Virtual Home to Train AI


A team of researchers has come up with a training ground for Artificial Intelligent characters to train them to do regular household chores.

The project is led by researchers at the Massachusetts Institute of Technology’s Computer Science and Artificial Intelligence Laboratory (CSAIL), the University of Toronto, McGill University, and the University of Ljubljana in Slovenia.

All about virtual home

VirtualHome is a virtual world where chores never end. It might seem like a nightmare for human but for AI, it can serve as a training ground. It is a Sims-like system which will help AI to perform everyday activities, one step at a time.

Unlike humans, AI does not have the talent for inference. If we tell a human to vacuum the rug, he or she will already know that first, they would have to grab the vacuum pump, move it, plug it in and then start working with it. But it is not so obvious for a robot. For them, one has to break down the task into the individual steps-like walk to the closet, take out pump etc. Hence, in case of a machine, one task is needed to break down in sub-tasks.

This VirtualHome goal is to help robots to learn tasks by first experiencing in a virtual system and then use that knowledge for real world. In its present form, the VirtualHome avatar can perform 1,000 separate actions in eight different settings, from the living room to kitchen and home office. They have created programs that laid out each step of a specific task in skeletal detail. For example, the task of watching TV involves subtasks like walk to the TV, Turn on the TV, walk to the sofa, sit on the sofa and watch the TV.

The researchers gave these programs instruction to characters in the VirtualHome system that had the character who performs these tasks and the videos of the agent performing the activities would be used to further train robots, giving them a visual example of what these actions look like.