Much of the attention in the virtual reality (VR) industry is focused on entertainment; creating a realistic world that humans can become immersed in for entertainment. Researchers at OpenAI see another use for the work that has gone into making virtual worlds as real as possible – training artificial intelligence.
Training AI robots is not a simple task. The algorithms that control movement in a robot need to be trained on thousands of real-world examples. OpenAI’s systems work using two distinct algorithms; one which interprets where everything is around it and another which tries to decipher why and how an action occurs. In the case of, for example, picking up and stacking a tower of wooden blocks could take hundreds of repetitions with the algorithms observing and interpreting how the action is performed before it would be able to replicate the motions.
By using a VR simulations of the activity, however, the AI was able to decode the actions needed after just a single recording, and then replicate them using a physical machine. The system was able to copy the movements made by a human without ever having moved the machine before. The developers expect that the VR training system will work well for training with other rigid objects, since VR is able to accurately simulate solid objects well, but is less successful at accurately modelling fluid or flexible objects, though that may change as technology develops.
“Nothing in our technique is specific to blocks,” says Josh Tobin, a researcher at OpenAI, in a video. “This system is an early prototype, that will form the backbone of the general-purpose robotics systems we’re developing here at OpenAI.”
You can view a video demonstration of the OpenAI system below.
VRFocus will continue to report on OpenAI and other innovative uses for VR technology.