Book review: Robot, Take The Wheel

Discussion in 'General' started by bwilson4web, Oct 9, 2019.

To remove this ad click here.

  1. bwilson4web

    bwilson4web Well-Known Member Subscriber

    Robot, Take The Wheel by Jason Torchinsky (Jalopnik reporter) is critical of Level 2, driver aids like Tesla's Autopilot. Citing three fatal accidents, in theory Tesla's Autopilot can make driving safer but he sees a severe problem with those other drivers who use it as if it were full autonomy. Sort of like someone with a 500 hp pony car driving on public streets like a drag or NASCAR track. But for three fatal accidents, he missed Tesla quarterly reports on accidents per million miles for Q2 2019, Autopilot had, 3.2, active safety features, 2.19, and none, 1.41.

    Accidentally, Jason gives powerful description of why Level 2 driving is uncomfortable:

    “... when you get behind the wheel and pedals of a powerful car; you feel powerful yourself , because all those 700 or so insane horsepowers are directly controlled by your own body; … many people who love to drive and drive aggressively are the most uncomfortable being driven fast and aggressively because they are no longer in control, and it feels wrong somehow.” (pp. 32.)

    Far from being an aggressive driver, I've ridden with three aggressive drivers and it wasn't fun. One was a former Navy pilot who did 'formation driving', either maximum brake or maximum accelerator in bumper-to-bumper traffic. The second broke a 12 hour drive into two segments at plus 15 mph and a single meal and bathroom break. In contrast, I normally took 14 hours with three breaks. I married the third who delighted in talking her way out of speeding tickets and waking me up with high speed, tire and wind noise. Whether Level 2 Autopilot or another human, cognitive driving style determines the rider's comfort level.

    Now Jason suggests autonomous systems could:

    “... download the fastest lap ever recorded on that track, and “replay” that lap. … download entire, curated road trips. ...” (pp. 193-194)

    In effect, full autonomy could be a hoot combined with playback. Unskilled drivers could experience the joy of riding 'second seat' around Laguna Seca Raceway.

    Can I recommend this book? It has a nice history of cars and autonomous systems (pp. 1-92.) He correctly calls a car a prosthetic with an emotional appeal and writes with a dry wit. But reading a book is different than forum posting. Each book page is like a three paragraph forum post and there are 240 pages with no breaks or commentary but your own.

    Still, I appreciate Jason because he was the only reporter at a Prius “press day” who had the good sense to run race car tuning software on his phone to get metrics Toyota was not sharing. Would that every reporter did that!

    Bob Wilson
     
    Last edited: Oct 9, 2019
    Domenick likes this.
  2. To remove this ad click here.

  3. hobbit

    hobbit Well-Known Member

    After watching that vid you posted a while back with the two "experts"
    expressing high levels of doubt, and seeing/feeling what a brandy-
    new 2019 car's "lane keep" does when an off-ramp opens up on the
    right, I ain't havin' none of it. And I can tell exactly when these
    unthinking boobs are behind me with their "autofollow" cruise
    controls -- still riding way too close, and I'm constantly having to
    do *their* driving for them and get rid of them. It feels like the
    people designing this stuff never actually go out on the road.

    No, the only way I expect autonomous travel to be feasible is
    with a lot of infrastructure support *in* the traveled paths, like
    trains. Even they don't always get a "perfect" navigation
    environment, but at least they're not having to figure out
    where the lane is through a construction zone on a rainy night.

    _H*
     
  4. bwilson4web

    bwilson4web Well-Known Member Subscriber

    Although less-bad, that remains a risk today. So it keeps me on my toes as left turns at angled intersections and curved, dashed lines from the left often lead to a 'split the difference' steering towards the right corner ditch or right lane. I have not tested the Mississippi crossings on divided highways. Regardless, one must alert behind the wheel regardless of simple or augmented driving.
    Distance following was less-good with the previous Autopilot (current version is 2019.32.11.1, downloaded Sept 29.) The distance scales are 1-7, 1 closest, but between 3-7, the previous version did not seem to have significant changes in following distance. The current version has larger and more distinct following distances but I've not had time to properly test on the highway. But there is a 'best practices' challenge.

    With our previous Prius Prime, alone on the highway in cruise control, there were a small number of too often 'raised' pickup trucks that took great delight in tailgating and illuminating the cabin. So I adopted the practice of following semi-trailer, towed trailers, and RVs as camouflage for both optimum cruise speed, 65-70 mph, and tailgate avoidance.

    Although following "big truck" works with our Tesla, the EV efficiency means driving at optimum speed is less important than it was with our earlier Prius. So now I run faster, 73-77 mph, between charging stations every 90-140 miles. Only when driving for maximum usable range, ~220 mi., do I revert to my Prius truck following style with close monitoring of the reserve range. When I get to 10 mi. reserve, I start mapping L2/NEMA 14-50 charging as reaching 5 mi. limits the car.

    Bob Wilson
     
  5. bwilson4web

    bwilson4web Well-Known Member Subscriber

    Running some informal tests, there appears to be a factor of two between following at 1 or 7. I’ll get better metrics following truckers on level highway.

    Bob Wilson
     
  6. bwilson4web

    bwilson4web Well-Known Member Subscriber

    I took the Tesla quarterly accident reports and generated this chart:
    [​IMG]
    • Initial Model 3 production shows a decrease in safety rate (accidents per million miles.)
    • Since Q4 2018, the rate has improved reflecting driver experience and software improvements.
    • Severe winter of Q1 2019 may have limited accident rate.
    Bob Wilson
     
    Domenick likes this.
  7. To remove this ad click here.

  8. Pushmi-Pullyu

    Pushmi-Pullyu Well-Known Member

    A couple of questions, Bob:

    1. The Y-axis is millions of miles per accident, is this correct? So a higher number on the Y-axis is better -- that is, a higher number means a lower accident rate?

    2. I don't understand your comment: "Severe winter of Q1 2019 may have limited accident rate". That would merely mean a lower number of miles driven for the period, which should not be reflected in the metric of n accidents per million miles, and therefore shouldn't affect the chart at all. Am I missing something?

     
  9. bwilson4web

    bwilson4web Well-Known Member Subscriber

    #1 - Correct. More million miles per accident is better.

    #2 - I prefer to list any and all hypothesis and we had a severe winter in 2019 Q1. This reduced the number of cars and trucks sold which indirectly suggests fewer trips. However, "limited" was probably incorrect and "affected" would be batter. Bad weather in Dixie often results in more accidents and there are better studies than my ad hoc observations:
    https://ops.fhwa.dot.gov/weather/q1_roadimpact.htm
     
  10. bwilson4web

    bwilson4web Well-Known Member Subscriber

    Living in Huntsville AL, I use what is available which is Autopilot, V10 2019.32.11.1. Perfect, no, but perfectible. So everyday I'm doing ~90% of my driving on Autopilot:

    10/10/19 11:10 AM 3.06kWh (15.3 mi)
    10/10/19 12:30 AM 6.39kWh (31.95 mi)
    10/09/19 01:32 PM 9.3kWh (46.5 mi)
    10/08/19 09:21 PM 8.28kWh (41.4 mi)
    10/08/19 12:26 PM 2.7kWh (13.5 mi)
    10/08/19 09:56 AM 4.42kWh (22.1 mi)
    10/07/19 06:28 PM 3.97kWh (19.85 mi)
    10/07/19 12:35 PM 1.9kWh (9.5 mi)
    10/07/19 10:37 AM 3.51kWh (17.55 mi)
    10/06/19 05:14 PM 3.63kWh (18.15 mi)
    10/06/19 01:54 AM 6.96kWh (34.8 mi)
    10/05/19 12:28 PM 6.1kWh (30.5 mi)
    10/04/19 09:39 AM 7.06kWh (35.3 mi)
    10/03/19 07:14 PM 7.61kWh (38.05 mi)
    10/02/19 09:56 PM 27.7kWh (138.5 mi)

    Anyone else? Actually yes.

    Perhaps I'm an extreme case because I'm curious about Autopilot limits and have mapped many but I'm not alone. Based on surveys of Tesla owners, ~75% are to a greater or lessor extent also driving on Autopilot out of an estimated ~500,000 Teslas ... ~375,000 fellow Autopilot testers.

    Not casting shade on Waymo and the others who have their approach. Just Tesla has the largest fleet, covering the widest driving conditions with enthusiastic owner/drivers and uses the lessons learned from this 'reality testing' to improve Autopilot.

    Bob Wilson
     
  11. To be fair, I don't think you can compare Tesla Autopilot to Waymo self driving. My son has Autopilot on his M3, which is needed just so you can have ACC (adaptive cruise control). So yeah, would expect most Tesla owners would have Autopilot, just for that. Can't imagine buying any car these days without ACC. I have that on both my Kona and older ICE car.

    I think the Tesla Autopilot is a very misleading marketing term. They should separate Driver Assist features (like ACC) and reserve the term Autopilot for true self driving (no driver input required from point A to B). I know my son was a little surprised when he got his M3, and no ACC. He was not happy having to shell out more money for Autopilot just to get this feature.
     
    Walt R likes this.
  12. To remove this ad click here.

Share This Page