Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and the University of Toronto are teaching AI to do household chore using a Sims inspired virtual simulator. While household chores are dreary, if not particularly complex for humans, for AI, making a cup of coffee or turning on the toaster is a bit more involved. Watching TV from the sofa, for instance, requires an AI to break the task down to specific steps, like: “walk to the TV,” “switch on the TV,” “walk to the sofa,” and “sit on the sofa.”
To accomplish household tasks, researchers collected common verbal descriptions of household chores, then translated into robot friendly code. Then they showcased the AI house assistants in a 3D VirtualHome stimulator that looks very much like the Sims. Not only did the virtual robot agents successfully make coffee, turn on the toaster and relax on the sofa, but researchers have created a database of household tasks described through natural language, which could help systems like Amazon's Alexa to carry out more complex tasks.
In the future, robot assistants could even learn through watching Youtube videos instead of stimulation videos. Researchers also want to implement a rewards system that gives the AI positive feedback when it accomplishes tasks correctly. In the future, robot assistants might be able to learn and carry out complex multi-step tasks just by watching a human accomplish them.
“You can imagine a setting where robots are assisting with chores at home and can eventually anticipate personalized wants and needs, or impending action,” said PhD student Xavier Puig, lead author of the research. “This could be especially helpful as an assistive technology for the elderly, or those who may have limited mobility.”
The team’s AI agent have already learned to execute 1,000 separate sets of actions in eight different scenes, including a living room, kitchen, dining room, bedroom, and home office.