The Information: Apple's autonomous driving project is a costly wreck

"Earlier this year, a test vehicle nearly hit a jogger who was crossing the street and had the right of way."

From Wayne Ma's "Inside Apple’s Eight-Year Struggle to Build a Self-Driving Car" ($) posted Monday to Information subscribers:

Last August, Apple sent several of its prototype self-driving cars on a roughly 40-mile trek through Montana. Aerial drones filmed the drive, from Bozeman to the ski resort town of Big Sky, so that Apple managers could produce a polished film, with picturesque mountains in the background, to show CEO Tim Cook how their costly and long-running autonomous car project, Titan, was making progress.

Inside Apple, executives hailed the demonstration as a success. The vehicles showed they could drive without relying on highly detailed, three-dimensional road maps, which most rival self-driving-car programs require...

The good vibes following the Bozeman demo didn’t last long. Apple’s test vehicles, which are modified Lexus SUVs, struggled to navigate streets near its Silicon Valley headquarters without the maps, smacking into curbs and sometimes having trouble staying in their lanes while crossing intersections, according to two people who worked on the program. And earlier this year, a test vehicle nearly hit a jogger who was crossing the street and had the right of way, one of these people said...

Getting rid of high-definition maps has caused numerous problems for the test vehicles, forcing Apple’s human backup drivers to take over many times due to safety concerns, according to data Apple is legally required to file with California regulators and a person with knowledge of the situation. And in the first quarter of the year, one of Apple’s test vehicles, driving around 15 miles per hour, almost hit a jogger who was crossing the street at an unmarked crosswalk. Apple’s self-driving software first identified the jogger as a stationary object before recategorizing it as a stationary person and then finally as a moving pedestrian, all of which took place in less than a second.

But rather than stop, the car only slightly adjusted its path. The backup human driver slammed the brakes at the last moment, and the car stopped within a few feet of the pedestrian, according to a person familiar with the incident. Apple later determined that the car would have almost certainly hit the jogger if the backup driver hadn’t acted, this person said.

Apple temporarily grounded its fleet to investigate what it internally called the “jogger incident,” this person said. It resumed driving within a few days after fixing the identification problem and adding the specific crosswalk to its maps database.

My take: Adding specific crosswalks one at a time is not -- repeat not -- the solution.

12 Comments

  1. Gregg Thurman said:
    Cook has said that AI is the mother of all projects, or something like that.

    What if project Titan isn’t about making a self driving car, but rather an AI platform that could actually think like a human.

    If I were to undertake such a project I’d want to test ii in the most complex activity engaged by humans – driving a car. Following a map wouldn’t do it.

    How would value a platform that could that?

    0
    July 11, 2022
    • Steven Philips said:
      I think every other semi-autonomous vehicle has had severe real world teething problems. Tesla still does. It’s expectable. Similarly it’s taken Siri a long time to develop on device AI that can compete with Google. True AI navigation must be difficult. I’m not sure this report is particularly meaningful. Maybe? Or not? It at least gives a bit more information about Apple’s directions and intentions.

      2
      July 11, 2022
  2. Alan Trerise said:
    Meanwhile … CarPlay seems like a much more promising thread to keep pulling on. Complex systems that work almost always evolve from simple systems that work.

    4
    July 11, 2022
    • Gregg Thurman said:
      Meanwhile … CarPlay seems like a much more promising thread to keep pulling on.

      If CarPlay is ultimately installed on 25% of new cars, why go to the capital expense and low margin of manufacturing an EV?

      I think 25% is a conservative number. Who, until Apple, thought of the dashboard as a competitive advantage? Who does UI better than Apple?

      8
      July 11, 2022
  3. Timothy Smith said:
    I hit 2 deer in one week last year on country roads (my first and second of my life), but neither resulted in damage to my car or serious injury to the deer, because I hit the brake when I saw them OFF the road, BEFORE they jumped into my path. Likewise, I barely avoided a big buck on the expressway a few years ago because I braked before he jumped into the road.

    I just drove my Mustang Mach E hands free on the expressway in western New York, and I can see becoming somewhat comfortable in sparse traffic, limited access driving. (I even experimentally watched an hour long video with my iPad mini on the steering wheel. I kept looking up, though, so not the best way to watch a movie.) Hands free, looking at the road, not watching a movie, I’m relaxed.

    But I can never imagine feeling comfortable in city driving, and I can never imagine feeling comfortable on an expressway unless I am also watching the road, because of the deer. But, if I have to watch the road, hands-free isn’t worth much.
    So I agree with Joe. First, let’s kill the deer.

    1
    July 11, 2022
  4. Rodney Avilla said:
    And earlier this year, a test vehicle nearly hit a jogger who was crossing the street and had the right of way, one of these people said…

    If a person is not in a cross walk and is crossing the street, they are jaywalking and do NOT have the right of way.

    0
    July 11, 2022
    • S Lawton said:
      Bottom line, right of way or not, the goal is not to hit the pedestrian. However this is why its a test car with a human driver who prevented tragedy. Data reviewed and program error corrected. That’s what tests are for.

      0
      July 11, 2022
      • Steven Philips said:
        That’s weird. I always thought pedestrians were worth 25 points! 🙂

        1
        July 12, 2022
  5. Jerry Doyle said:
    Wayne Ma is under the erroneous assumption that Apple’s short range project for 2025 delivery is an “autonomous vehicle.” Apple has a long range plan to build a fully self-driving car the same as Tesla and other EV manufacturers. For the short term targeting 2025, though, I suspect Apple is planning either a level two or three.

    0
    July 11, 2022

Leave a Reply