A Better Solution for Autonomous Cars | Hacker News Reader

A Better Solution for Autonomous Cars

medium.com

9 pointsRohanarun11 comments

So a centralized control model that becomes a single point of failure? No, thanks.

Centralized, monolithic models fail. They're slow to upgrade, slow to adapt, don't fit the edge cases, and suffer from our inability to completely model a problem with 100% accuracy.

And that's not even counting the up-front infrastructure costs that would NEVER make it past any political process. And the 15 year standardization process for the technology.

That's a good point, thanks I'll add it to the post under the top arguments.

My answer would be that the current upgrade speed is to buy a new car, and how many people do you expect will do that in 15 years?

This alternative approach brings up the lower bar, but still allows people to buy special cars like Teslas, if you want. So you can get the independent car if it's important to you, but you would also get the benefit of every other car on the road being safer.

This solution is more geared towards optimizing in the numbers in car crashes

Yes, that's entirely the problem: Seeking an optimal solution when a less-than-optimal solution gives us "good enough" workability while avoiding all of the heavy costs and obsolescence risk of a centralized system.

You're pushing for a technical solution that ignores political problems. The political effort to even get this to the planning stage is already astronomical. Then you get into the endless committees, finding a mayor with the cojones to confront his constituency with the up-front costs, somehow keeping everything compatible between each and every city you do this in (because each will be done by different contractors, each with their own ballooning expenses - 10x cost is not uncommon), and hoping to god the other cities don't get cold feet after watching the vanguard get eaten alive come election day for blowing through $5 billion or more (did I mention ballooning costs?). And then of course there's coordinating with the car companies...

The fact is, autonomous cars as they're being done now will win by default, because besides the already huge political momentum they have, they also involve little-to-no cost for municipalities. Private companies pick up the R&D tab.

That's very true, but in the alternative scenario, you would have citizens who never buy new cars driving in semi autonomous cities.

No matter how hard it may be to get past the political problems, this solution would guarantee more lives would be saved. Whether that's worth it requires weighing against the alternative: lives lost and more years for the lower bar to raise. There's an $800 Billion+ in savings available if cities implement, and they don't even have to be the ones to develop the technology. Politics also change quickly when there's money involved from the players who can implement such a system, like google or apple.

Also a point on obsolescence risk: are you confident enough to buy an autonomous car now? When better autonomous reconstruction systems come out, you would be obsolete unless you buy a new car. A city can upgrade daily.

This feels akin to saying a subway or hospital or giant computing entity can easily upgrade its mission critical systems. Does not feel realistic. These things tend to move slowly once stable. A client is usually easier to upgrade than a server, if all the brains live on the server and the system achieves stability it would rarely be touched; now it's responsible for life and death.

Furthermore it seems that the paradigm of automatically updating each car independently at night ( like modern computers/etc ) would be less prone to major failure. Updates can be rolled out automatically when the vehicle is not use and can be done in stages, not all in one big shot.

Lastly what about bugs and fault. With the cars and systems still being private we get to reuse and evolve from the legal structure we have now, especially including 'fault'. In this utopia idea proposed we have a big official system keeping us all safe. Would imagine the first big failure to be expensive with a lot of finger pointing and a broke city that won't want to pay for all the injuries and damages, or would need a bailout to do so.

This idea is not fault tolerant or distributed enough to to live with our current amount of chaos. Getting to utopia is harder than it sounds.

I don't want to live in a vehicular system with a single point of failure even if that single point of failure is an organization such as a government and this is seen as a utility.

Possibly this day will come, but the idea is way ahead of its time. Good luck!

The day may come sooner than you expect though, because one of the "big players" is actively interested in the idea now..

Your concerns are all valid of course, but this is actually considered a better solution from the people working on the solutions too. It takes a lot to admit you are wrong, so imagine how much better this solution must be for them to do that.

Would appreciate any feedback on how to improve the simulation for testing. For now turning off traffic lights makes the efficiency go up by 10 cars/minute. It would be interesting to test real city models, and different traffic algorithms

I don't understand how this is hacking-resistant; if the car is relying on external information, then any flaw in its authentication of the external information can be exploited, regardless of any changes deployed by the city.

If a car starts moving independent of a city's models, it would easy to identify as a possible hack in real time. If a car gets hacked independently, there may not be much evidence of how it happened. The idea is a city based system can send help faster. A completely independent system can be hacked in multiple different ways to trick the controls, such as through the model, but a city based system would also have minimal interfaces in the actual car.

That makes sense, but I wonder if the city could respond in time to prevent casualties, given the close proximity of autos and pedestrians in most cities.

I agree any real car hack will likely not end well. This solution also introduces the possibility of a new vector for state sponsored attacks, which should be concerning to some.

The utilitarian decision is still to upgrade everyone, because the people who would hack cars to kill would likely do it another way. This may become the new improved way to "cut brakes." A state mandated system could upgrade hacking charges to serious felonies, which deters most except state-sponsored and psychopaths. I have a feeling that real car hacks will be as rare as poison Halloween candy

Also in a situation where someone is hacked, having a dense recorded model of all the roads in a city would be very handy for investigation, since the hacker would likely have to be near the car.

Reply on news.ycombinator.com