toolspool
Dobb-E logo

Dobb-E

verifiedFreedobb-e.com

Open-source research framework for teaching home robots new manipulation tasks from a few minutes of demonstration.

What it does

Dobb-E is an open-source, general-purpose framework for learning household robotic manipulation, developed by researchers at NYU and collaborators. Using a cheap iPhone-based demonstration tool called The Stick, it collects home demonstrations and trains Home Pretrained Representations, letting a mobile robot learn a new task from about five minutes of demonstration. It reports an 81% success rate across 109 tasks in 10 NYC homes.

Core features

Learns new tasks from ~5 minutes of demonstration
'The Stick' low-cost iPhone demonstration tool
Home Pretrained Representations (HPR) models
Runs on the commercially available Stretch robot
Fully open-source software, data, models and hardware

Best for

Research into home/household robot manipulation
Rapidly teaching robots new tasks by demonstration
Benchmarking imitation learning in real homes