Nvidia Video of Self-Driving Car in Rain, Country Lanes

Burney Simpson

A new video from Nvidia researchers offers an extended view of an autonomous car driving on public roads in New Jersey, managing rain, an unmarked road, and other challenges.

Nearly all of the 14-minute video presents the point of view of a car observing the autonomous vehicle. There are a few minutes showing the autonomous car’s view as it travels a rolling country road.

The researchers used an Nvidia DevBox and Torch 7 for training and an Nvidia Drive PX self-driving car using a Torch 7. The system operates at 30 frames per second.

The video appears to have been shot around Holmdel, N.J., where Nvidia opened an office in February in what had been a Bell Labs site.

It’s not clear if it is legal to operate a self-driving vehicle on a public road in New Jersey. The state legislature has been considering proposals on the technology since 2012 but has yet to pass anything.

Much of the video shows the autonomous vehicle in a business park, and in what appears to be a blocked off parking lot.

A link to the video was included as part of a paper “End to End Learning for Self-Driving Cars” from 13 Nvidia researchers based in Holmdel.

The paper’s abstract reports the researchers “trained a convolutional Neural Network (CNN) to map raw pixels from a single-front-facing camera directly to steering commands.”

This system soon learned to drive on local roads “with and without lane markings,” on highways, and on unpaved roads, the researchers report.

The researchers says their system automatically learns how to detect road features with minimal human intervention.

They conclude that their system will bring better performance and smaller systems because “the internal components self-optimize to maximize overall system performance, instead of optimizing human-selected intermediate criteria, e.g., lane detection. Such criteria understandably are selected for ease of human interpretation which doesn’t automatically guarantee maximum system performance.

“Smaller networks are possible because the system learns to solve the problem with the minimal number of processing steps.”