The OCD-Friendly Cleaning Robot

Robots designed to provide assistance in the home have long been a dream of many, and while they may not be commonplace fixtures in most people's homes just yet, with the present rate of advancement in the field of robotics, that goal may not be too far off. These helpful machines have the potential to revolutionize the way we live and perform daily tasks, freeing up our time and energy for more meaningful pursuits.

It is no wonder why we have been captivated by the idea of personal robot assistants for so long. Who wants to fold the laundry, clean the house, or cook dinner in their free time? There are dozens of routine, mundane tasks we all do everyday that take our time and energy away from more important or enjoyable aspects of life.

But for all the help that robots could provide us, there are a lot of people that cringe at the thought of a machine moving things around their home and not putting them back in exactly the right way. For those that are very particular about how they keep their homes, such a robot may feel more like it is a bull in a china shop than a valued assistant. There is no purpose in having a cleaning robot if you have to follow it around to clean up after it.

A team of researchers at Princeton University and Stanford University who apparently believe in the notion that there should be a place for everything and everything in its place have developed a cleaning robot that factors user preferences into its actions. Aptly named TidyBot, their robot collects a small amount of information from a user, then uses machine learning to turn that information into a set of guidelines that can guide the robot in a general way that can be applied to future scenarios. To prove the concept, the initial version of TidyBot focuses on picking up objects and putting them in their place.

One of the better known features of large language models (LLMs) like Google’s Bard or Open AI’s ChatGPT is their ability to summarize texts. The team leveraged LLMs to accept statements of user preferences, like “red colored clothes go in the drawer while white ones go in the closet.” After collecting a number of these statements, the model summarized them to provide a concise set of guidelines that could be used as guiding principles for TidyBot.

Language-based planning and perception algorithms were combined with the summarized guidelines to govern the robot’s actions. These methods were evaluated in a series of trials that were conducted to assess how well TidyBot could clean up a room, while respecting an individual user’s preferences. First, a benchmarking dataset was tested on the algorithm, and it was found to achieve better than 91% accuracy in dealing with previously unseen objects. Next, real world tests were conducted with TidyBot. In this case, the robot correctly put away 85% of objects in accordance with the user’s personal preferences.

With a bit more work, this research could lead to the development of robots that can help with all sorts of tasks around the house. But as it currently stands TidyBot is limited to some very specific activities, like picking up objects with a top-down grasp, then delivering them to a known location. The researchers believe that the development of a more advanced algorithm and perception system will be the key to a future filled with home assistant robots that do not drive us crazy.