As with many of things related to self-driving cars, the target doesn't have to be "perfection" so much as "better than people on average". Human drivers do dumb things in large groups all the time, like highway pileups in bad weather, or rubbernecking traffic jams. I don't think all the humans on earth ever stopped driving at the same time, though, so...that'll be a first :p
I have an issue with that argument, because it's typically being used together with highly disingenuous statistics (like Teslas our FSD has less accidents per mile than normal drivers) to give the impression that autonomous driving is anywhere close to being "better than people on average". This is just not the case at least not in any region with a broad variety of weather conditions.
> Human drivers do dumb things in large groups all the time, like highway pileups in bad weather, or rubbernecking traffic jams.
Considering that most autonomous driving doesn't even work in really bad weather, I'd argue that humans are still much better despite the pileups.
Agreed. I think the argument is ultimately true, but it's also easy to abuse with bad stats, and being skeptical of each new claim that we've "crossed the line" is a good idea.
Got any facts behind that assertion?
You see, I completely disagree with this assumption. They should be perfect or they shouldn't be allowed on the road at all.
You wouldn't accept a bread slicer that sometimes erroneously slices off your fingers, as long as it does so less frequently than people slice their own fingers on average, would you?
If I posed the question as: Would you prefer to place your hand next to a machine which will accidentally cut your finger off once every 100,000 cuts, or a human chef that will accidentally cut your finger off every once every 50,000 cuts, which would you prefer?
I don't know about you, but I'd give the machine a crack.
I see it the same way - the current "self driving" systems shouldn't be allowed on the road, period, no matter how much safer they are than a statistical driver, unless they can be demonstrated to be completely 100% perfect in all scenarios(other than actual hardware failure - they shouldn't for instance run into a truck that's turning across a highway just because the system chose to ignore it).
What I was thinking about, was that seatbelts are hardly "100% perfect in all scenarios." I mean, they can make it a more difficult to get out of a burning vehicle quickly, just as an example. Even ignoring the possibility of malfunction, or jamming and such, they're not perfect.
The fact is, even if seatbelts decrease your risk overall, they are nowhere near 100% perfect. Nevertheless, many of us choose to use seatbelts, and indeed even chose to do so before widespread legislation.
"sometimes erroneously slices off your fingers" sounds scary, but if it happens less often than people in the wild, this is absolutely a net gain.
I'm not sure I understand where you're coming from here. Are old fashioned injuries with old fashioned knives preferable to newfangled injuries with newfangled equipment? It seems pretty straightforward to me that fewer injuries would be better, all else equal.
Maybe you're imagining all else will not be equal? Maybe the new injuries are more severe, or more likely to happen to innocent bystanders, or something like that. It could be. But I think it's important to be specific about what we're worried about here. Otherwise "fewer is fewer" is all we have to go on.
And yet I have certainly seen bread slicers that are exactly like that. Possibly even worse than "other hand with knife". Maybe the analogy is also not very helpful though, because (at least for those bread slicers I'm thinking of) the user is still majorly in charge of his own safety, unlike FSD.
I would definitely use a robotaxi system that drives better than an average person. I already take taxis driven by someone else, which means that I take some risks anyway.
If a core goal is really to make thoroughfares safer and reduce traveler injuries/fatalities, we seem to be ignoring or discounting some pretty obvious options, like minimizing car traffic
This vision where we all get around in low-passenger-count autonomous robot cages on existing roadways is just one possible future but we seem to fixate on it. At the end of the day the car is a tool but we've allowed it to burrow into the collective psyche
What about making cities walkable, bikeable, segway-able, wheelchair-able, scooter-able, whatever, with proven public transportation options like trains or buses running fixed routes for longer commutes? Much more energy efficient, way easier than autonomous cars, and conducive to much more livable urban areas imo
It will take some effort to change habits and expectations and revamp infrastructure but surely that's as worthwhile as all the time and money going into FSD?
In parts of the city I live in it has become very uncomfortable to drive a car, there are just too many pedestrians and bikers, and this has encouraged be to bike even more. For distances of less than 6km there is no time advantage when using a car. And in the highway I just use normal driver assist functions like lane keeping and adaptive cruise control, and it's enough.
I will go even further and say that reducing injuries and fatalities is just an excuse for FSD development, so the public thinks it's a good idea to go with it. The real reason is that a fleet of autonomous taxis can be very profitable.
There is really no evidence that FSD fleets will reduce car usage or fatalities, so really it could be the wrong solution for the wrong problem.
Electric rideables are the future, and if we could give them storage on public transport and a lane of road, they would go a long way.
This is part of the problem imo
Remote employee "drivers" take over when a car signals that it doesn't know what to do
Because it has lost connectivity due to a technical glitch, an outage in the cell network, lack of coverage (tunnel, …)
I'm extremely confident that self-driving cars will never make a significant improvement on car safety.
I'm also confident that it will never be applicable to anything but highway driving in good weather conditions - we will never see a Level 5 car, those designations have about as much practical relevance to us as the Kardashev scale.
First of all, the safety statistics that we have with modern cars today is such that you need to run 10 million cars for several years before you have gathered enough statistics to say whether a new development has actually improved the situation. For instance if you take the Volvo XC90, there has not been a single fatal accident in the US or UK since the model was launched in 2002, with over 1 million such cars sold in those markets combined. That kind of lag in your dev cycle means it's extremely hard to do any sort of meaningful development of a machine learning system in a way that demonstrably improves safety.
Second of all, human factors such as being distracted by a cell phone or being intoxicated, plus not wearing a seat belt, accounts for the overwhelming majority of fatal accidents. It is, beyond a shadow of a doubt, much easier to build a system that detects such conditions and then deliberately slows down or stops the car. It's already being deployed today and will only increase in the future, so most of the potential safety improvement from self-driving will be eaten up by other tech.
The more immediate concern is self driving vehicles that are clearly not ready to be on public roads being allowed to test (or not being pulled off the roads).
The only way for either of these things to get better is through real-world experience.
Have they not already completed many millions of miles in simulation, back in 2018?
To me, it looks like self-diving capabilities have reached a maxima. The resources poured into it have barely been needle movers in objective metrics.
Yeah, but videos aren't objective metrics. For me, an objective metric is "went $FOO miles in $BAR conditions without needing any intervention".
What would you consider an objective metric?
But who decides whether intervention was necessary? This is still too vague.
Irrelevant - until the car has actually gone without an intervention for whatever the median is for human drivers[1], we can't very well claim that it is safe for use.
[1] The average human driver goes hundreds of thousands of kilometers without crashing.
And I work in robotics.
Driving is terrifying.
But how does one defend themselves when the vehicle does not want to do so due to inexperience or negligence? Best you can do is hope it doesn't decide for you that your infant's carseat is the best place to collide with a semi it just noticed because it was gray in fog, and oh by the way the CEO decided not to use lidar or radar, and oh this particular ice patch was not in the training set. If you don't trust lvl 5 in those conditions, you don't really trust lvl 5.
if you are going to make that argument, you have to explain why we are socialising losses and privatising profits (again)
so some number of people will be injured of killed while self driving cars gain experience, and then private corporations will extract profit from it?
Until there is a emergency vehicle thats unable to get through.
Some genius 10x'er forgets to carry the 1 or something, and all of a sudden an entire city full of cars all collide with pedestrians all at once.
I don't know enough about this stuff to know if this is a realistic thing or just the premise of a bad sci-fi movie.
2 - each satellite is under observation by a team
3 - how many satellites do you think are there actually? Not that many, unlike the driverless cars we are setting up nowadays
4 - Earth is actually big. Satellite pouring down all at once and you'll have maximum like 100 casualties - worse case scenario if 1). doesn't happen. Cars OTOH are in busy cities. One battery to explode and can easily do 100 casualties
That’s definitely not true. A 66KWh battery is not a tactical nuke
One battery fire isn't a problem, but if every 2nd car in a city or country crashes into something in a way that starts a fire, that becomes a very big problem extremely quickly.
I had heard about the rate at which Starlink was launching stuff a while back which is likely what put that idea in my head.
And new ways for us to better destroy ourselves and each other.
Imagine, say, a pandemic like lockdown, but now only authorised vehicles may travel at certain times, or at all, and that software is buggy, or gets hacked, or used malevolently by one or another political faction.
No need to imagine. This was de facto policy for quarantine in most south american countries.
You couldnt drive unless: - you were a doctor or politician, then only if - the weekday corresponded to the last digit of your plate, then if, - only before x time, like 10pm, - etc.
The better approach is decentralized, with each car built to always fend for itself without a server, imho.
Right now it's just a small fleet of driverless taxis. Doesn't this demonstrate the potential for it to be, say, every single Honda in the city? Why even limit it to a city? A single manufacturer's critical infrastructure has some kind of fault (malice, shit code, hardware failure - dealer's choice!) and suddenly Fords stop running nationwide, or internationally.
Then I start thinking of the impact to emergency vehicles caused by a cavalcade of stopped cars, and other similar issues.
Again, I need to point out that I'm not trying to be sensationalist or hyperbolic. I'm ignorant here and am genuinely curious how viable these types of scenarios could be. It's clear that it can happen, I'm just curious to what extent.
I hadn't really thought about new single points of failure vulnerabilities as an aspect of this before.
I'm still a big SDC fan, the pros are so massively bigger than the cons, but you've given me a new con to be aware of.
If there's something weird about the location where they are stuck -- maybe somebody is jamming GPS or whatever -- then they should at least be able to pull over using local sensors, right?
On the other hand, if there's some effect that is tripping up their sensors -- maybe some odd geometry that, for whatever bizarre reason, looks like a wall to them -- this pretty embarassing, but probably an inevitable growing pain for this technology.
I mean yeah, there will be different manufacturers so it's not like every car in a city could come to a standstill. And I'm also speaking from a place of ignorance, so please correct me if I'm off-base. But isn't this kind of a thing (mass stoppages of entire fleets) a cause for concern as we see increased FSD adoption?