Human drivers keep rear-ending Waymos
arstechnica.com
arstechnica.com
It's possible confuse other drivers without technically making a mistake. The article doesn't say whether Waymos are rear-ended more often than human drivers, but it's plausible that this could be the case. Human drivers have a mental model of what other drivers will or won't do, and that model might not work well for autonomous vehicles. It's very possible that Waymos are faster to put on the brakes in unclear situations than a human driver would be, and/or that they categorize different situations as dangerous than a human driver would. I'm sure human drivers will eventually adapt to the behavior of these robots sharing the road with them, but that might take a bit of time and experience.
You might also be involved in less serious accidents: a low or moderate speed rear end collision is safer overall than hitting a pedestrian, T-boning someone, or colliding at the higher speed from before braking.
That's a trade off that Waymo presumably know about, and are taking intentionally. It would be interesting to know what they prioritise: is it fewer overall deaths and injuries, or fewer accidents where they were determined to be at fault?
I was talking to a kid once that was bragging about how his dad always got rear-ended, due to aggressive braking, but it was 'never his fault'. I was like "your dad is an asshole, that isn't something brag about."
The insurers want to know who those people are so they can jack their rate preemptively because there's a non-trivial chance that a) both parties in a collision have the same insurer b) the rear ended party's conduct is both indefensible and caught on camera and their insurer will wind up paying out anyway.
If you've been rear-ended, chances are the person behind you wasn't a very good driver. If you've been rear-ended 7 times, chances are that you are the person who isn't a very good driver.
They tailgate them
That’s not how insurers are permitted to set rates by state regulators and also it’s a good way to go out of business due to what’s known as adverse selection.
There's only one state in the US that prevents data collected from cars being used to increase your rates. Drive at night? Your rates go up. Drive "too often"? Your rates go up. Hit the breaks and successfully avoid hitting an animal or person suddenly running into the road? Your rates go up. Good luck counting on captured regulators to protect you. They did nothing to stop insurance companies from using records purchased from data brokers for ages. The EFF is calling on them to do more but don't expect much. https://www.eff.org/deeplinks/2023/06/steering-mobility-data...
What you’re describing is segmentation, which is another way of saying geico charges the people without speeding tickets less than those with speeding tickets.
If ACME co decides to charge everyone the same, then the drivers without speeding tickets will go switch to geico leaving ACME holding the bag of speeders… and losing a ton of money since prices weren’t set for that.
Switch speeding tickets with hard breaking or whatever other thing correlated with accidents.
Also, as far as fairness goes, using actual driving behavior sound far better to me than using credit scores. But strangely I don’t see nearly the same level of complaints about that
So is it ethical to brake aggressively to avoid a 1% chance of an accident up ahead? 5%? 0.1%?
This is a real life trolley problem, and the answer can't be as simple as "always brake, feel no remorse because it's the other driver's problem" even if that might be a good heuristic for a human to teach in driver's education.
Human drivers are taught to avoid scenarios like that. During drivers' lessons you will be taught to check your mirror before braking, and you will be taught to activate your alarm lights when there's a sudden stop on a highway to alert the drivers after you. Basic self-preservation makes it pretty clear that you don't want a car crashing into your rear.
Waymo doesn't need to care. If the other vehicle is considered at fault, their insurance is going to pay for your damages. It's at worst a minor inconvenience, so why bother spending engineering time on writing code to avoid it?
If I'm going to crash into something I'm going to be slamming the brakes without checking my mirrors first.
OTOH, if you're driving and the light changes to yellow and you have room to stop, you should check your mirrors to see if there's a vehicle behind you that's unlikely to stop.
Additionally, in the impending collision decision, once you've taken initial action, you'll want to check on traffic to see if you have an option to evade the obstacle. And if you don't have an out, if there's a vehicle behind you and try to maintain space ahead and behind if possible.
If there's room to stop, then I'll still be braking immediately because braking immediately gives me the most time to brake, which makes the deceleration as gradual as possible and gives the person behind me the most time to react. I'm not sure what the alternative is supposed to be. Delay braking but having brake harder later? Gun it and try to run the yellow?
It sounds like people defending their terrible driving.
It's your problem too if you stop for a yellow and they don't. They may be 100% at fault, but you're delayed at best, at worst may be injured and may not have a driveable car, and they may not carry insurance, so you may have a large, uncompensated loss.
By 'unlikely to stop', I really mean 'unlikely to stop without striking your vehicle if you were to stop', so things like following too closely, clearly not paying attention (which is hard to tell in a quick glance, but maybe you noticed their lack of attention before and haven't had time or space to make space), has a large vehicle that needs more room to stop, etc. I likely wouldn't stop on a fresh yellow if there was a trash truck following me closely, for example. Even at the extreme where I end up running a fresh red, it's more likely a better result than getting rear ended by a vehicle that's around 10x the weight of mine.
At best that's an argument for not slamming on the brakes on a yellow if you already know that someone is following you, not for you to waste precious seconds diverting your attention to your rear view mirror and trying to make a judgment call. In critical moments like that you really can't afford to divert your attention for whatever is happening behind you, because there could be important stuff happening in front of you as well (eg. the car in front decides to slam on the brakes). If you're really in a situation where you think you can't safely come to a stop because there's a truck tailgating you, you really should get yourself out of that situation rather than trying to run the next yellow.
Which would make it a bad driver.
That vehicle will still be taken out of rotation for repairs. In situations like that, does the insurance typically cover lost revenue as well? Otherwise there's still a ton of risk for Waymo if it's a typical deductible + repair costs. I genuinely know nothing about insuring a business like Waymo.
https://www.mwl-law.com/wp-content/uploads/2019/01/LOSS-OF-U...
But generally, yes, loss of use is generally something that courts recognize as a valid claim.
Thats a very human-centric viewpoint, understandably. A human driver can only see in one direction, and should be looking in the direction where things are happening the vast majority of the time. A self-driving car can see in all directions and in fact is looking in all directions at all times. Perhaps the rules will be updated at some point.
Human drivers are required to be aware of vehicles in all directions at all times.
Cars have mirrors. Any halfway-decent drivers' education will teach you to regularly check them, especially when you're about to brake.
You’d never design your system to intentionally be at fault.
There are relatively few jurisdictions that have a concrete quantitative definition of tailgating. So it's not yet widespread, but law enforcement certainly has the option available now.
In all serousiness, eventually, driving will be removed from humans.
And it's usually in the direction of their phone. Even while driving.
I'm curious if Waymo does any sort of warning brake lights like that if chance of panic braking > 15% or something.
Is that statement based on evidence?
While that seems plausible, it also seems counter-intuitive to me. I suspect that many crashes happen because someone was afraid to apply maximum brakes, either out of fear that someone might hit them from behind, or simply due to being unfamiliar with braking hard.
Once you've tuned your parameters to be driving pretty safely and minimizing accidents, which is what Waymo is surely doing for some definition of "minimizing accidents", it must be true that every further parameter change involves a tradeoff like this.
Not to be rude, but American drivers are not normally considered great, and San Francisco is number 7 on the list of cities you are most likely to get in an accident in the US.
https://www.forbes.com/advisor/legal/auto-accident/cities-mo...
One of 4 cars on a city street? Better try to push the car in front of you along, even if the other 2 in front of them are going the same speed.
Driving 2 miles on a 45 mph road? Better crowd the car in front of you that is already going 50 mph.
And then you get to the freeway where people think it's a great idea to try to push the car in front of you out of the way so you can weave through a little bit of traffic.
Hopefully human drivers learn to apply the same wisdom to their own driving.
Most countries' driving laws for good reason demand that you must maintain enough distance to the vehicle in front of you so that if it hard-breaks surprisingly for _whatever reason_, you will not touch it.
Lots of people ignore that. They are bad drivers. This normalises bad driving.
Robot cars being very good at breaking quickly just makes that more obvious, and will hopefully educate more people to drive in practice how they should have been driving all along.
Basically, it is impossible to keep that distance on half busy road.
I don't really see autonomous vehicles as adding some vastly different curve at me. Wait, what?! They obey the law? How will I ever adjust to that?
Waymos on the other hand, they are so new that other traffic participants don't know what to make of them. Maybe it improves in the future because people get more experienced, maybe it doesn't because the alien intelligence is not compatible with human intelligence and those things are inherently unpredictable. We'll see.
Edit: I've been down voted so as clarification I mean a sudden change without any obvious/predictable reason. Think the person that realized too late that they wanted to turn here and instead of continuing and turning around they slam the breaks. Yes you need to be ready for this, but it's still more risky.
Having a child who's a runner made it clear just how much the world is literally asphalt ribbons of death right out everyone's front door.
Bingo. (Emphasis mine.)
Design knowing full well that humans are error prone, and account for that. Wisdom popularized by Donald Norman's book Design of Everyday Things [1988].
We added a 3rd brake light, distinct from left & right turn signal lights, to reduce rear-ending. It worked, right?
Changes will be needed to accomodate robotaxis. For instance, one suggestion I read, repeated here without any judgement, is adding blue running lights, so that humans know a vehical is autonomous.
--
From the hip, because I don't know anything about robotaxis or road safety rules: Waymo could use those blue lights to signal when someone is tailgating too closely. As in too close for the robotaxi's own stopping distance.
Who knows? Someone will figure something out.
Let’s not get caught up handwringing about predictability since it’s very possible that being less predictive is safer overall.
We have the safety data to prove that Waymo is safer than human drivers, so I don’t think this predictability thing is that important afterall.
...no orange turn signals, then?
The right thing to do for the autonomous software would be to break hard to avoid killing that person.
If there was a car behind that Uber car, it would likely rear end the Uber. That would be preferable to killing the human. The software would be right. Unpredictable but right.
If the read ending happens because Waymo erroneously did phantom break (i.e. it did break because software was confused), then software would be wrong.
Without analyzing rear ending accidents to decide if the software was right or wrong, we can't say if Waymo did the right thing or not.
In human-to-human crashes, rear-ender is presumed guilty.
The person in front may also be guilty, if they stop for no good reason, or a malicious reason (aka brake checking).
In ye olden days before 1080p everywhere it was considered somewhat common courtesy to hang around and make a statement to the police if you witnessed a rear ending that was undeniably the fault of the party who got rear ended.
Often, situations like:
* You are stopped in traffic, the driver in the front reverses into you.
* Traffic comes to a quick stop, you stop, and the driver behind you rear ends you and pushes you into the vehicle in front of you.
Brake checking and unsafe lane changes might result in a split of fault in places that assign comparative negligence, which might be more of what you're thinking of. Often the idea here is that the driver in the rear could have done something more to avoid the accident. This is less likely to be determined to be the case when a vehicle is completely stopped in traffic and the driver has no options.
Yup. Just drive the exact speed limit on every highway you're on. And if you're on larger width highways, you can do that in the left lane, since you're at the legal maximum allowed speed.
Driving safely and driving perfectly legally are not always compatible.
Exactly this. One of the most important things that enhances safety is to avoid doing anything that surprises others.
We don’t need to guess here, Waymo is way safer than human drivers as a car and a pedestrian. So clearly optimizing for “not confusing human drivers” is not correct.
Human drivers frequently run into stationary objects and plow over bikers/pedestrians so maybe there’s a huge category of situations where they *should* be surprised
That's why I said "one of the most important", not "the most important". Driving in the least surprising way is one of the methods of avoiding accidents.
If Waymo's driving style means that they get rear-ended more than normal, that means that Waymo is behaving in a way that surprises other drivers. While it's true that the fault is the driver that did the rear-ending, it can also be true that the accident may have been avoided if Waymo drove in a manner that didn't surprise the tailgater.
Yeah I experience that in my own truck (a 2022 dodge ram provided by and for work) has this awful delay in acceleration. It’s almost like it needs to “spin up” before there’s power to move, and it’s generally slow when it comes to moving up gears.
This results in people misjudging my vehicular body language, thinking that I’m waiting/indecisive when I’m not. So by the time I’ve started moving from a stop, the other guy got convinced I was letting them go first and is also moving.
Activating “Tow Haul” mode every time I start up seems to mostly, but not entirely mitigates the lethargy.
Also the wheelbase is too damn long, which makes reasonable turns in parking lots into curbs scraping or 3-7 point affairs to get out of.
I‘ve driven boats with better handling than this stupid vehicle.
- AV approaches intersection where some obstacle blocks cross-street visibility.
- AV slowly enters intersection to get better visibility.
- AV detects cross traffic and stops quickly.
- Trailing human driver following too closely rear-ends AV.
Back when Google was testing in Mountain View, this was the most common accident. One particular intersection had a tree in the median strip, at a height that blocked the vehicle-top LIDAR, forcing a slow, careful entry to the intersection. At least two Google AVs were rear-ended there.
As the article says, there are no recorded incidents where a Waymo AV entered an intersection with legit cross-traffic and was hit. There's one where a human driver ran a red light, clearly not the fault of the AV.
One partial solution would be to have AVs flash their brake lights rapidly when they're in this may-brake-suddenly situation. That would warn humans to back off. AVs know when they're being tailgated.
I've said this on here many times before, but one of the reasons I love riding in Waymos is because they (in my experience) obey all traffic laws to the letter of the law. So if there's a stop sign, they all stop.
Would love to know the specifics of these rear-end collisions because I'd bet that they're either California rolls at stop signs or doing the same roll behavior when turning right on red traffic lights.
We put far, far too many entrances and exits directly into traffic here in the US. The behavior to normalize should be directing folks onto back roads with much saner speed limits.
Where are you finding these unnecessary STOP signs?
But, having used and been behind other assisted driving vehicles. (Which one thing I find exhausting with auto cruise is that people pass, usually on the right, and then cut in constantly if you have the distance set safely)…
But as another commenter said about mental models of drivers. I wonder if part of it is they brake sooner, and perhaps more completely.
Example: someone turning left. Most human drivers will slow and go around if safe. Does waymo brake far away, and abruptly?
Again, I can’t comment on waymo, but I know my own vehicle too will sometimes brake late and harder if set to auto because of sensor distance. (I try to not let this happen and over ride and brake sooner of course, gradually slowing vs flying up, detecting a stop or slowdown, and then braking aggressively )
I’d submit aggressive drivers aren’t used to that.
Waymos indeed make odd and non-human maneuvers. A quite frequent one is inching forward jerkily as it pulls over (<5mph) at which point there's usually no-one moving nearby either.
Waymo has veered into oncoming lanes a few times, though traffic was stopped at that intersection. I've really never witnessed any risky maneuver that would endanger property, humans, or even warrant a ticket.
BUT... motorists and pedestrians HATE Waymo with a passion. I was literally spit upon when riding a scooter-share and it may get worse in those comfy white Jaguars.
No overt incidents yet, but several passers-by have vocally expressed their ire, and there is no shortage of folks who deliberately block our path just because I have no control. But I suspect, violence will arise as the pressure rises.
Still, it is and always has been completely and totally insane to test these heavy fast robots on public streets. I never in a million years would have thought we would do that.
The obvious thing to do is:
A) Start with light slow robots, like a golf cart covered in foam, and work your way up to Knight Industries Two Thousand. Going directly for KITT is a fetish.
B) Test it in one of those artificial cities with, you know, people who have at least signed waivers to be human test dummies for your two-ton high-speed killer robots?
E.g.:
> Mcity is a 32-acre simulated urban and suburban environment that includes a network of roads with intersections, traffic signs and signals, streetlights, building facades, sidewalks and construction obstacles. It is designed to support rigorous, repeatable testing of new technologies before they are tried out on public streets and highways.
https://news.umich.edu/u-m-opens-mcity-test-environment-for-...
A. They used to have these cute little cars (e.g. golf cart sized) that couldn't go over 30 mph, long before they rolled out the Chryslers and Jaguars
B. My memory is fuzzy on the details, but I'm sure there's public info on this: Waymo owned or leased some massive plot of land where they did exactly that, testing all sorts of chaotic scenarios (weird construction, skateboards coming out of nowhere, etc)
None of this "well the manufacturer says it works so let's just let 'er rip" that has already gotten people killed.
And yes human drivers should have to pass the same test in the vehicle they intend to drive.
Just like the existing uniform state run test that shows how all licensed human drivers are safe? It shouldn't need to be pointed out how any kind of standardized test isn't going to prove or even realistically evaluate the safety of a system like this. We can't even get a standardized test for GPU performance that manufacturers don't cheat on.
> The test would also measure how the computer driven car reacts when one or all of its sensors are damaged.
I do agree with this, though: we already expect cars to behave well in the face of hardware failures. Autonomous or no, cars should have requirements for failure modes.
Another comment pointed out that Google did roll out in an extremely safe way, but to this point: do you still consider this to be the testing phase? You might call it that but if we're at a point where "they are, as this article points out, already safer than human drivers" (and in particular, nearly 2^3 times better in some cases) that would suggest that the fundamental concept of the vehicles no longer need testing.
It's simply not possible to develop autonomous vehicles completely in simulated environments. There's two kinds of testing:
1. Testing that the car behaves appropriately such that it's suitable for public use. This is the phase where you're not sure that it's safe.
2. Iteratively improving the vehicle after it's been deemed safe. This is where you're rolling out changes that you believe will improve safety and comfort on an existing safe product.
Google already did #1. They did it for over a decade. If you believe that #1 and #2 are not two separate phases, then if #1 can't be evaluated with real human drivers you can't ever get to a place where you deem the vehicles safe. There's no threshold where you can confidently say the cars will behave well because you've never put them in front of real, unpaid humans on real, non-simulated roads. How could you call them safe in good conscience if you've never tested them outside of a test environment? How do you know you've controlled every last variable? You simply don't and can't.
Proposal accepted.
(The first "auto-" is self-moving as contrasted with the cart that needed an external motive (horse etc.), then the second "auto-" is self-directing with sensors and feedback systems forming a crude "mind" with a "self image" (all efficient regulators contain a model of themselves[1]). A self-motile and self-guided machine is an auto-auto-mobile. I love it.)
Comparing self-driving cars with human car drivers is stupid, a more fair comparison is trains or trams, which also are less prone to accidents and you can't be in a hurry when taking a ride and make them go faster.
I had to take over the driving in these places because cars would pile up behind me, drivers would get impatient, and in some cases people would get angry and tailgate or try to pass where they shouldn't.
Turns out there is a FSD setting that lets the car adjust to conditions, and it will "go with the flow", even going over the speed limit, but matching the traffic and harmonizing.
Note there is a concept traffic engineers use called "the 85th percentile rule" that helps design highways by setting the speed limit for maximum compliance. Using this rule leads to the safest roads.
Some folks seem to be outraged that people don't comply with the speed limits, but when they are set artificially low, speed variance increases and roads become more dangerous. I think driving the speed limit or below in these places might amplify this.
data on speed variance vs safety (fig 4):
https://www.fhwa.dot.gov/publications/research/safety/17098/...
The fundamental flaw in this concept is that it completely ignores that 1) the ideal speed depends on external circumstances and intended road use, and that 2) road design heavily influences driving speed. If you want to have a safe road network, you have to design it for a specific speed. When done properly, drivers will naturally be driving at the intended speed. Speeding doesn't happen because it doesn't feel safe to speed.
Highways have to be straight and wide, without any level crossings. People are supposed to drive at high speeds, so it is designed in such a way that it is safe to do so, and it feels safe to do so. Local access roads are narrow, twisty, have relatively poor visibility, and are equipped with things like speed bumps. You're supposed to drive at slow speeds because there are lots of level crossings and driveways, so they are intentionally designed to make high-speed driving unsafe to the point of being physically impossible.
If you want a lower speed limit, you can just set it, and if below the 85th percentile speed, you'll get a less safe road.
The proper way to do it would be to adjust the characteristics of the road so that drivers slow down and the speed variance is minimized.
Same for increasing the limit. A wider or smoother or straighter road may work.