Hacker News Reader

Comment by nsxwolf | Hacker News Reader

Having trouble accepting Elon Musk’s claim that Teslas will be capable of fully autonomous driving in 3 years.

One of the difficulties I am having is that I’m not aware of another advanced technology that just popped into the consumer market overnight like we’ve been told to expect with self driving cars.

There’s always a gradual evolution of a technology over time. There’s very little of that evolution seen with this technology. There’s self-parallel parking cars, there’s autopilot-style cruise control systems, but we’re going to instantly leap from that to fully autonomous driving?

It's like aliens visiting and giving us some far-future tech overnight.

View on HN

These haven't just sprung into existence. They have been worked on consistently for at least two decades. Carnegie Mellon has research dating back to the mid 1980s on this topic.

Self driving cars have been driven up and down various parts of the SF bay region for several years now. Google's self driving cars have driven over 1 million miless. They report that their vehicles have been involved in 12 accidents, none of which are the fault of the vehicle's self driving capabilities.

The only reason they will appear to 'pop' into the consumer market overnight is because this technology is politically charged and simultaneously so important to society that it is being overengineered to improve success and safety. The moment self driving cars has its first accident after being released to consumers, the media will likely flip, ignoring mountains of data demonstrating how much more safe these vehicles are to provide a sensationalized report on the dangers of self driving cars. The safer they are and the more data backing this up, the less likely this will derail the technology.

1 million miles: https://plus.google.com/+SelfDrivingCar/posts/iMHEMH9crJb

12 accidents: http://www.latimes.com/business/la-fi-google-cars-20150603-s...

CMU ALV: https://www.cs.cmu.edu/afs/cs/project/alv/www/index.html

Agreed. I saw most of the technology pretty matured in Japanese automotive almost 10 years ago.

Serious question:

What do these cars (with no steering wheel) do, if involved in an accident?

How would the owner / police move the car to the side of the road?

With a tow truck.

Some places have laws requiring you to move the vehicle off the road if it steerable. There has to be some way to move the vehicle manually.

Exactly this. Minor collisions should move to the side of the road ASAP.

They have an "urgent stop" button, which passengers can use. Presumably that's unnecessary, though, as I'd expect the vehicle can detect an accident and will stop. Tow trucks could take over, at that point.

Emergency steering wheel app.

And who pays for the insurance? And who is blamed for the accident? I'd be surprised if Google is accepting these costs.

In accidents involving self driving, it could be easier to go to the root cause of accident, as these cars shall have lots of data for that incident and environment (like a black box).

wow 12 accidents per 1 mio miles.

2009, humans in US:

185 crashes per 100 million VMT

1.13 deaths per 100 million VMT

all weather, incl. intoxication, bad luck, etc.

current google car tech relies on brute forcing visual mapping, breaks down in snow.

self-driving cars are controlled by software, written by humans, riddled with bugs. see Nest on how awesome this works for physical stuff. the aircraft industry has the correct coding practices, NONE of which are being applied by the current research teams. multi-redundant systems in play, with even more training for the operator. competing with human drivers is really hard. a shit ton of signal processing, etc. going on. all this techy-BS is belittling the awesomeness of the human brain and central nervous system.

You're underestimating the awesomeness of the human brain to tackle big problems. I want to be drunk, watching movies, pleasuring myself in the backseat of an autonomous Uber during my two-hour commute in and out of the city. Open your mind.

Interesting stats, and gives a good estimate but it's off by at least an order of magnitude.

1. Autonomous google crashes have been entirely small rear end fender benders. It might be 10x or more safer when accounting for this. You must look at the speeds and types of crashes and account for this in the safety calculation.

2. For safety it's less relevant but for insurance purposes Google has yet to (technically/legally) cause a crash so already it's cheaper for your wallet than a human driver. Progress =)

As far as I was aware, in none of the 12 accidents Google's autonomous vehicles have been involved in has the autonomous system been in control. They've all been manually operated at the time of collision.

Do you have any other information?

how are they off? straight out of a car and driver report, google it.

but i guess even this will get downvoted, rather than PROVEN wrong. why even bother debating anything on HN anymore?

Oh I meant its off in that its not an apples to apples safety comparison. Not all crashes are created equal. As of yet atonomous cars have only had specific types of (safer) crashes.

This should be included in the safety comparison as looking at raw crash number is not enough.

Technically, crashes caused by Google's self driving car is not "12 per 1 million," it is "0 per 1 million." which is significantly better than "1.85 per million" for humans. Inversely, crashes caused by humans involving a self driving car is 12 per 1 million. Not a good record for humans. :)

It's also important to take note where the driving is. In densely populated regions, you are more likely to have minor accidents such as fender benders than on the open road. However, the open road is more likely to get you killed if you are involved in an accident due to speed.

All this "techy-BS" is not belittling humans by creating self driving cars. I strongly believe the opposite is true, humans are amazing because we can build artificial intelligent systems to handle these kinds of tasks for us and let us focus on bigger and better problems than steering a car. Your same argument falls flat with other technologies. Why have a computer add numbers when we have a wonderful brain that can do the same? I argue that it is because we build systems to augment and work around our deficiencies. One of those deficiencies is our driving skill. Over 30,000 in the US alone die due to auto accidents. Self driving cars has the potential to eliminate most of these deaths. Ideally, we'll get to a point where a death on the road makes national news because of its rarity.

> we’re going to instantly leap from that to fully autonomous driving

If you had lived and owned a car in Mountain View, you would have been used to driving alongside fully autonomous cars for years now, and the move wouldn't seem nearly as instant.

Yeah, they're everywhere around here, including the tiny pods that can only go 30 mph and don't seem to have a steering wheel.

I legitimately can't tell if this is a sarcastic comment... I don't live in Mountain View, but haven't seen any of these on my visits.

Not sarcastic :-). I see one or two almost every day. Their number seems to have been increasing rapidly in recent months, have you visited recently?

It's been about a month. I guess I just missed them.

As a side-note, I love that I can't tell if a comment is true or sarcastic-faux-science-fiction, and it ends up being true. We live in a really cool time.

It might just be where I walk vs where you walk - I usually see them on Shoreline and between Shoreline and Castro, sometimes on El Camino in that span. A few weeks ago I saw one of the slow pod cars holding up a bunch of traffic in he right lane on El Camino because its top speed is too low for the flow there. :-D

http://www.usatoday.com/story/tech/2015/06/25/googles-autono...

I found a YouTube video from someone who seems to have encountered one: https://youtu.be/FKETvrCZhzU

I'm surprised there aren't more videos like this, though.

I haven't seen any of the pod cars yet but I've seen a couple of the Lexus SUV's. I've even seen a few in San Francisco but they're not Google's. I seem to end up biking next to the same one every week or so in Dogpatch. It's a dark red Nissan Leaf without any labeling. I wonder what company owns it.

> There’s always a gradual evolution of a technology over time. There’s very little of that evolution seen with this technology. There’s self-parallel parking cars, there’s autopilot-style cruise control systems, but we’re going to instantly leap from that to fully autonomous driving?

In addition to cars that park themselves and cruise control that handles full stop/start, modern cars also keep you in the lane, prevent imminent collisions, monitor blindspots, detect pedestrians and more. If you can't see autonomous vehicles coming in the near future you aren't looking.

On the one hand, Musk has been known to exaggerate timelines. All of Tesla's cars have ended up being released behind schedule (the first Model X cars are being delivered tonight, multiple years late). I doubt we'll really have fully autonomous Teslas in three years.

On the other hand, there is one interesting, unique aspect to the approach that Tesla is taking: their cars all have internet access, and could potentially learn from each other. Within a year they're going to have partial autonomous driving on over 100k cars in the wild, which is going to allow Tesla to generate an enormous amount of training data very quickly.

I'm no expert, so I don't really know how valuable this kind of data would be for building new autonomous driving systems, but it seems like it could have the potential to dramatically accelerate their progress in this area.

Keep in mind Google's autonomous car team says they'll have a product commercially available in 4 years. So a Musk 3 years might well equal a Google 4 years.

Sometimes I wonder if his public/private financing model lends itself to a politician's style of promises.

Well Musk has not a good track record with announcements like that.

That said, not quite sure what else he could really do. Car production cycle are long and capital heavy. The buying cost is also very large, so to maintain a media presence you can't really iterate every year like with other tech widgets.

Other manufacturers make prototypes between model refreshes and go to various car shows, so far Musk is able to that by just talking to the press. Good for him, saves Tesla a bunch of money until the competition in electric car market really picks up !

Tesla doesn't make prototypes and go to car shows because they did away with the yearly update model. Instead they are constantly adding new features and improvements to the car instead of waiting for some big update. So you might get a car today that has something a car made yesterday doesn't have.

"Well Musk has not a good track record with announcements like that."

Yup, very much like a politician. You get the sense that he is playing their game. They'll give him a hand but he has to speak their language. Promises made to the public are more like optimistic goals in which potential complications aren't accounted for and instead reserved as explanations for revised promises.

I'm a Musk fan, don't get me wrong, but there is a level of dream fuelling manipulation behind his approach that feels political.

That whole [i]warning the public about dangerous AI thing[/i] actually felt like a big political money push to me. Designed to overwork the imaginations of the public and get government more interested in throwing money at the tech.

What did Musk actually say here? I have great trouble trusting TechCrunch's interpretation of them, and they don't quote him on that specific bit.

"We're probably only a month away from having autonomous driving, at least for highways and for relatively simple roads. My guess for when we'll have full autonomy is about three years, approximately three years. However, regulators will probably not allow full autonomy for maybe... at least one to three years after that. So, it depends on the particular market -- some markets the regulators will be more forward leaning than others. But in terms of when will it be technologically possible, three years."

https://youtu.be/ktyMmzZ6WeM?t=474

Thanks!

I wonder just what he means by "full autonomy" though. When he refers to "autonomous driving...for highways" that means that the car manages the steering and acceleration but the driver is alert and involved at all times, and ready to take over any time the car can't handle something. I wonder if he's referring to a similar concept that works in more situations, rather than a car that can drive with nobody in it.

Or he might just be optimistic and mean full automatic driving. Hard to tell.

I thought fully autonomous cars had already existed long enough to have racked up statistically meaningful (and impressive) track records?

They have, but there's a big list of caveats that go with that. Google's cars are very clearly extremely safe on dry highways in good visibility when humans are ready to take over if there is construction on the highway and so on and so forth. That's where they've got hundreds of thousands of miles.

The cars may well be substantially better than that right now, but they don't have a million miles of experience in more challenging conditions.

Provided the roads are also very well-mapped.

Google (as well as some auto manufacturers) have made impressive progress. I'm actually encouraged that full autonomy (as in I can legally read a book) on a subset of limited access highways in appropriate weather conditions is something we might see within 10 years on a mainstream luxury car.

I am sadly far more skeptical about the general case.

I've been thinking about this. What do humans do when they don't know exactly where they are or there are detours/other off map difficulties? They mindlessly follow the cars in front of them. Waze at at small scale! The autonomous cars just need to a) not be willing to crash and b) follow the car(s) in front of them when in doubt...

The problem can be separated into navigation and situational awareness. The autonomous car won't ever be lost, and I guess it would be easy enough to prompt the passenger for input if it got stuck deciding where to proceed.

Not just well mapped but also what do they do when there are temporary detours? I can't imagine a situation where an automated driving car copes well with a temporary human based stop/go situation (without signs - I'm talking really temporary here like a construction site with a vehicle backing up). Maybe I'm wrong.

My biggest gripe with hearing about fully autonomous cars is northern climates and their inability to handle bad weather.

To be fair, as a human I don't feel particularly adept at driving in bad weather, so an autonomous vehicle that simply slows down to where it "feels confident" and uses some kind of radar to avoid collisions won't be doing too badly.

In theory, as storage costs approach 0, computers get faster and and vision becomes more sophisticated, road conditions shouldn't matter too much.

These cars are going to be scanning constantly in nice weather, it shouldn't be a stretch to then figure out relative position with GPS and then use the distance to 'known' trees / signs / landmarks from previous trips to determine lane position.

I'm sure there are barren areas where this would be tough (I've driven I-80 through snowy Nebraska about a dozen times en route to ski vacations in Colorado) but "stay between the snow banks" should be a decent start for autonomous driving.

Some humans have an inability to handle driving in good weather, heavy traffic, at high speeds, to avoid drinking, to avoid driving when tired, let alone bad weather.

There's a mostly straight street on my very ordinary and safe commute home. When it's raining at night and the road is wet and reflecting street lights, it's actually very difficult to see the lane markings and make sure you're sticking to your lane. And in those conditions, it's not always safe to be constantly checking your blind spot for a car beside you. This is in an urban environment, no snow, no crests or crazy curves, no risk of fallen trees, and I find that a stressful drive despite having driven that road thousands of times.

I mean, even pilots have to take control sometimes. In great conditions autonomous cars will arrive quicker than many believe. However, most people (myself included) would NOT trust an autonomous car, essentially operating in beta, to navigate a new england winter. That functionality is going to come and can even be pushed OTA later.

I now wonder if you will just pay $199 for the "blizzard package" or something. I suspect Tesla wouldn't do something like this, but just like Mosaic, they get to shape a lot of the early thought behind this.

If a car company charged more to simply turn on a feature that ought to exist, like driving in a blizzard, then I would never buy a car from them. ever. I only have one life to lose and it's not going at the hands of some scummy company that intentionally holds back software it can install in an instant so it can nickle and dime me.

Upgrades generally not just(or possibly even) safety software, will have an assigned value of the developer hours assigned to creating and maintaining upgrades and will be built into the cost of cars. Unlike 4 wheel drive, which legitimately needs to be more expensive having much more hardware, software releases can leverage pre existing APIs and hardware interfaces.

If we assume that either all cars have the software priced in, a "good actor" would allow you to pay less for "blizzard navigation software" if you lived in S. California.

At this point, Tesla is so new that they will make it as attractive as possible and push all of their updates for free. I also have a high degree of trust, at least relative to other auto companies, in Tesla. One would imagine GM, or another less scrupulous company would try to do something like this if they were the front runners.

edit: Point taken though. I am also not keen on some of the other car and boat manufacturers producing identical engines, and then having you "pay" to unlock more horepower.

Agreed. It would be very impressive to see an autonomous car that could handle blowing snow: https://www.youtube.com/watch?v=VHWNQVVY-LI

Have any manufacturers even fixed cruise control for icy roads? That should be trivial, but no car I've ever driven has handled it.

Plus, batteries don't perform as well in freezing temperatures.

My biggest gripe about hearing about the invention of planes is the lack of runways.

My biggest gripe about hearing about the invention of the automobile is the lack of mechanics and gas stations.

My biggest gripe about hearing about the invention of the internet is the lack of home computer ownership.

etc. etc.

"the lack of runways"

As an historic note with little relevance to your point, this wasn't solved until the US entered the second world war and the government built a bunch of them for heavy bombers, which were later converted into civilian use.

That's why the flying boats, like the Boeing 314 Clipper, were the go-to plane for long distance travel pre-war. There's no need for a runway when you can land and take off on water.

"lack of mechanics"

This isn't really true. The early cars could be fixed by most any mechanic or blacksmith. Remember, farm equipment was already pretty mechanically complicated.

The lack of gas stations was much more critical.

The specific examples I used were not the point.

The point is that you must give new technology time to grow and become mainstream before you can expect them to be everything for everyone.

Lots of new tech will never be everything for everyone

Hence "As an historic note with little relevance to your point".

At best, treat it as comment about how hard it is to think what detractors to X might have said before X became wide-spread. I believe your example of 'lack of mechanics' is poorly chosen; akin to saying 'My biggest gripe with hearing about fully autonomous cars is that they will be so expensive that only the five richest kings of Europe would own them.'

As a real-world examples I think are better, 'My biggest gripe with hearing about telephony is there won't be enough operators to handle the number of calls.' 'My biggest gripe with hearing about computers is that they are too bulky.' 'My biggest gripe with television is I'm at work when my favorite show is on.'

Here's the Hyundai SDC test last year on a rainy day. It's fascinating to see where it succeeds and where it fails.

https://www.youtube.com/watch?v=xs9Gr9V2mOE

Reply on news.ycombinator.com