Through and DriveWorks DriveNet, NVIDIA Support Autonomous Vehicles

8:25 AM | Labels: ces2016, drivenet, driveworks, nvidia, self-driving car

NVIDIA and DriveWorks DriveNet only part of NVIDIA'S big plans to provide a complete platform for autonomous vehicles. (Image Source: NVIDIA)

NVIDIA believes that autonomous vehicles will revolutionize society. The technology could potentially reduce deaths and accidents, traffic and increase mobility for those who could not afford to operate the vehicle. To maintain momentum while autonomous vehicles have been available, NVIDIA announced a DriveWorks software and DriveNet which is deep neural network hosts NVIDIA.

DriveWorks was born with the spirit of other platforms such as NVIDIA GameWorks DesignWorks and both of which were presented to provide a tool for game developers and graphic professionals to enhance their work. Not much different, DriveWorks is a set of tools, libraries, and modules of software that aims to accelerate the development and testing of autonomous vehicles. DriveWorks has also become a solution for training autonomous vehicle which covers the entire process of perception, localization, planning, and visualization.

DriveWorks allows calibration of sensors, data retrieval regarding the environment surrounding the vehicle, synchronization, and recording. Lastly, DriveWorks also facilitates the processing a lot of data through diverse sentor complicated algorithms running on the processor Drives PX 2. A software module used in each stage of the process of autonomous vehicles ranging from object detection, classification and localization and segmentation to the planning line.

DriveWorks can handle input data from the entire configuration of the sensor so that the mobile companies have the freedom of choosing their own settings. (Image Source: NVIDIA)

Deep neural network

DriveNet NVIDIA will also become a big part of NVIDIA'S efforts encouraging the advancement of autonomous vehicles. NVIDIA CEO Jen-Hsun Huang, stressed that deep neural network has exceeded the human capacity in several areas. For example, a network of Deep Speech 2 belong to Baidu has outperformed humans in voice recognition test. Then there's Microsoft who collaborated with the Hong Kong University of Science and Technology student surpasses in tests of IQ.

Neural network does not require programming on the part of the developer. After being trained using large data, they can learn how to recognize and respond to the new situation. It is easy to see its usefulness on autonomous vehicles, since it is impossible to give the program a vehicle the right response against a large number of ever-changing road situations and difficult to predict.

DriveNet carrying 37 million virtual neurons and need 40 billion calculations to process information through the network. DriveNet is very precision making it can classify and recognize each pixel and specifies an object based on the pixels. Nevertheless, Drivenet designed as reference neural network and NVIDIA are hoping each vehicle manufacturer could create their own network.

DriveNet is the reference for future deep neural networks developed by the manufacturer of the vehicle. (Image Source: NVIDIA)

Other interesting possibilities that could be presented by the neural network is a collection of autonomous vehicles experience within the collective network. So, when one vehicle learn something new, then the entire vehicle others can learn from the experience so get smarter.

Ultimately, the goal is to create the NVIDIA platform from upstream to downstream for autonomous vehicles. As a prefix available network training NVIDIA DIGITS, deep learning network NVIDIA supercomputer, Drive 2 PX will execute what it has learned from the network.