SoylentNews is people
SoylentNews
Lofty promises for autonomous cars unfulfilled
The first driverless cars were supposed to be deployed on the roads of American cities in 2019, but just a few days before the end of the year, the lofty promises of car manufacturers and Silicon Valley remain far from becoming reality.
Recent accidents, such as those involving Tesla cars equipped with Autopilot, a driver assistance software, have shown that "the technology is not ready," said Dan Albert, critic and author of the book "Are We There Yet?" on the history of the American automobile.
He questioned the optimistic sales pitch that autonomous cars would help reduce road deaths—40,000 every year in the United States, mostly due to human error—because these vehicles themselves have caused deaths.
As a result, self-driving maneuvers in the technology-laden vehicles are limited to parking, braking, starting or driving in a parking lot.
[...] "Automation may be used in areas such as closed campuses, where speeds are low and there is little or no interaction with other vehicles, pedestrians or cyclists or inclement weather," said Sam Abuelsamid, engineer and expert at Navigant Research.
The big problem is "perception": the software's ability to process data sent by the motion sensors to detect other vehicles, pedestrians, animals, cyclists or other objects, and then predict their likely actions and adapt accordingly, he said.
And that part is key, said Avideh Zakhor, engineering and computer science professor at the University of California-Berkeley.
"The perception part is not solved yet. The most advanced publicly available is 80-85 percent (reliable). That means that 15 percent of the time, it's going to hit objects and kill and destroy them," she said.
(Score: 1, Interesting) by Anonymous Coward on Monday December 16 2019, @04:45AM
Maybe it's just the big players that are over-promising. Smaller startup operating in small areas, simple areas, office areas and real estate developments, but still public places is promising driverless cars in 2020:
https://arstechnica.com/cars/2019/10/how-self-driving-shuttles-could-enable-car-free-living-in-the-suburbs/ [arstechnica.com]
I think May Mobility may do it about the same time.
Deep Learning isn't solving all the world's problems like promises suggested, and it's not only cars where the promises are falling through. What has the newest, compute-intensive artificial intelligence really accomplished? It seems like just more (albeit _better_) classification. Granted, in the self-driving car arena, it can help classify stop-lights and cars and pedestrians, but it won't realistically develop the rules, obedience to the law, braking behaviors, or avoid throwing passengers from one side of a vehicle to the next on a turn.
These are systematic principles created by people, not by AI that solves all the world's problems. They're developed by iteration, testing, further iteration, learning from example, and so on. I'd suppose that Google, Apple, Cruise, Toyota and others have thrown enough deep-learning at this problem that if it hasn't been solved already it simply won't be solved by AI. (Really, what is there to develop if the AI gives you what-to-do at the next step? why isn't this done already?) Like everything, like the development of the LCD, like the development of computer processors, none of AI nor anything else will allow anyone to skip from step 1 to step 967.
It will all be incremental. Solve a small problem, then solve a slightly bigger one.
Even IBM's Deep Blue was meant to replace doctors (Watson). Diagnose cancer. Find new drugs. Largely, it hasn't. Even back in the 1990's, software was supposed to make treatment plans. Better your care. Allow everyone to live longer, healthier. Maybe, in fact, it's the press who is over-representing everything, or else heavily leaning on the most extreme offering of any spokesperson out there.
Lets assume: How long, with computers helping in design, development, and manufacturing, did it take to get CPU processors from 5um to 7nm? (How many process generations were skipped/advanced because of Deep Learning?) Then, how long might it take driverless cars to get from the light pedestrian areas to car-filled urban downtown? How long do _you_ think it will take for self driving cars to permeate society?