Humans Still Quicker Than Robots at Learning to Drive

Self-driving cars may be threatening to take the jobs of millions of human drivers, but there are still some areas where mankind has the upper hand.

An artificial intelligence algorithm designed to teach an autonomous car to drive managed to do so using just 72 hours of driving data. While that may be impressive for a robot, it is still twice the amount of time it takes the average human to learn to drive.

Computer giant Nvidia developed a convolutional neural network (CNN) to learn how to steer a car in any weather condition, using only the data taken from cameras and a car's steering wheel.

Nvidia's project, called Dave, first began in 2015 in an effort to bypass the need to program specific features—such as lane markings and guardrails—into self-driving software. By collecting data from roads in New Jersey, Nvidia was able to train the CNN to steer a car the same way a human does in the same conditions.

"We have empirically demonstrated that CNNs are able to learn the entire task of lane and road following without manual decomposition into road or lane-marking detection, semantic abstraction, path planning, and control," a blogpost describing the technology stated.

"A small amount of training data from less than 100 hours of driving was sufficient to train the car to operate in diverse conditions, on highways, local and residential roads in sunny, cloudy and rainy conditions."

The latest development for driverless cars follows three years of sustained investment in artificial intelligence and deep learning from Nvidia.

"Three years ago, we dedicated ourselves on the single greatest endeavor in the history of our company," Jen-Hsun Huang, CEO of Nvidia, said at the GPU Technology Conference earlier this year.

"We decided to be all in on AI. I think we are going to realize looking back that one of the biggest things that ever happened is AI."