Live data from Hacker News

Self-driving cars turned out to be harder than expected

vox.com

91–100 of 287 posts

Re: Self-driving cars turned out to be harder than expected

#91

I've been beating this drum for years (but I'm a nobody, so it's like pissing in the wind): 1. Anything less than 100% FULL automation is MORE dangerous than manual driving, because the "driver" will almost certainly lack any situational awareness. When the need for manual intervention happen, it will be at the moments where you need maximum awareness and split second reflexes. 2. There are SO many edge cases and nev…

It's too expensive to built the infrastructure to accommodate them--current infrastructure spending isn't even enough to eliminate potholes and other bad pavement conditions.

Re: Self-driving cars turned out to be harder than expected

#92
post #33

Alternate headline: Some people overstated how simple this problem was because they didn't know what they were talking about, could promote themselves as experts, could get funding for a self-driving startup or some combination of the above. I guess that isn't as pithy but it's closer to the truth. When it became clear that Uber's strategy (under Kalanick anyway) was premised on replacing drivers with AIs before the…

As usual the article conflates the (very difficult problem) of self-driving everywhere with various other sub-problems. The self-driving problem on highways is much more solvable, and can further be improved by convergent infrastructure evolution (forewarning of traffic jams from central control, embedded rfid to aid pathfinding in bad weather, weather reports, wildlife nets/barriers, highly standardized signage, int…

I suspect that part of it is that a lot of people in the tech and tech journalism space live in and around cities. Many of them are mostly focused on being driven around and not owning a car.

Long highway drives may be something they don't do that often. So the ability to own a car that can automate that subset of their driving probably doesn't strike them as all that interesting.

Personally, as I've said elsewhere, this seems like a much more tractable problem and gives me a lot of the total benefit of self-driving.

Re: Self-driving cars turned out to be harder than expected

#93

To anyone with a detailed understanding of the current limits of machine learning, this should be no surprise; unsupervised learning is far from solved and ML in its current state will always be plagued by the cat and mouse game of edge cases. The reality is the industry has decided to go this way anyways because it has a good profit outlook if you can get it to work, which is the only thing funding the endeavor. Con…

So I do not think that this is an unsupervised problem. As a matter of fact, this should be one problem where labeled data is not an issue.

I.e. just let humans ride the self-driving car, and record the human actions as well as all the sensors (at time t-1).

Combined that with all the humans on road X, and you have the a model for road X.

The problem here is that each road needs its own model as well as model per time of day.

Another problem is that each company will not share its supervised data.

Re: Self-driving cars turned out to be harder than expected

#94
post #33

Alternate headline: Some people overstated how simple this problem was because they didn't know what they were talking about, could promote themselves as experts, could get funding for a self-driving startup or some combination of the above. I guess that isn't as pithy but it's closer to the truth. When it became clear that Uber's strategy (under Kalanick anyway) was premised on replacing drivers with AIs before the…

So you're saying we need to program our driverless cars to occasionally go on road-rage rampages if they're cut off too many times?

No, that is reductio ad absurdum. He is simply saying that things which are too predictable are easily gamed, and this changes others' behavior in a game theoretical sense. Irrationality can have utility: see the "Ultimatum game."

Re: Self-driving cars turned out to be harder than expected

#95
post #82

I think car companies try to rely way too much on machine learning. You get some promising results fast, but it is all inside a black box, both verifying the correctness and changing what is wrong are almost impossible jobs. Maybe machine learning can be used to tell the difference between a dog and a plastic bag, but you'll need some hard code to describe how to react to either.

>ut you'll need some hard code to describe how to react to either My understanding is that's largely how it's done. The ML part is mostly about recognizing objects. But the car doesn't "learn" how to drive. It's told how to drive depending on what's happening in its field of view. Which is why there's probably misperceptions about the importance of miles on the road. It uncovers un-programmed situations but it's not…

There are some levels in between, like feeding an ML algorithm recordings of human drivers and expecting it to learn how to drive from watching those.

Re: Self-driving cars turned out to be harder than expected

#96
post #3

When I was starting my engineering career, it was right around when the first DARPA challenges had started. The hype was beginning, and my optimism towards technology was strong. I thought the predictions and timelines would be correct, and I still feel strongly that self-driving will be safer than humans in the long term. Recently, I bought a newer Subaru, with EyeSight. It has adaptive cruise and lane keep assist.…

I tried out a relative's Subaru on a 6h drive over the holidays. I really liked the adaptive cruise control for following behind folks who were not keeping a consistent speed. I just set it to a reasonable value at the maximum following distance and stopped worrying about my speedometer. However, at one point the guy in front of me turned off onto a small side road. It was at night, and I don't think the car realized…

Yeah this is interesting. My Tesla M3 does similar behavior, and so I often am ready to punch the accelerator. In the Tesla, this is how you solve that problem. The driver's push on the accelerator contraindicates the AI's decision to slow down, and so the car follows the driver's direction.

Where it gets dicey is the scenario where the "imminent collision" (hazards on, seatbelts tightened) detection is triggered, and the driver continues to push hard on the accelerator. Tesla has a fairly lengthy statement in the manual about this scenario. The bottom line is there are all kinds of heuristics at play that may or may not result in an override depending on the specific sequence of events.

Re: Self-driving cars turned out to be harder than expected

#97

I've been watching Lex Fridman's youtube podcast and there is a recent interview with Jim Keller [1]. Keller is a chip designer famous for his involvement in multiple chips at Intel, AMD, Apple and he was co-author of the x86-64 instruction set. He also worked for Tesla. There is a point in the conversation where Lex and Jim clearly disagree about how "easy" self-driving AI should be. Lex is clearly pessimistic and J…

Jim Keller went back to Intel in 2018.

Re: Self-driving cars turned out to be harder than expected

#98
post #66

Earlier quoted context omitted.

> It's a wonder only 35,000 people are killed on US roads every year. vs > do an exceedingly bad job of it. No, they're doing an amazingly good job of it. Given the cumulative number of miles driven and the conditions driven in it should impress you rather than that you come to the conclusion they are doing a very bad job. If the current crop of AI would be unleashed in those same conditions the carnage would be unbe…

Respectfully, it sounds like you've only read about driving and have never been in or driven a car. People are _terrible_ drivers. Cars are safer than they've ever been. The number of deaths is not the right metric to evaluate driver quality.

Respectfully, with about 30 years of driving experience and some years (fortunately long behind me) with crazy mileage, it sounds like you're wrong, and that experience includes Canada (including winter), a very large portion of the United States, Europe (both East and West) and Latin America.

People are great drivers if you take into consideration the conditions that they drive in, and that traffic deaths include all vehicle types, not just cars.

Yes, cars are safer than they've ever been. But there is also much more traffic than there ever was (both because there are more people, people are more affluent and some countries have been more or less designed around vehicle ownership), and infrastructure has not always kept up, and - again, in some countries - the car as a status symbol translates into 'right of way for the best protected', which leads to unnecessary accidents.

The number of deaths is - indeed - not the right metric to evaluate driver quality, instead, the number of deaths per total number of miles driven is much better, and further breakdowns into occupants, pedestrians, motorcycles and so on should be made before drawing conclusions. And on that scale - again, taking into account the conditions that people drive in - they are doing very well indeed, some countries excepted.

My own best tricks to avoid getting into accidents: don't drive when there is ice / snow / excessive wind, never drive when you're tired, stay away from countries where driving is a bloodsport, keep your car in excellent shape and never ever drive impaired, and that impairment includes cell phone usage.

FWIW I'd be totally for a law that punishes cell phone usage while driving with immediate vehicle confiscation when spotted, as well as instant revocation of the driving license of the driver.

Re: Self-driving cars turned out to be harder than expected

#99
post #6

Earlier quoted context omitted.

> Cop directing traffic by hand at an intersection. That’s going to be really challenging. Good one!

The cop can carry a transmitting device which will communicate with cars telling them what to do.

Who will pay for that?
Post reply on HN