Hands Off the Wheel at 65mph – And Alive to Tell About It

Look, no hands. An AImotive-equipped autonomous vehicle drives itself 65 mph on California’s crowded US101 south of San Francisco.

Autonomous vehicles (AV) are safe, much safer than human driven vehicles, we are repeatedly told. But when the time comes to sit in one, especially one about to go off at highway speed, it can be a bit scary.

AImotive invited ENGINEERING.com to experience one of its self-driving cars go full highway speed.

I shouldn't have been worried. The 20 minutes on US 101 south of San Francisco was handled with aplomb by El Capitan, as the Prius has been renamed by its owners, AImotive. The car was traditionally driven—by a human—onto a center lane of the highway. It was around 1pm,so while not snarled in the Bay Area’s notorious rush-hour traffic, it was still fairly busy. A chime indicated we had gone into self-driving mode. The car maintained a safe driving distance from vehicles in front of it, kept itself centered in the lane, braked smoothly when traffic slowed and sped smoothly back up afterward. 

At one point the car wavered ever so slightly. That was as exciting as it got. I wouldn't have noticed except László Kishonti, AImotive founder and CEO, sitting beside me in the back seat, perked up. El Capitan had a moment of confusion, less than a second, from the shadows cast by an overpass, he later explained. The car quickly found its bearings and settled down, completing the rest of the journey without the slightest event.

Pretty boring, I said, as we parked the car at the AImotive offices.

That’s the point, Kishonti responded.

AImotive licenses it AI-based technology to auto companies. Peugeot (PSA) and Volvo are customers. Porsche may become a customer, but I can’t get them to divulge.

AImotive does not make the car itself, nor most of the hardware added to the car to make it run. Instead, it has assembled the parts and adds the AI to make its small fleet of Priuses into highway-ready AVs. A key component is NVIDIA, whose graphics processing units (GPU) are the brains of the AV.

aiDrive is an autonomous driving software suite. aiSim aids in the development of aiDrive. It is a software simulator that provides “photorealistic” simulations of driving situations with a game-like interface. It is capable of simulating different systems but AImotive only offers it to customers and prospective customers of the aiDrive system. AImotive licenses aiDrive to auto manufacturers.

Kishonti sees cost as his company’s big advantage. He said by the time you add the cost of the car, electronics and processors, atypical AV costs close to $300,00—AImotive can outfit a car for less than $15,000.

AImotive got its start in Budapest and keeps most of its 170 staff in Hungary. It has added some Silicon Valley funding and hopes to make its AV technology available on a bigger scale to more automakers.

The AI-based technology learns from a traffic scene that is filmed and annotated. That data prepares the AI for different scenarios. It is collected from around the world in different locations. AI trained on a dataset collected from a single location would act similarly in different situations in different environments. It learns how people drive, even avoiding local obstructions. It is the only system that can do so, says Kishonti. Other systems rely on AI that is trained but acts similarly despite a myriad of situations found in various regions of the world.

I remember a recent trip to India, which has big cities with some of the most chaotic traffic in the world. First-time visitors are often traumatized, certain of death or causing it, on the taxi-ride to and from the airport to the hotel. It’s not just the crush of vehicles and pedestrians that make the Bay Area traffic pale by comparison, it’s frequently a large animal.

“How’s it with cows?” I asked, sure that will throw Kishonti for a loop.

“We have cows,” he said, unruffled. “Kangaroos, too.”

László Kishonti, AImotive founder and CEO, with El Capitan, a well-instrumented self-driving Prius that took us for a ride on the highway.
Brains take the back seat. AImotive has adapted NVIDIA GPUs for its system but is working on its own hardware, aiWare, for the future. Such hardware provides the processing needed for the car to react autonomously. The goal is to shrink the processing unit to the size of a shoebox. Heat dissipation becomes an issue with temperatures reaching100 degrees Centigrade. The heat degrades the performance of the GPUs, according to László Kishonti, AImotive founder and CEO. Not what you want when the car needs to “think” quickly to avoid accidents.