MIT researchers are training virtual AI agents to complete helpful tasks and chores around a virtual home, with the possibility of using the same method to train physical robots in the future. In a 3-D world reminiscent of the Sims video game, the agents carry out tasks like making coffee, retrieving objects and using a TV remote.
The system is called “VirtualHome” and is described in a paper called “VirtualHome: Simulating Household Activities via Programs,” which will be presented at the June 2018 Computer Vision and Pattern Recognition, or CVPR, conference in Salt Lake City.
The virtual robots were trained in about 3,000 activity programs, wherein each task is broken down into sub-programs. The artificial agent can carry out 1,000 of the actions in eight different rooms, according to MIT News.
“Describing actions as computer programs has the advantage of providing clear and unambiguous descriptions of all the steps needed to complete a task. These programs can instruct a robot or a virtual character, and can also be used as a representation for complex tasks with simpler actions,” MIT PhD student and Lead Author Xavier Puig said.