Dobb·E

Dobb·E is an open-source framework for teaching robots new household tasks via imitation learning. It solves the problem of enabling general-purpose robots to learn new tasks quickly in real home environments. The target audience includes developers, researchers, and robotics enthusiasts, as well as businesses and consumers interested in home robotics. It is positioned as an affordable and versatile system, using a low-cost demonstration collection tool and pre-trained representations, and is delivered as an open-source software stack with models, data, and hardware designs.

Key features

• Learn new tasks in 20 minutes
• Demonstration collection tool (The Stick)
• Home Pretrained Representations (HPR)
• Open-source software stack and models
• Dataset of 13 hours of interactions
• Hardware designs for The Stick