Hacker News Reader

Comment by zellyn | Hacker News Reader

I don't understand why everyone keeps pretending there's a smooth continuum: "He says they contain GM’s road map to autonomy, starting with “Driver in charge” in 2010, progressing to “Driver mostly in charge” this year, to “Car mostly in charge” in 2020, and finally “Car in charge” in 2025."

That just doesn't make sense. To my mind, there's "cruise control" as it exists today, and "car drives itself completely and safely even in adverse conditions". Pretending that it's safe to suddenly hand over to the human in an emergency is just crazy.

View on HN

> To my mind, there's "cruise control" as it exists today, and "car drives itself completely and safely even in adverse conditions". Pretending that it's safe to suddenly hand over to the human in an emergency is just crazy.

Fine but where do you draw the line? Is it still safe if "cruise control" adds speed matching, steering, lane changing, or braking? As far as safety today, when I'm on the highway, I'd rather have a Telsa on autopilot than a human driving behind me.

Pretending that it's safe or not is an issue for 2025 cars technology as there are no cars on the road that "drive itself completely and safely even in adverse conditions". Not even Google's self-driving cars are there yet. They still don't drive on in inclement weather and have a professional driver monitoring them at all times.

> Is it still safe if "cruise control" adds speed matching, steering, lane changing, or braking?

That ship sailed years ago. Even in the 2015 model year (mid-2014 release) all of those features are available.

My car does speed matching, minor steering to keep you in the lane (at highway speeds), and auto-braking (both for auto cruise control and for collision avoidance). It works fine.

and on your car, if it does any of those things wrong, and you don't correct it (because you are paying less attention, and reacting quick enough would be an impossibility), who is to blame?

Legally me. But I suppose I could sue the manufacturer.

We're rapidly heading toward this very "interesting" time where cars will have (indeed, some high-end cars are very close today) assistive systems, especially for highway driving, that really don't require drivers to pay much attention. It is absolutely guaranteed that people will start reading books and watching movies when they're nominally driving even though they're legally responsible.

The manufacturers will fall back on the "these are only assistive systems." It says so in bold type in the user manual.

This is what scares me most about self-driving cars. The "last mile" of making them self-driving is by far the hardest part (Google's engineers have stated that they're at least 10 years out, and that is admittedly an optimistic estimate), so I think this is an era we'll be stuck in for quite some time.

There's a really good video of a John Leonard talk at CMU on the topic: https://www.youtube.com/watch?v=x5CZmlaMNCs

He basically took video of his commute in the Boston area over the course of a few weeks and pulled out snippets of situations that would be really difficult for an AI. IMO, a lot of people are far too eager to extrapolate from works-mostly under a controlled set of conditions to letting AI cars loose on the road in a wide variety of real-world conditions.

Even if you accept that we'll never have perfection (and humans certainly aren't), you do have to get to a point where serious failures are fairly rare events as they are with driving today in the context of the millions of road miles driven.

I have tried the Tesla autopilot and completely agree with you. The tesla cruise control is very good at controlling the speed and the brakes, but once you give it control of the wheel it gets very dangerous.

The car is prone to making a sharp turn towards the curb and then beeping and handing control back over to you.

Honestly, I think it's because the auto makers are completely out of their depth when it comes to machine learning, and are either pretending they're not while they get their act together, or pretending they're not while the ship slowly sinks…

Why wouldn't a GM, VW, Toyota work as they always have and work with partners with experience in that space? Google gets the eyeballs because it's the darling, but there are lots of companies working on lots of advanced and dull things under the radar. There is no reason why these behemoth multinationals can't use third party tech just as google, apple and amazon do for many of their products.

It's not as if all these new entrants are taking Ford's erstwhile philosophy of building everything inhouse. Rather tbey cobble lots of things from different sources together.

Well with technology available as it is today I think Google is going to find building the car is more difficult than creating the technology.

See its not just GM looking to do this, almost every make of components for cars is into this space so a solution is already in the pipe. however when it comes to delivering cars that people want; not what government agencies might be coerced into buying; google doesn't stand a chance.

They could license the tech. Car manufacturing isn't really an exciting business to be in.

I fear the day that google has broken the political barriers and car manufacturers release different software that turns into headlines of automated car crashes. I sincerely hope that google's software is licensed at a reasonable rate such that every car manufacturer will use it.

I've never trusted Google's technology with my life. (Never had occasion to.) I've trusted every car manufacturer's technology with my life at one point or another. I'm not betting on Google having a better safety record. They can learn, but dealing with physically dangerous situations is a new area for them.

I think politically that the driver assist systems that are in millions of cars already are at least as big a contribution as the research projects.

I agree.

Supply chain management, government, warranty, unions (UAW), this is not an easy industry to operate within.

Alphabet's market cap is 490B, Fiat Chrysler is 19B. Toyota is (afaict) 150B. GM is roughly 50B.

I don't think Google's gotta worry about building cars. They can just buy someone who does.

Ugh Google just buy Tesla already /s

They said somebody who can build cars. I assume that means on a large scale. Much as I like Tesla, they are not currently up to the task and their market cap makes then undesirable for anyone looking to buy a company that can make cars.

hence the sarcasm tag

Where does Tesla's autopilot fit "in your mind?" I doubt if Telsa is done. More hardware and software will be in future cars, but we probably won't have your final vision for several years.

Detroit is a little more willing to build killing people into its business model.

Hence the 30,000+ death toll on the roads for the past century, their big push in the 1990s to replace lightweight cars with big heavy SUVs, and their present push to build big automated distraction screens into the middle of cars.

Are you saying the US car industry is more willing than European, Japanese, Chinese and Indian auto manufacturers in producing cars which are marginally more dangerous, on the whole?

I think you are overstating your case. They all provide as much as their market demand. So, in some markets you see fewer safety features (see south African auto market)

SUVs probably do a better job of protecting the driver and passengers in a collision, and are no more efficient at killing non-motorists than "light weight". You can easily be mowed down by a Smart Fortwo.

Reply on news.ycombinator.com