
TidyBot
Overview
TidyBot is an advanced mobile manipulator designed with household cleanup tasks in mind. It combines language-based planning and perception with the learning capabilities of large language models to quickly learn and generalize user preferences from just a few examples. This approach allows TidyBot to understand where different items should be placed according to each individual’s tastes and cultural background—whether that means putting shirts in a drawer or on a shelf. In controlled benchmarks, it has achieved high accuracy with unseen objects, and in real-world tests, it successfully organized 85% of the items it encountered.
In a domestic setting, TidyBot could serve as a personalized cleaning assistant that adapts to the unique organizational habits of each household. By accurately identifying and sorting items, it not only tidies up rooms but also creates storage solutions that align with the owner’s preferences. This capability makes it particularly useful for reducing clutter, streamlining daily routines, and supporting individuals who may have difficulty performing these tasks independently. As TidyBot learns and adapts over time, it promises to make household management more efficient and tailored to each user’s lifestyle.




Credits
Jeannette Bohg and the Interactive Perception and Robot Learning Lab
Zi-ang Cao, Jingyun Yang, Rika Antonova