Tesla’s Fatal Error: Defensive Driving | Hacker News Reader

Tesla’s Fatal Error: Defensive Driving

eetimes.com

"Defensive Driving" seemed like a vague name even though I use it in speech myself, so I looked it up:

https://en.wikipedia.org/wiki/Defensive_driving

> defines defensive driving skills as "driving to save lives, time, and money, in spite of the conditions around you and the actions of others."

Looking at a few sources it seems to describe anticipating other vehicles, as well as the environment, and adapting to them.

I'm sure most drivers would say they do that, just like most drivers think they're better than average.

From my personal experience riding a fixed-gear bicycle around London definitely did improve my driving. There's a nice amount of fear in not being able to stop on a dime when a pedestrian randomly walks out in front of you or a taxi swings a left at the last moment and the only way to avoid it is see it before it happens.

That really is the essence of TFA... anticipation and seeing it before it happens is better than the fastest processor reacting and seeing it as soon as it happens.

Anticipation provides the advantage of avoiding the situation in the first place... typically by shedding speed before the event occurs.

There's nothing inherent in computer vision vs LIDAR that precludes computer vision from building a scenario and anticipating though. So TFA feels a little flawed in damning Tesla (and I am not a Tesla fan boi).

>Do regular folks think an aircraft autopilot does everything?

Yes, I think you will find they do.

Hop into a Cessna RG158, and no. "Autopilot" in that case is roughly comparable to "cruise control" in a car.

Pilot a modern commercial Airbus, and almost yes[0]. Those things can literally take off, fly, and land without human intervention if all goes well, though each step is currently unconnected. They're not often called on to do that, but they can.

0. https://www.businessinsider.com/airbus-completes-autonomous-...

I don’t understand why so many take issue with the feature being called Tesla Autopilot. I’m a pilot and believe it is a pretty accurate name.

Do regular folks think an aircraft autopilot does everything? Traditionally, it is just a system that holds heading and altitude. Some are even simpler and can only hold a heading. There are more recent advanced implementations that have a sequence of courses and altitudes to cover a flight route and departure/approach procedures along the way —- but it is still a fairly unintelligent system.

Most aircraft autopilots are ignorant of traffic, weather, flight conditions, air law/regulations, etc. I say most because aviation technology continues to evolve and there’s some neat stuff emerging.

But the majority autopilots in aircraft in service today look very similar to Tesla Autopilot if you go down the features they cover.

There's a whole lot less to run into at 40,000 feet.

Firstly yes most people don’t understand how an autopilot work, not to mention that there are varying degree of autopilot for aircraft ranging from what is essentially cruise control to those capable of take off and landing.

Secondly unlike pilots drivers aren’t trained on how to drive with an auto pilot and they don’t require to be certified for every car model and software version / package combination like pilots do.

So I as a passenger don’t need to assume or think about what capabilities the autopilot in an A380 has that an A350 might not I can be assured that the pilots that fly these aircraft were certified to fly them and understand exactly the limits of their aircraft and how to utilize autopilot effectively and safely.

Contrast it with Tesla’s advertised self driving capabilities and it’s a completely different picture.

Not only that I as a driver do not know under which circumstances the autopilot should or shouldn’t be operated on and exactly what level of attention is required from me but also as another road agent I can’t make the correct assumptions about other autonomous agents because I don’t know their capabilities or even if they are in autonomous mode or not. The fact that there for example isn’t an indication light when a Tesla or any other “autonowannabe” vehicle is driving under no or partial driver control is still mind boggling.

I agree with all your points, my comment was mostly my confusion around many articles getting upset with the name "Tesla Autopilot" which seems like an honest description of the feature.

There are practical implications of having an autopilot in a driving context as you've mentioned and those are perfectly valid criticisms that Tesla needs to address. Also, it is perfectly valid to criticize the less honest name the feature goes by -- "Full Self Driving"

It's sold as FSD, Full Self Driving, but the reality is more like a aircraft autopilot. 3 year owner here and I can tell you it's deeply flawed and if aircraft mistook the top of cloud layer as the ground, it'd be a good analogy for the kind of mistakes it makes.

Full Self Driving is a much less honest name that people should be legitimately upset about.

But I see article after article pick on specifically the "Tesla Autopilot" name which I always found strange.

Autopilot in video games and movies does everything. More people exist that have experience with autopilot in video games and movies than has experience with autopilot in actual aircraft.

It's more an outcomes and operator requirement perspective: When it gets used it gets a plane safely from A to B, without crashing, and pilots are not required to have split-second reactions to it failing. It has severe limitations in what it can handle, but the usage scenario is designed to make that acceptable. (That's not to say they are foolproof or never involved in accidents, but the failure scenarios are differently)

This can't be said of Tesla Autopilots, since they are not intended to be only used in environments without physical obstacles and other traffic being kept at safe distances by ATC and warning systems. (they would be a lot safer if those environment restrictions existed, but most people would find them pretty useless then) Instead they rely on "... but you need to be always alert and react very fast to prevent a crash", which is a very different requirement on the operator. One that is fundamentally difficult, and evidently ignored by operators.

As a non-native speaker, it took a while for me to notice that (a) the article is ironic and (b) he's using "offensive" as in "offending others", not the opposite of "defensive" ("attacking" others).

> he's using "offensive" as in "offending others", not the opposite of "defensive" ("attacking" others).

I guess that largely depends on whether you consider a vehicle a weapon :)

I interpret the point of the article as being that normal defensive drivers will learn to handle autonomous vehicles common mistakes and anticipate them too, saving everyone from accidents.

This is since "autonomous" vehicles are plainly just as dumb, but in maybe other ways, as the most reckless of drivers out there. Going beyond this state to handle all kinds of "special cases" (that are still quite common) is really hard.

The article is not about any specific brand (Tesla or not), or implementation details like teraFlops, LiDAR or not, etc. so those discussions seem off-topic.

He has a point.

I've thought about this for awhile.

In order to be a safe driver you need to break the law sometimes.

You program it into a vehicle, and their is an accident, the lawyers will have a happy day.

I don't feel like a debate. I want a autonomous driving vehicle. I want my chariot to drive me home after the hootenanny.

Reply on news.ycombinator.com