University of California, Berkeley researchers demonstrated that a few self-driving cars can prevent traffic jams, and in some cases double the average speed of surrounding vehicles.
The team used a video game-style interface to control simulated cars on multiple virtual roadways; in each scenario, the researchers tested four algorithms that used reinforcement learning to learn how to navigate the streets. However, the algorithms can still be improved, which is why the team is making its programs public. If others think they have a more advanced algorithm, they can use this framework to test new ideas, the researchers say.
It is unknown when self-driving cars will become commonplace in real life, but some of the new techniques being developed could be applied to modern vehicles. For example, patterns of traffic-reducing acceleration and breaking could be used by adaptive cruise control systems in cars, saving time, gas, and possibly lives.
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA