settings.
The website presents Dobb·E as a complete stack rather than a single model. The Stick is a hardware tool built from inexpensive parts to collect robot demonstrations comfortably in homes, while HoNY provides RGB-D data gathered across many homes and environments. HPR is the pretrained visual backbone used to initialize policies for new tasks, and the platform ties these pieces together for deployment in novel homes. The project emphasizes open-source access to software, models, data, and hardware designs.
Dobb·E is aimed at robotics researchers, engineers, and teams working on household manipulation, imitation learning, and embodied AI. It is especially relevant for people who want to prototype home-robot learning without building a full data-collection pipeline from scratch. The website also makes clear that the system is intended for simple household tasks rather than broad general-purpose autonomy. It is positioned as a research platform that can accelerate experimentation in real homes.
The strongest selling points are the low-cost demonstration setup, the curated home dataset, and the ability to adapt to new homes with limited training time. The website also highlights an 81% average success rate across 109 tasks in 10 homes, showing that the system is meant to work beyond a lab-only environment. Because Dobb·E is open-source, users can inspect the implementation and reuse the components for research. That makes it especially useful for labs that care about reproducibility and hardware-software co-design.
Dobb-E pricing
Pricing model: Free
The website describes Dobb·E as open-source and does not present paid plans. The core software, models, dataset, and hardware designs are made publicly available, and the project emphasizes low-cost components for the demonstration tool. Based on the site, there is no subscription pricing, free tier, or commercial plan listed. Access appears to be free for research and implementation, with costs mainly tied to assembling the Stick and any supporting hardware.
Dobb-E pros
- Open-source software stack
- Open-source hardware designs
- Open-source models
- Open-source dataset access
- Learn new tasks in about 5 minutes
- Low-cost demonstration tool
- Ergonomic demonstration collection
- Built for household manipulation
- Uses real home data
- RGB-D dataset included
- Covers 22 homes
- Covers 216 environments
- Pretrained vision representation model included
- Designed for novel homes
- Useful for robotics research
Dobb-E cons
- Focused mainly on simple household tasks
- Requires hardware setup
- Relies on a custom Stick device
- Needs human demonstrations
- Not a fully autonomous home robot
- Best suited to research users
- Limited to the provided home-task scope
- Performance depends on fine-tuning data
Frequently asked questions about Dobb-E
What is Dobb·E?
Dobb·E is an open-source robotic imitation learning framework for household manipulation. It combines a hardware demonstration tool, a home-robot dataset, and a pretrained visual model so robots can learn tasks from a small amount of human teaching.
What does the Stick do?
The Stick is the hardware tool used to collect robotic demonstrations in homes. It is designed to be affordable and comfortable to use, and the project says it can be built from low-cost parts to support data collection in real household environments.
What is the Homes of New York dataset?
Homes of New York, or HoNY, is the dataset used by Dobb·E. The website describes it as a large home dataset with RGB-D frames collected across 22 homes and 216 environments in New York City.
What is HPR?
HPR stands for Home Pretrained Representations. It is the pretrained lightweight vision model in the Dobb·E system, trained on the HoNY dataset and used to help initialize robot policies for new tasks.
How quickly can Dobb·E learn a new task?
The website says Dobb·E can learn new household tasks in about five minutes of training data, followed by about 15 minutes of fine-tuning for HPR. The project presents this as a fast way to adapt to new home settings.
What kinds of tasks is Dobb·E meant for?
Dobb·E is meant for simple household manipulation tasks, such as home-robot actions that can be learned from demonstrations. It is not presented as a general-purpose autonomous system for all robot behaviors.
Is Dobb·E open source?
Yes. The project states that its software stack, models, data, and hardware designs are open source and publicly released to support research and reproducibility.
Who is Dobb·E for?
Dobb·E is aimed at robotics researchers, engineers, and teams working on imitation learning, household robotics, and embodied AI. It is especially useful for people who want to study or prototype learning in real home environments.
What evidence does the website give for performance?
The site highlights an 81% success rate across 109 tasks in 10 homes. This is presented as evidence that the system can generalize to novel household settings beyond a controlled lab environment.
Does Dobb·E list pricing plans?
No pricing plans are shown on the website. The project is described as open-source, and the site focuses on access to the system components rather than subscription tiers or paid plans.