The study presents the Open-World Mobile Manipulation System. This system enables robots to interact with articulated objects such as doors, cabinets, and refrigerators in unstructured environments like homes. The robot uses an adaptive learning framework to learn from a limited set of data and subsequently from online practice on novel objects. The researchers also developed a cost-effective mobile manipulation hardware platform for safe and autonomous online adaptation. The system showed a significant increase in success rate from 50% to 95% with less than an hour of online learning for each object.
Publication date: 25 Jan 2024
Project Page: https://open-world-mobilemanip.github.io/
Paper: https://arxiv.org/pdf/2401.14403
