Nvidia Teaches Autonomous Cars to Drive by Mimicking Humans
Technology company Nvidia has taken a unique path to developing autonomous driving technology by teaching driverless cars to pattern their driving after humans.
Through a project called DAVE2, which built off of the findings of DARPA Autonomous Vehicle (DAVE) research, Nvidia used a variety of technology to teach a Lincoln sedan, codenamed BB-8, how to mimic human drivers on the road.
Using their Nvidia DevBox and the Torch 7 (a machine learning library) for training, and the Nvidia Drive PX self-driving car computer to process the information, a team of Nvidia engineers trained a convolutional neural network how to navigate the roads with help from a time-stamped video from a front-facing camera in the car that is synchronized with the steering wheel angle applied by the human driver. While there was a human in the car, video footage of the drive showed the car steering itself while the person held their hands outside the window.
The real-time driving data that was collected to teach the car came from the streets of New Jersey, and the car was driven in California. A variety of information was collected in different conditions, including clear, cloudy, foggy, snowy and rainy conditions, at both day and night. Information collected also took into account things like lane markings, residential streets with parked cars and unpaved pathways.
While there is no plan currently in place to develop this technology for the roads, Nvidia has partnered with different automotive companies including Audi, Tesla, Volvo, and Mercedes-Benz to develop autonomous cars.
