Comment by dlisboa

21 hours ago

ADAS + human driver seems enough to solve most of the shortcomings of both human and robot drivers.

Full self-driving doesn't strike me as necessary to prevent deaths.

There is a moral hazard with ADAS + human drivers:

> Once drivers move from the novelty phase to the post-novelty operational phase, behavioral adaptation may begin to occur and overreliance and over-trust in the automation features potentially develop, including a greater willingness to look away from the forward road (as shown in the VCC data set). [1]

I'm not sure if this is more dangerous than strictly human-operated vehicles, but I would wager that it is more dangerous than strictly autonomous vehicles.

1. https://aaafoundation.org/research/understanding-the-impact-...

  • I'm not sure why a semi autonomous system of equivalent driving capability would be more dangerous than strictly autonomous vehicles because of "over-trust" unless one assumes that the driver that overtrust the autonomous vehicles and looks away is also inclined to try to override the semiautonomous system whilst doing so. Humans overtrusting a system capable of imperfect Level 5 driving will certainly not be alert to correct all its errors, but on the other hand they should be capable of correcting more than the baseline (especially if the autonomous system is good enough at general driving under most conditions for the human driver to be restricted to "get me off this road train track or flood the system shouldn't have stopped in" and "follow instructions" type edge cases rather than misidentifying objects at high speed type edge cases.)

    If there are additional problems caused by human drivers, it's surely that they are inclined to intentionally override the system with dangerous behaviours which is a separate problem entirely, or that their driving ability has atrophied so much from not driving they are unable to handle the boundaries of Level 4...

  • There was research some years ago in Sweden that cars with 4WD get into accidents 20% - 30% more often than cars with 2WD. And cars with 4WD averaged 10% higher speed at the moment of the crash.

    The cause was found to be drivers with 2WD losing a little traction early in their drove and subsequently slowing down, while drivers with 4WD just kept driving like normal until they were suddenly off the road.

    Another statistic is that intoxicated or otherwise impaired drivers cause more than half of all fatalities. So making a solution with machine learning and cameras that could retrofit into any existing car and just deny starting the car for such drivers, and having insurance companies cut premiums for having such gadgets installed, would be much more effective than any form of self driving or similar.

    https://mb.cision.com/wpyfs/00/00/00/00/00/3E/5A/A2/wkr0011....

    • The gadgets would have to meet an impossibly low false-positive rate and they'd need to be mandatory and non-removable for this to be effective. Constantly impaired drivers won't voluntarily install the devices, and any non-impaired driver who is falsely prevented from using their car will get rid of the device and not care about the insurance discount.

      Also, they really need to be able to detect the other really bad cause of impairment: mobile phone usage.

      1 reply →

  • It's like those approval prompts in coding agents. At the start you are reading the commands intently but after being bombarded with them for days you just hit accept on them.

Automatic braking alone is a huge deal: if you're a bit late in braking, you get a warning. If you go past that, brakes auto apply. When configured correctly, I don't even notice it's there 99.9% of the time...and the times I do I'm glad it's there, because the beep pulls you out of a momentary distraction.

The only problem is not every car on the road has it yet.

  • >Automatic braking alone is a huge deal

    Except I had to turn it off for my Ford E-Transit, because it will freak out at nothing. Sometimes it screams at me for cars that aren't even in my lane that suddenly slowed. Sometimes it misjudges a ripple in the crappy asphalt here as an obstacle I'm going to hit. There is a high chance its going to brake for no reason at all and cause an accident.

    I left the warning system on, but I will never give it control over the brakes.

    • Any time I crossed a bridge near me that is kind of steep and has a structure over it, the automatic braking engaged HARD. Like hard enough that I nearly caused accidents until I figured out how to turn it off.

      Not great.

    • I wonder if there will be some extra liability if the driver kills someone and the logs show automatic braking would have engaged. Something similar happened near me where a lamborghini driver disabled traction control, spun out and killed a girl on the way to school. They had to add new laws increasing the liability if you disabled TC.

  • Anecdotally, the only times I've ever noticed these chimes and the auto-braking (even automatic rolling-up of windows), it was false positives, and it was mildly terrifying: for a brief and unexpected moment I was no longer in control of my vehicle. [Recent Honda Civic and Audi A3 models]

    • I've only had a couple weeks in rentals with automatic emergency braking, and was very nearly been rear-ended twice due to false positives. Anecdotal, and I'm sure they'll improve with time, but I would never allow the system I experienced anywhere near a car I owned. It was profoundly unsafe.

      5 replies →

  • > The only problem is not every car on the road has it yet.

    Yep. ADAS should be mandatory like seat belts and airbags. A regulatory body should be able to come up with a minimum standard for it.

    • Even airbags were phased in over decades. Front-seat airbags were first available in the early 70s and didn't become mandatory until 20 years later.

  • Automatic braking can also be hazardous! In a front-wheel drive car, understeer is a fact of life. When you're approaching a bend in the road with your foot in the accelerator, you're going to be pointed at the guardrail, since the car's going to rotate around the bend when lift your foot off the gas. Now imagine you're driving a car with AEB and it slams on the brakes while you were lightly accelerating towards the guardrail, right before you were going to lift your foot off the gas. What was a perfectly routine curve just turned into complete loss of control, and there's a very good chance your car's going straight into the guardrail!

    Even when it doesn't slam on the brakes, if you were concentrating on the road, those jarring beeps can just as easily break your focus on driving as they can break your focus on a distraction. I think it's better to avoid distraction by keeping cruise control off, and you can compensate for having human-level reaction time by following at a safe distance and staying in an appropriate gear or setting an appropriate level of regenerative braking to avoid mindlessly getting too close to the car in front of you.

    I think that too often, driver assist features are being added to cars to compensate for the other features (like adaptive cruise control) that encourage drivers to mentally disengage from the task of driving. There's also the fact (well, opinion) that driving a car without electronic assists and nannies is fun and engaging (especially on back roads with the windows down and the radio up). Riding in an autonomous vehicle is just as dull as riding on public transportation, all while being less efficient, less social, and worse for the environment.

    It would be interesting if we could somehow test how the reduction in crashes with the introduction of driver assists would compare with if we had just banned cruise control, introduced minimum visibility and minimum suspension stiffness standards, and mandated manual transmissions.

    • This is the perfect expression of chronic windshield brain. If you are driving anywhere near the limits of dynamic stability on a public road, then you are already driving dangerously. Also, computerized stability control works better than 99.99% of human drivers.

      4 replies →

An ADAS would be L4 if it was designed to be truly autonomous. If it's not designed to be truly autonomous, you have to somehow address the attention problem. Even a good attention monitoring system isn't sufficient, because people have the ability to appear superficially attentive by all external metrics, while not engaging with the driving task mentally.

The data to differentiate these levels is pretty limited for a variety of reasons, and the problem only worsens when you conflate AEB with ADAS as the IHS/IEEE do in this article rather than use an OEDR-based definition.

I recently implemented OpenPilot for my vehicle and used it on a 9 hour road trip.

Due to the nuances of my vehicle, the built-in hands-on-wheel nag was still active and required (small) driver wheel inputs every 10 seconds to prevent the beeps. Just keeping both hands on the wheel was enough to pacify the nag. But this forced me to keep attention on the vehicle, more than OpenPilot's face direction detector. Anecdotally I am certain this kept me more engaged than if I could have gone hands off.

The net effect was I had more capacity to be aware of what was going on around me on the road, AND I didn't feel nearly as fatigued at the end of the drive. Keeping the car between the lines consumes a little bit of cognitive energy constantly, which compounds over a long drive. Freeing up that load allowed me to pay more attention to what was happening on the rest of the road. And the mental burden was still half as much--I felt like I had driven maybe 4-5 hours instead of 9.

I have some ideas for how I might disable the built-in nag, but I'm pretty sure doing so would make my car less safe. The current configuration might already achieve the safety maximum for ADAS based on front-facing cameras only.

It increases situational awareness while reducing fatigue, and I think that anything further I try to do will adversely affect one or both of those effects.