Home / Business / Driverless cars may be decades away, no matter what Elon Musk says.

Driverless cars may be decades away, no matter what Elon Musk says.



In 2016, Lyft CEO John Zimmer predicted they would “end” all car ownership by 2025.

In 2021, some experts aren’t sure when people will be able to buy a self-driving self-driving car out of the area.

In contrast to investors and CEOs Academics studying artificial intelligence, systems engineering and autonomous technology have long said that building a fully self-driving car can take years or even decades. Now, some of them are continuing. It said that despite the investment of up to 80 billion dollars. But we may never get a self-driving car as we promised. At least there are no major advances in AI that almost no one predicts will arrive anytime soon—or completely redesign our cities.

Even the most fanatic of this technology ̵

1; in 2019, Musk doubled his previous predictions. And saying that the autonomous Tesla robot will be launched by 2020, it has also begun acknowledging the public that experts who say so may have a point.

“A critical part of real-world AI needs to be addressed to create completely self-driving tasks without supervision,” Musk recently tweeted. Translation: To make cars drive like humans. Researchers need to create AI in equal measure. Researchers and academics in the field will tell you that’s something we don’t know how to do. Musk, on the other hand, seems to believe that’s what Tesla is going to accomplish. He continues to deliver the company’s next-generation “Full Self Driving” technology. which is a driver assistance system with a misleading name. which is currently in beta testing

A recently published article titled “Why AI is More Difficult Than We Think” sums up the situation nicely. In this regard, Melanie Mitchell, a computer scientist and professor of complexity at the Santa Fe Institute, noted that when the deadline for the The arrival of autonomous vehicles has slipped away. People in this industry are redefining the term. Because these vehicles require a test area with geographical constraints and ideal weather conditions. Not to mention security drivers or remote monitors. The manufacturers and sponsors of these vehicles therefore include all of these warnings in their definitions of independence.

Dr. Mitchell wrote with an asterisk: “None of the prophecies came true.”

In vehicles you can actually buy Autonomous driving shows nothing but improved cruise control like GM’s Super Cruise or optimistic Tesla Autopilot. in San Francisco A GM Cruise subsidiary is testing a self-driving car that doesn’t have a driver behind the wheel. But there are humans monitoring the performance of the car from the backseat. And it’s the only commercial robo-taxi service that operates in the US. without a driver at all Small operations are limited to the low-density portion of the Phoenix metro area. from Alphabet’s Waymo family of companies.

An autonomous vehicle powered by General Motors subsidiary Cruise during a test drive in San Francisco in 2019.


Photo:

Andrej Sokolow/dpa/Partner Pictures/Getty Images

Still, the Waymo vehicle was in a minor accident with the car wrapped up in the trunk. And their confusing (to humans) behavior has been cited as a possible cause. Recently, people have confuse them with traffic cones on a construction site.

Nathaniel Fairfield, software engineer and head of Waymo’s “behavior” team, said, “I don’t know if we’re bumped or winded more than the driver.” The company’s self-driving vehicles are programmed to be vigilant—“in contrast to the driver. Legislative teenagers,” he added.

Chris Urmson is the head of autonomous truck startup Aurora, which recently acquired Uber’s self-driving division (Uber is also investing $400 million in Aurora). Something useful in the next few years But to become widespread takes time,” he said.

Initially, Aurora’s vehicles will only operate on highways for which the company has created high-resolution 3D maps.


Photo:

Aurora

The key to Aurora’s debut was to work on highways where the company had only created high-resolution 3D maps, Mr. Urmson said. It was a long journey from the highway where it was initially launched, but Mr. Urmson declined to say when that would happen.

Slow launch of “automatic” cars that are limited and controlled by humans. is slowly is predictable and predictable even years ago. But some CEOs and engineers argue that new self-driving capabilities will come if these systems can record enough road distances. Now, some are in a position where all the test data in the world can’t. Compensate for fundamental AI flaws

Decades of advancements in artificial intelligence known as machine learning has yielded a specific pattern “Intellectual intelligence” only in its most primitive form. Ministry of Defense on AI

To measure today’s machine learning systems She has developed four levels of AI complexity. The simplest thinking starts with “bottom-up” skill-based reasoning. The current AI is pretty good at teaching itself to stay on the highway. The next step is to learn the rules and reasoning (e.g. what to do when a stop sign), then knowledge based reasoning. (Is it still a stop signal if half of the branches are covered with twigs?) And at the top is expert reasoning: the unique human skill of being left in a situation where Unconventional and brings out our knowledge, experience and skills in one piece.

The problem with driverless cars is actually on the third level. Today’s deep learning algorithms which is a variety of machine learning It cannot achieve a knowledge-based representation of the world. Dr Cummings said. and human engineers’ efforts to compensate for this shortcoming, such as creating ultra-detailed maps to fill gaps in sensor data. They aren’t often updated enough to navigate vehicles in every possible scenario, such as encountering an unspecified construction site.

Machine learning, which excels at pattern matching, is terrible at inference—transfer what they learn from one domain to another. pedestrian But it cannot be said to be an inanimate object that is highly unlikely to cross the road.

“When you were a child You will be taught that a hot stove is hot,” says Dr Cummings, but AI cannot transfer knowledge of one stove to another. she added “You have to teach that for every stove that exists.”

Some researchers at MIT are trying to fill this gap by going back to basics. They have put a lot of effort into understanding how babies learn in terms of engineering. to translate back to AI systems in the future

Dr Cummings said “Billions of dollars have been spent in the self-driving car industry. And they won’t get what they think they’ll get.” This doesn’t mean that eventually we won’t get that car. It’s “self-powered” in one form or another, she said. It just “will not be what everyone promises.”

But she added that A small, low-speed shuttle that operates in well-planned areas. Loaded with sensors such as lidars, engineers could reduce the level of uncertainty to a level acceptable to regulators and the general public. (Pictures of shuttle buses to and from the airport, e.g. driving along the expressway, etc.)

Nathaniel Fairfield, a software engineer and head of behavior at Waymo, said his team didn’t see the fundamental technological barriers to making self-driving robotaxi services as widespread as his company.


Photo:

Caitlin O’Hara/REUTERS

Waymo’s Mr. Fairfield said his team didn’t see any fundamental technological barriers. To make self-driving robotaxi services like his company widespread. “If you are too conservative and ignore the facts You said it would take 30 years—but it wasn’t,” he added.

More and more experts Suggesting that the path to complete independence was not based primarily on AI, engineers have solved countless other complex problems, including landing spacecraft on Mars. by dividing the problem into smaller pieces So smart humans can create systems to manage each part, says Raj Rajkumar, a professor of engineering at Carnegie Mellon University with a long history of self-driving cars. Optimistic about this path “It won’t happen overnight. But I saw a light at the end of the tunnel,” he said.

This is the main strategy Waymo has taken to achieve autonomous shuttles on the road. and hence “We don’t think you need a full-fledged AI to solve driving problems,” Fairfield said.

Mr. Urmson from Aurora said his company combines AI with other technologies. To create a system that can apply common rules to new situations. as a human being wants

share your thoughts

When do you think we’ll see fully autonomous cars? Join the discussion below.

Getting to self-driving cars the old-fashioned way too Tried and true “system engineering” will still mean spending big bucks on equipping our roads with channels and sensors to guide and fix robotic cars, said Dr Mitchell. And they will still be restricted to certain areas. and certain weather conditions—where the machine teleoperators Humans are on standby if something goes wrong, she added.

Our self-driving version of Disney’s futuristic animatronics is far from building artificial intelligence that can be dropped into any vehicle, immediately replacing a human driver. It could mean safer human-powered cars. and fully self-driving cars in a number of carefully monitored areas. But it’s not the end of car ownership—not anytime soon.

—— For more WSJ Technology analysis, reviews, advice and headlines, sign up for our weekly newsletter.

write to Christopher Mims at christopher.mims@wsj.com

Copyright ©2020 Dow Jones & Company, Inc. All rights reserved. 87990cbe856818d5eddac44c7b1cdeb8




Source link