The article discusses the development of a versatile, cost-effective robotic system named Dobb E which can learn and perform tasks in a home environment. The system learns new tasks in just five minutes using a demonstration collection tool constructed from inexpensive parts and iPhones. The study involved collecting data from 22 homes in New York City and training Home Pretrained Representations (HPR). The system was tested in 10 different homes, performing 109 tasks with an overall success rate of 81%. The experiments revealed unique challenges in home robotics like strong shadows and variable demonstration quality by non-expert users. The Dobb E software stack, models, data, and hardware designs have been made open-source.
Publication date: 27 Nov 2023
Project Page: https://dobb-e.com
Paper: https://arxiv.org/pdf/2311.16098
