Researchers at the Autonomous University of Barcelona's (UAB) Computer Vision Center in Spain have developed Synthia, a simulator employing convolutional neural networks and deep learning to enhance how vehicle artificial intelligence (AI) systems manage environmental factors such as obstacles.
UAB professor Antonio Lopez says the project originally focused on detecting pedestrians based on commercial video games. "Now with the sensors we use, we can see what the content of each pixel in an image is," Lopez notes. "We also know how far these objects are from the camera, which is crucial information for vision systems."
Vehicle AIs are being trained on a massive image dataset to recognize various elements and differentiate between key objects despite poor visibility, for example; the software utilizes this labeled information to interpret input from the vehicle's cameras and formulate a response.
"We've modeled an autonomous car within Synthia so we can make tests and be sure the vehicle does execute the orders it's receiving," Lopez says.
He sees the "complex and uncontrollable" urban environment as the main challenge for self-driving cars, but still envisions a partial rollout of such vehicles within a decade.
Lopez's team plans to further augment Synthia to manage more data and different types of situations.
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA