• May 2016
    M T W T F S S
    « Mar   Jun »
     1
    2345678
    9101112131415
    16171819202122
    23242526272829
    3031  

Nvidia GPU-Powered Autonomous Car Teaches Itself to See and Steer

Network World (04/28/16) Steven Max Patterson

An Nvidia engineering team built an autonomous car that combines a camera, a Drive-PX embedded computer, and 72 hours of training data. The researchers trained a convolutional neural network (CNN) to map raw pixels from the camera directly to steering commands. Three cameras and two computers were utilized by the training system to obtain three-dimensional video images and steering angles from the vehicle driven by a human. Nvidia researchers watched for changes in the steering angle as the training signal mapped the human driving patterns into bitmap images recorded by the cameras, and learning was enabled using the CNN to generate the internal representations of the processing steps of driving. The open source machine-learning system Torch 7 was used to render the learning into the processing steps that autonomously saw the road, other vehicles, and obstacles to steer the test vehicles. The steering directions the CNN performed in a simulated response to the 10-frames-per-second images captured by the human-driven car were compared to the human steering angles, teaching the system to see and steer. On-road testing proved CNNs can learn the task of lane detection and road following without manually and explicitly deconstructing and classifying road or lane markings, semantic abstractions, path planning, and control.

MORE

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: