Major companies are testing autonomous car software within virtual reality (VR) simulations of cities, mainly to spot and correct operational flaws without imperiling actual passengers.
Scientists also are devising techniques to enable cars to learn new behavior from VR, accumulating skills faster than human engineers can program them.
Toyota Research Institute CEO Gill Pratt says Toyota is using images of simulated roadways to train neural networks, and they are sufficiently similar to the real world to reliably educate the systems that operate the cars.
Other firms are exploring reinforcement learning to gain skills within virtual environments via trial and error.
One issue concerns the fact that because these programs learn by analyzing more information than anyone ever could, it can be difficult to audit their behavior and understand why they arrive at specific decisions.
Nevertheless, machine learning is expected to be vital to the continued progress of autonomous vehicles in the coming years.
From The New York Times
View Full Article - May Require Free Registration
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA
No entries found