About > Let's Go Humans

Teaching the Car to Drive

Artificial Intelligence and Connectivity

The first building blocks that we outlined for the self-driving car focused predominantly on the car identifying its surroundings and positioning on the road, but it still needs to actually interpret that data and information into action. The next step to making the car fully autonomous is to teach the car how to use that data to drive itself. According to Chris Gerdes, a mechanical engineering professor at Stanford University, cars will soon have the skill-level of the best human drivers, maybe even better.

Through artificial intelligence (AI) and connectivity, the car uses the data it collects to present an accurate picture of its environment, determine actions, and communicate those actions in an understandable way.

Building Machine Intelligence

The effort behind the self-driving revolution is truly an attempt to make the roads and all passengers safer. In order for us to ensure this safety, we need to depend not only on the cars’ mechanics, but also on its intelligence. The thought of relying on what a machine “knows” is still a bit unnerving for many consumers, but as driver error is the cause of 94 percent of accidents in the United States, AI presents a key solution to learning from and preventing those errors.

At CES 2017, NVIDIA Co-founder and CEO Jen-Hsun Huang suggested in his keynote address that the emergence of new technology is creating the opportunity for the car to be your most personal robot. “With deep learning we can now perceive the world, not just sense the world and also reason about where the car is; where everything else is around the car,” Huang said.

AI allows a car to create its own knowledge base from algorithms, which will determine the car’s future actions. As Huang elaborated “we can use technology to teach a car how to drive just by watching us, observing us and learning from us.”

Even the system of algorithms has improved over the short period of time that AI has been implemented. Instead of programing the vehicle on manually-coded algorithms that cannot change dynamically, deep neural networks (DNNs) allow for fluid responsiveness and reaction from the vehicle itself.

Connectivity and Cloud Computing

Connectivity, much like AI, relates to the car’s understanding and communication of its surroundings. Currently, we use connectivity for the purpose of telematics or infotainment – through either the installed technology or our smartphone’s connection to the vehicle. But according to research from CTA and Vision Systems Intelligence, connectivity will be essential for total autonomy in the following use cases:
  • Localization and path planning – by transmitting the data from high-definition maps, connectivity will enable the car to localize based on those geographic attributes.
  • Network training – by crowd-sourcing data from various cars and companies and collecting it on a network server, the algorithms that determine the cars’ actions will learn from and update based on previously-made mistakes.
  • Safety – by facilitating communication with infrastructure, such as traffic signals, connectivity will help prevent collisions.
The complexity of this technology has broadened the landscape of the traditional automotive industry. NVIDIA, as described above, and a host of other companies have made efforts to apply their technology to the self-driving car. In terms of connectivity, Microsoft has entered partnerships that pair its software with the data being collected and compiled from sensors. Car makers have then agreed to implement this software so the vehicles can access the massive data source, and interpret it with an impressive level of sophistication. 

Cooperation for Consumer Access

It is not enough to develop simply the tools and mechanics that help the car drive on its own. In order for a car to become fully self-driving, we have to “teach” the car to interpret those tools in a way that humans can…only better.

According to Chris Gerdes, the Stanford University engineering professor mentioned earlier, the issue is not whether the technology will be created quickly enough, but whether it is accessible and affordable as a transportation option for everyone. If access is restricted in the consumer market, the societal benefits of self-driving cars are greatly decreased. It will take cooperation not only between the technologies, but, perhaps more importantly, between the companies, industries and establishments that control it.

The partnerships of companies in and out of the traditional automotive industry provide encouragement that all hands are on deck to make this future a reality. All of these elements and improvements are starting to come together to create the framework for a fully self-driving car. Just look at the “AI car supercomputer,” Xavier that NVIDIA introduced at CES 2017 to get a glimpse of a future where cars can “perceive, mobilize, reason and drive.”