Self-driving cars get a reality check

TOKYO -- The annual ITS World Congress arrived here last week just as Typhoon Wipha did. The timing was apt: The typhoon's lashing rains weren't alone in throwing cold water on optimism over self-driving cars and interconnected vehicles.

At the annual gathering to discuss intelligent transportation systems, engineers and operators alike injected doses of realism about the hurdles ahead for those technologies.

It was an ironic about-face for the annual gathering. In years past, the congress was largely a platform for experts to tout the potential of intelligent transportation. But as self-driving cars have grabbed the public's fancy, those same experts have become the voices of caution.

The roadblocks are plenty, even as companies such as Nissan, Tesla and Google race to unveil plans to develop autonomous cars by 2020.

Panelists presenting at the congress -- which was forced to cancel technical sessions, workshops and outdoor demonstrations the morning the typhoon bore down on the Japanese capital -- outlined a plethora of issues that could push these high-tech rides further into the future:

• Technical challenges.
• Lack of industry standards.
• Unclear or nonexistent regulations.
• Implementation costs.
• Liability questions.

"I think it's longer than a lot of people think," John Capp, General Motors' director of global active safety, said of the path to autonomous driving.

Consider the technical challenges to managing the massive streams of data being broadcast among vehicles and between vehicles and the infrastructure over Wi-Fi-like networks.

Those systems, which could use cellphone networks or a wireless communications channel called Dedicated Short-Range Communication, are considered key to a rollout.

But when a car drives down the street communicating with other vehicles and the roadside sensors that warn of blind curves, stray pedestrians and hidden obstacles, the vehicles are throwing out 1 gigabyte of data each minute over the wireless networks, said Christoph Hagedorn, CEO of Continental Japan, the local unit of German supplier Continental AG.

"That is a huge amount of data," Hagedorn said, adding that the stream of auto data can be too heavy even for today's top-speed 4G LTE cellular networks.

"It's no longer a challenge of the automotive industry. It's actually becoming an IT challenge," he said.

A demonstration drive through an ITS network set up along Tokyo's waterfront district showed big limitations. The signal, which was supposed to alert the driver to oblivious pedestrians or motorcycles in the car's blind spot, failed to trigger warnings in the Mitsubishi Outlander test vehicle.

At other crowded intersections, the "Caution, Oncoming Vehicle" buzzer sounded -- but it was unclear about which vehicle the system was warning.

Meanwhile, if every passing car were equipped with vehicle-to-vehicle communications, the continual bombardment of warnings in a crowded city could be annoying.

"That is one issue we are still weighing," conceded Yuichi Takei, manager of the vehicle control group at Isuzu Motors Ltd., which demonstrated the technology in a 2-ton Isuzu Elf truck.

Part of the problem is there is no one-size-fits-all solution.

Autonomous cars, such as the kind proposed by Nissan and shown in a prototype Leaf electric vehicle, could get around network problems. Those vehicles would brim with radar sensors, lasers and cameras to scout out the surroundings and independently plot a course for the car.

But autonomous cars face two challenges:

1. Their sensors have a hard time seeing around corners, into blind spots or miles down the road.

2. Their computers need to account for millions of driving scenarios.

"The limit of the car's knowledge is the car in front of me," said Klaus Kompass, vice president of vehicle safety at BMW Group. "As soon as you want to know more about the car in front of the car in front of me, you're limited by these existing technologies."

Also, verifying the software in autonomous cars for safe use will take a long time.

"The biggest obstacle is the millions of different driving scenarios you have to face," Hagedorn said. "There are so many driving scenarios it will probably require years of validation."

Because of these hurdles, networked systems -- such as vehicle-to-vehicle or vehicle-to-infrastructure setups -- are seen as a crucial first step to self-driving cars.

Roadside sensors and radios would keep the car connected to the network and share information about construction ahead, traffic jams, oncoming cars, blind spots, emergency vehicles or hidden stop signs. Japan already uses a limited system, and South Korea and European countries have plans to introduce their own in the coming years.

But that technology requires a universal standard, not only for the transmission device and receivers, but for the brand of transmission used. Regulation is also a big unknown.

Operators of transportation systems are pushing back for other reasons. They dread the cost of outfitting every intersection with the radio devices and the additional expense of running and maintaining the network.

New York City alone has 13,000 intersections with traffic signals that would need to be equipped, says John Tipaldo, director of systems engineering at the New York City Department of Transportation.

"Do the math," he said. "It means I've got to find a lot of money, which I don't have. All areas have the same issue. We're not at the forefront of this, because of the economics."

Or worse, operators fear being sued for accidents if the system should malfunction or suddenly crash -- as sometimes happened during the demos at the congress.

"The second we become involved with that interaction, we're going to pick up the liability," Tipaldo said. "Let's make sure we fully understand what we're doing before we go do things that we can't maintain. Pull the horse back a little bit."