Understanding the Myths and Realities of Autonomous Vehicles

The topic of autonomous vehicles, also known as self-driving cars, has been at the forefront of technology news for the better part of this decade. I get the appeal, as the vision of vehicles that drive, or fly around on their own, has historically been the stuff of science fiction. Autonomous vehicles are fascinating as they use almost every kind of technology imaginable, including network connectivity, messaging, cloud services, graphics processing units (GPUs), artificial intelligence (AI), video analytics, and more.

Over the past year, I've noticed the self-driving topic has become more widespread. Last month at Microsoft Ignite, we heard how the autonomous vehicle is one of the core components of Goodyear's strategy, as outlined in a previous No Jitter post.

And earlier this month at Nvidia's GPU Technology Conference (GTC), Porsche and Audi had concept cars on display, and Volvo announced it had selected the Nvidia Drive AGX Xavier computer for its next generation of vehicles. Its announced goal is to have a Level 2 vehicle in production by 2020. Level 2 vehicles have two or more advanced driver assistance systems, which are able to control braking, steering, or acceleration of the vehicle when needed.

Drive AGX is a fully integrated AI car computer that enables automakers to streamline the deployment of self-driving capabilities while lowering the cost of development and support. The platform runs Drive software that incorporates features such as drive monitoring, data collection, obstacle, and computer vision that are critical to vehicle operation, while the Drive AGX kit includes a vehicle harness, power supplies, camera sensor, and everything else required to make deployment close to plug and play. Nvidia has been the industry leader in making GPUs easier to consume and the Drive AGX platform is a great example of that.

This news follows a June 2017 agreement between Volvo and Nvidia to jointly develop a Level 4 automated driving system (high automation; no driver intervention) that would use Drive in partnership with automation companies Veoneer and Zenuity.

We're likely no more than a couple of years away from seeing a wider range of self-driving cars available, as I discussed with Nvidia executives Rob Csongor, VP and GM of Automotive, and Danny Shapiro, senior director of Automotive Marketing, while at GTC. The conversation helped me formulate my thoughts, and made me realize how many misconceptions people have about automonous vehicles. Here I clarify a few of the misconceptions.

1. In-Vehicle AI Isn't Only for Self-Driving Cars

The first wave of cars with in-vehicle AI use the technology to provide better driver assistance. Having an AI watch the road doesn't mean it needs to take over operations. Instead, it can stop the car when it senses that an accident is going to occur. Or, it can constantly monitor the driver's eyes and issue an audible alert if they've been off the road for too long. AI can make cars easier, more comfortable, and safer to operate.

2. Self-Driving Technology Isn't Just for Cars

Autonomous operations are for any kind of moving vehicle. We're likely to see a wide range of other autonomous vehicles long before we see self-driving passenger cars. We should see self-driving capabilities in anything that moves, including trucks, trains, buses, drones, and aircraft -- and should expect some disruptive opportunities.

For example, last week at the Gitex 2018 event in Dubai, I saw a prototype of a self-driving grocery store on wheels from Robomart, which is part of Nvidia's AI incubator. Using an app, shoppers will hail the vehicle to their location, pull food out, and be automatically charged -- making shopping much more convenient than driving to a store. The idea behind the vehicle isn't to go zooming down the highway at 60 miles per hour to deliver groceries. Rather, the first use cases will see the vehicle move through neighborhoods at safe speeds.

Shopping redefined with Robomart

Other early adopter operations will include construction vehicles that are confined to local sites, airport shuttles, or mobile office buildings.

3. Autonomous Vehicles Are Already Here

While the technology seems futuristic and a bit scary, the fact is self-driving vehicles are already here and are safe. Many trains operate without a driver, relying on sensors to stop in case of track obstructions. The inter-terminal shuttle at San Francisco Airport has been operating without a person at the helm for years, and it safely moves tens of thousands of people daily. These low-speed, controlled environments are ideal use cases to test and refine the technology.

4. Perfection Isn't a Barrier to Entry

It's a common misconception that self-driving vehicles need to be perfect to hit the road. But while perfection may be required for airplanes, it needn't be for most vehicles.

I thought about this in my conversation with Nvidia, when Csongor asked me if I would feel safe in an airplane with a safety record of 99.9%. I said absolutely, as the odds of hitting that 0.1% is miniscule. Based on the number of planes that fly globally per day, that would equate to hundreds of plane crashes per day, making 100% the bar.

For self-driving cars, the barrier to entry only needs to be better than human error. People have been driving for about 100 years now, and in the U.S. alone there are thousands of car accidents every day in which many are killed. We text, read, drive drunk, fall asleep, or do other things that cause us to be distracted and crash. Once we can show the rate of accidents with self-driving vehicles falls below the current rate then, as a society, should want to embrace autonomous vehicles for their ability to drop accident and death rates.

5. Self-Driving Vehicles Don't Require a Painful Learning Curve

One of the societal fears on the journey to autonomous vehicles is that many accidents will occur as manufacturers test their cars on our already congested streets. During a Q&A with the analysts at GTC, Nvidia CEO Jensen Huang discussed this topic at length. Nvidia has built a virtual simulator that allows drivers to steer cars through real-world scenarios it generates. He demonstrated this capability during his keynote, showing a 100-mile route around Silicon Valley that virtual self-driving cars can operate in.

On the virtual road with Nvidia

Once a manufacturer has put a vehicle the virtual simulations, it can upload the data into the onboard computer and begin on-the road operations. In many cases, the virtual road system is better than the physical one. For example, if the automaker wants to test the impact of something like a sunset on the sensors, the simulator can hold the sun in place and for hundreds of miles. Also, the virtual vehicles can drive nonstop for any length of time, making data collection easier.

While most people today wouldn't jump into a car with no driver, steering wheel, or other human controls for transport across town at high speed, they would be willing to have an autonomous delivery vehicle drop off a package or hop on a driverless bus to get them between the airline terminal and rental car building safely. I believe we should embrace instead of fearing autonomous vehicles; they will change our lives in ways we never imagined.