Slow Teslas everywhere: 24 hours into ‘the FSD button’ | Hacker News Reader

Slow Teslas everywhere: 24 hours into ‘the FSD button’

electrek.co

I don't get it. The article clearly outlines the factors accounted in the safety score, and vehicle speed is not one of them (at least not directly). Yet the author and apparently others are intentionally driving slowly to boost their score? That doesn't make sense.

You can definitely "game" the score (to an extent) by making unnecessary trips, i.e. driving to give it more data, but driving abnormally slowly on those trips doesn't help anything. That is, unless you think driving 55 mph will help in the measures, but I'm skeptical of that.

The forward collision measure is based on number of miles, so if you drive 100 miles with no warnings, that's scored the same way if you were driving at 55 mph or 80 mph.

For the others (hard braking, aggressive turning, and unsafe following), each computes the percentage of time you spend doing the unsafe thing relative to doing it safely. If you want to improve those measures, you need to actually spend more time braking, or turning, or following at >50 mph and less than 3 seconds distance. Just driving a slow constant 55 mph on the highway won't change those measures at all, as I understand it.

The forced disengagement measure is just whether or not the car put you into autopilot jail for not paying attention while autopilot was active. I'd hope no one is getting put in that jail, but especially not if they're trying to be on their best behavior.

Anyway, from the Tesla FAQ, I find the 'predicted collisions in 1 million miles' formula fascinating:

    Predicted Collision Frequency (PCF) = 0.682854
      x 1.014495^{Forward Collision Warning per 1,000 Miles}
      x 1.127294^{Hard Braking}
      x 1.019630^{Aggressive Turning}
      x 1.001444^{Unsafe Following Time}
      x 1.317958^{Forced Autopilot Disengagement}

Hard braking and forced autopilot disengagement are, in their model, big predictors of a collision. I'm surprised to see unsafe following at such a low weight.

"vehicle speed is not one of them (at least not directly). Yet the author and apparently others are intentionally driving slowly to boost their score? That doesn't make sense."

Makes LOT of sense, higher speed or driving at speed of rest of the traffic will require hard braking(>6m/h in 1sec) and turning.

Most insurance companies also seem to be offering these dongles which rate you based on your driving. What doesn't make sense is person can be "aggressive driver" without considering type and condition of their car. A performance car(BMW, Porsche etc.,) will be able to slow down or turn faster while being more safer than any CUV/SUV.

I don’t see how higher speed equates to hard braking and turning. I regularly drive 75mph on the I-5 in Washington (state) and never have to hard brake or turn. It’s just a matter of not following too closely—I aim for about 2s following distance myself, a number that was impressed upon me as a teen in Driver’s Ed.

If I drive slowly the likelihood of spotting a forward collision early is higher and I have more time to react, and certainly the speed differential as I approach some stopped traffic is lower reducing my braking intensity. If I'm driving slower I can make the same radius turn with lower Gs improving that rating, and I'm also going to be much more likely to be following safely since at the lower speed I can be closer whilst still having a safe headway.

So basically every metric is easy to improve by simply driving slower.

That's assuming you're varying only speed and not e.g. following distance. I try to leave at least 3 seconds' distance pretty much always (often more), so I have a pretty consistent amount of time to react regardless if I'm driving 30 mph or 55 mph.

Additionally, most of the metrics are cut-offs. It doesn't matter if you brake very gradually or if you brake very close to the area where it penalizes you. Based on my driving so far, the hard braking threshold is pretty forgiving. Yesterday, I let off the accelerator completely and went full-regen braking and it didn't care.

Anyway, I've driven 99 miles (highway/city mix) with the Safety Score thing turned on, and I'm still at score 100. I haven't slowed down at all.

I don't know what criteria they use for unsafe follow time but I remember a government suggestion that was pretty impractical for rush hour motorway type driving. There were just too many cars packed into the road space to spread them out that much so even good drivers end up close to the vehicle in front.

Yikes. sounds like they’re model, in this case, for what is “safe driving” is too simplistic. driving like granny and pissing everyone one else around you off, most likely, makes for a less safe environment. road-rage is a thing.

but this leads to another thought.. if Tesla has this data stack and, and lets say your a current non Tesla user, road-raging pass a granny Tesla driver. Tesla probably could pick up your plates and other forms of identification and label you as not so safe. Some data sharing later, Your insurance rates go up. or rather when when you decide to buy a Tesla, you already have a bad score.

Their model doesn’t include driving speed, so what’s simplistic here is Electrek’s interpretation of it.

My take-away is that Tesla is selecting for drivers with easy roads.

My take-away is that I’m no longer interested in owning a Tesla.

What is a “hard road” and how does it lead to higher rates of hard braking (-0.3g) and turning (lateral 0.4g) and <1s following distance at speeds over 50mph?

Hard roads and highways are ones where lots of cars need to change lanes in not much space. For example sections where entrances are quickly followed by exits always end up with more hard braking.

I find this fascinating, because Tesla is meant to be able to completely autonomously drive a vehicle, so it knows how to drive safely, so what you would expect is just running the self-driving programme and getting it to judge how safe the real driver's actions are - ie, Tesla saw hazard X at time T, driver responded at time T+delta etc.

That's what I would expect. Yet instead of this they've got something that could've been hacked together in 15 minutes by an intern.

This must be what they will be doing once they are farther along with the Dojo supercomputer. 100%

If instead of buying $10k FSD 5 years ago you bought TSLA stock, you would have $213,330 today.

And if you bought BTC, $340,000

It wasn't $10k 5 years ago. It was like 4k or something.

And hopefully that overhyped bubble will soon burst..

What's the point of "Full Self Drive" if you can't eat and drink or write something on your laptop while driving

Whats the point when your car can't drive you home when you are drunk as fuck?

In this case it is also necessary that "in control of an automobile" is not defined to include sitting behind the wheel ready to take over if something goes wrong. In a world where that phrase encompasses sleeping it off in your car, that will be a difficult change, I feel.

That, too. It really needs to be hands free.

I suppose this will need 50 further years of development.

Edit: Or other streets...

That is the eventual goal, and this is the beta where this is not yet possible.

Marketing.

A brief explanation of FSD can be found here:

https://www.smbc-comics.com/comic/fsd

I guess it's just a matter of time until those safety scores are sold to insurance companies...

"breaking". WTF is braking

Since you may be serious (no downvote from me), braking is applying the brake in a car. A brake is a device for slowing a vehicle, usually via friction (though regenerative braking in EVs/hybrids is different, as it involves using the electric motors as generators instead).

Reply on news.ycombinator.com