Image transcript:
White background meme with the text “People want to build a massive connected network of driverless cars to ‘solve’ traffic. Doesn’t that sound a bit like giving SkyNet control of all the nukes?”
There are illustrative images of the proposed “network of driverless cars” as well as a menacing image of a terminator robot and an image from a news report about a deadly crash caused by a self-driving Uber.
As someone working in embedded machine learning, I can say two things:
- Deep learning models, as powerful as they are, are the epitome of black box. We know they work, but we’re still struggling to understand how they work. Further, they fail in oftentimes really weird and unexpected ways. So if we want to build safety-critical systems that depend on deep learning, that’s a very tall order. And I don’t think we’re there yet with the tech.
- I’ve never worked with automotive embedded software, but my understanding is it’s very hard and very complicated, and there’s a real shortage of embedded engineers going into automotive. Trying to build incredibly complex, fault-tolerant embedded software systems for a widespread network of driverless cars sounds like an absolute nightmare and an absolutely gargantuan task.
Theoretically, I don’t think it’s impossible to build this vision of the future, but I think it would be stupendously difficult, take a lot more time than tech bros would like to lead you to believe, and ultimately just be a worse transit system than automated trains.
Not to mention being a pedestrian crossing the street in such a world would be a nightmare.
Now how much easier to program would this all be for say… A train?
Waaaaay easier. We already have numerous automated metro systems in the world, from Copenhagen Metro to Vancouver SkyTrain to Montreal REM to Honolulu Skyline to airport people movers galore to probably a whole bunch more systems I can’t name off the top of my head.
Having rails, dedicated infrastructure, and a grade-separated right-of-way works wonders for eliminating variables. You don’t even need AI for automated trains.
W for trains once again
Trains can’t stop winning.
I think 2 would be massively easier to solve if there were no human drivers. Drop private ownership, every car a self-driving uber. We’d need much less of them, they could coordinate without having to account for irrational human behavior. And they could be mandated to always give way at crossings.
Public transport is still superior of course, but I think it’s something that could make cars less objectionable.
Pretty much my thought process. It’s the humans doing human things that are difficult to code for. Honestly just putting Bluetooth in every car so you can see how fast they are moving and maybe predicting their projectory would be useful.
Im thinking birds eye view like in GTA or something. You can see where you want to go and what obstacles are in the way. Gives you a lot longer to react than if it’s just your view out of the window.
That can be done more reliably with lidar. Adding BT and other wireless dependencies to critical control systems opens up new vulnerabilities and attack vectors. Sociopath hackers would use it to fake an imminent collision and cause crashes.
The most efficient path forward is higher density cities, less low density sprawl, and free mass transit (trains, light rail, and buses) to remove the dependency on private cars, then gradually upgrading to driverless only lanes and roadways as private vehicle traffic is reduced over time; much more realistic than waiting for tech companies and politicians to solve complex technical and regulatory problems that could take many decades.
The government should also be directing investment and subsidies to smaller single-person “pod” transport options too, as it’d be cheaper, easier, and more efficient overall if we could accommodate the majority of traffic in both directions within a single car lane, freeing up space for future transition work — after-all, most road traffic is a single person in a 5+ seater car.
Lidar can’t do that. It can ping things but not beyond those initial ones. ? Correct it’s radar but for light ? Obviously BT would create a whole new havock but it might also make life a lot easier because you know where things are that you can’t see.
You can’t see past the lorry in front of you but you can see a car is a few miles away up the road. You’ve time to pass. In the dark you can’t see but BT can.
Hackers can’t do shit. This always gets blown out. Hackers will screw systems and kill people, hackers will do this and that. Yeah I’m sure maybe a hundred can actually do that. So you’ve got 100 deaths.
I think there’s already been 30 deaths on our roads this year. I think the hacker thing is a red herring.
Absolutely agreed but I don’t live in a city so none of those things help me. Cities are pretty easy to solve some traffic woes. But for country it’s much harder and where I need driverless.
Absolutely agree. I was just thinking about this the other day. I just need a tiny transport vehicle for me. I don’t need 5 seats and a boot 90% of the time.
It wouldn’t. Have you seen those massive drone light displays? they lose drones during the shows. Nothing indicates that machine learning would fare better. Because reality is way more complex than a computer can simulate. I mean, even in an entirely humanless car network, a single misplaced traffic cone could send the whole system into shutdown. No system is failproof.
I don’t know about that, it’s all about tolerances. Losing a drone from a cloud is probably considered an acceptable loss.
How many trains crash every year (leaving out poor maintenance for the moment)? Those systems are highly complex and almost fully automated. AI’s not even really needed.
Exactly, in a swarm system, losing a drone is fine. A car fully automated network, as CGP suggest, is a swarm system. If they are cars with people inside of it, it isn’t acceptable to lose units, we can’t accept even a single autonomous car randomly losing control into a tree. No matter that humans do that. The system has to be better than humans, not equally bad. Train systems are inherently free from this variation.
Skynet theories aside, the whole “AI will solve trafic” thing is just dumb. Sure, I suppose we could one day perfect a system of cars working as a hivemind to optimize the flow of trafic, but that just makes it impossible for anyone not part in that hivemind (cyclists etc.) to use the road.
I suppose the “just one more lane bro” joke sort of applies here in that the problem caused by cars would be “solved” by doubling down on cars.
Banning human-operated vehicles and private car ownership would be two huge, cultural hurdles to clear. So why don’t we do it in two steps? We can ban private car ownership right now to prepare. It would be pretty easy to transition over to everybody driving a vehicle from a car-share system. It could be phased in over time while we’re working on perfecting self-driving. We could probably reduce a lot of parking and vehicle demand, too, since private vehicles sit idle over 95% of the time. Then, when self-driving vehicles are ready, the operators of the car-share vehicles would be in position to switch them all over to autonomous mode, en masse.
If you think I’m insane to suggest that it’s politically feasible to just ban private car ownership, hey, that’s exactly the point I want to make.
I feel like we could do driverless buses right now if they had dedicated right of ways.
That’s another two step item where we could do the first step because we feel good about the way the tech is trending, and even if the tech fails, oops we only made a better life for everyone with dedicated bus lanes.
Yeah, I’d be down for this. Having the dedicated infrastructure would reduce the number of variables and allow the opportunity for sensors along the route for more checks so you don’t have to 100% rely on a black-box model. Plus, it solves one of the biggest operating expenses of buses (labor), which would make it much cheaper to run very high-frequency bus services.
As a bonus point, you could make them trolleybuses, so if they do something too crazy, they just lose power and stop.
I’d trust driverless cars more than most human drivers, to be honest.
I wouldn’t and don’t. I can’t tell at all what some software is thinking so I don’t know how to interact with it as another driver or as a pedestrian. When I’m crossing the road in front of it, it may stop because it sees me, but then randomly accelerate for impossible to comprehend reasons. With a human behind the wheel when I look at them and see that they acknowledge my presence I am way more confident that they won’t run me over. Also with being in a car the movements of other human drivers make intuitive sense to me.
Regardless, the solution to the worlds car / traffic problem is not more cars. It’s condensed public transit so the point is a bit moot.
![[meme] I'm look at you, CGP Grey (I still love hexagons tho)](https://lemmy.world/pictrs/image/2443592a-4af3-4f31-8c36-b7fc3944e74c.webp){kind=link}