Fully driverless cars could be months away | Hacker News Reader

Fully driverless cars could be months away

arstechnica.com

32 pointstnash13 comments

I've always been a big proponent of self-driving cars, but only because I imagined them as fully autonomous. A fleet of self-driving cars that rely on being backed by, essentially, a call center, sounds like a nightmare.

"When Waymo tested in Phoenix earlier this year, drivers sometimes had to take over the wheel to prevent the cars from holding up traffic because it took too long for humans in the command center to answer the cars’ requests for help."

I guess the ability for a single person able to essentially 'drive' an entire fleet of cars is an improvement but it's not exactly the utopia I imagined.

Fully driverless cars are here today, for that matter. But if you want want tamperproof driverless cars that perform well under adverse conditions and are thoroughly tested, then you're going to have to wait a lot longer than "months".

I think this snippet is pretty telling: "Waymo chose the Phoenix area for its favorable weather, its wide, well-maintained streets, and the relative lack of pedestrians." (Emphasis mine.) Probably wise, but I'm sure they've already carefully calculated the risk/return vs pedestrian fatalities and are coming out ahead.

How could a driverless car ever truly be tamperproof, though? At some point we'll just have to accept it as a potential risk.

Good question. Tamperproof to me means they can't physically be spooked into doing things, as well as being secure from hacking. In theory, they just have to be as safe as human drivers, who manage to kill thousands of people each year. But as a practical matter, people will probably demand that they be much safer than humans. And it will take awhile to prove that out.

…if ‘fully’ means ‘remotely operated, when the going gets tough’, such as, I guess, in bad weather (rare in Phoenix), or when there is a lot of traffic in one place, times when demand for taxis is highest.

This is something you should never hear about a product which is life critical that it works correctly:

"Efrati reports that Waymo CEO John Krafcik faces pressure from his boss, Google co-founder and Alphabet CEO Larry Page, to transform Waymo's impressive self-driving technology into a shipping product."

Combined with releasing that product in the area with the least consumer and safety protections:

"Another important factor was the legal climate. Arizona has some of the nation's most permissive laws regarding self-driving vehicles."

So the TL;DR is that someone has been pressured to rush a product to release in an area with few safety regulations that could cause a lot of harm if it malfunctions.

On the flip side, the status quo is killing 40000 people per year in the US alone:

http://www.nsc.org/NewsDocuments/2017/12-month-estimates.pdf

Which invites the question: how long before human-driven cars are no longer allowed on the public roads?

I'm guessing year 2035.

So we should push out half-baked solutions?

Depends on how half-baked.

From a PR perspective it's a definite "no". Even a single self-driving fatality would lead to global news headlines and a lawyer feeding frenzy.

But from a protecting people perspective? If making it "live" quicker would speed development and adoption... then maybe.

E.g. the first 5 year increase fatalities net by 1.1x. After that tech is actually dialed in and fatalities drop to 0.5x. Holding off gives you 400K fatalities whereas pushing earlier adoption gives you 320K. That would be 80K lives saved. (Under the assumptions of this shoot-from-the-hip model).

> Even a single self-driving fatality

It's more specific than that. It will be the first self-driving crash that would be glaringly obvious to a human driver. Like the Tesla crash where the sensors mistook a white semi for clouds and merrily plowed though it.

If the tech is adopted too soon it might face a death knell when computer assisted driving proves to be much safer than fully autonomous. It's much easier to fill a human's blind spots than replace the driver entirely.

Another plausible hypothesis is that the tech is already better-than-humans today, but not orders-of-magnitude better. Killing 30 thousand people a year instead of 40 thousand is not politically viable. So we just keep it in the lab while 10 thousand extra die because of politics.

Reply on news.ycombinator.com