Yesterday, we reported on a Cybertruck that held up quite well in a stupid wreck it seems to have caused itself. Well, not really itself: the inattention of the person who was supposed to be driving was the major factor. The real issue seemed to have been how humans interact with these sorts of advanced driver assist systems, known as ADAS systems. A new study published in Transportation Research Interdisciplinary Perspectives about ADAS systems gives some interesting perspectives, and reinforces some findings that we’ve been aware of for a while, but are worth remembering.
One thing this study introduced me to that I hadn’t really considered before was the classification of ADAS systems into safety-enhancing and comfort-enhancing, but that’s a good way to think about these systems and how they’re intended to interact with us. Perhaps unsurprisingly, the study found that comfort-enhancing systems like cruise control and even adaptive cruise control do not seem to bring safety improvements.
The results were really quite interesting; here’s a summation from the abstract (The exclamation points after all the initialisms is how the copy appears in the study, which makes everything feel weird, just so you know):
The data analysis showed that LKA! (LKA!) (−19.1%) and DMS! (DMS!) (−14%) had the strongest crash rate reduction effects, followed by AEB! (AEB!) (−10.7%). However, systems like ACC! (ACC!) and CC! (CC!) were associated with increased crash rates (+8%, +12%). Categorizing systems by either functional class or interaction type revealed central tendencies favoring safety of longitudinal control and intervening systems, while comfort-enhancing systems showed detrimental effects.
LKA is lane keeping assist, DMS is driver monitoring systems – the thing that makes sure you’re paying attention – and AEB is automatic emergency braking. ACC and CC are adaptive cruise control and regular cruise, which actually were found to be associated with more crashes!
I think the explanation for why that seems to be the case has to do with the same issues we’ve seen with Level 2 semi-automated systems like Tesla’s FSD or GM’s SuperCruise: the vigilance problem. If a system does most of the work, it’s hard for people to remain focused and vigilant on what the system is doing. Couple that with, as the study states (emphasis mine):
“However, it is not a requirement that ACC! detects stationary or slow-moving objects (International Organization for Standardization, 2018), and users might place too much trust in the systems’ capabilities, tempting them to engage in distractions, which could both be detrimental to safety (Moeckli et al., 2015, De Winter et al., 2014, Hoedemaeker and Brookhuis, 1998).”
The problem, as always, is with us. We’re humans, we get distracted easily, and comfort-enhancing systems tend to just make us even more distracted, because of course they do. That’s sort of their point, perversely, even if we pretend like it’s not. We couch these systems as being advantageous for safety, when what they really do is allow our focus to wander away from the task of driving.
The study also is good for thinking about these systems taxonomically. The paper breaks down these systems into two main classes, Functional and User Interaction:
User interaction
refers to a categorization of the way that systems interact with the driver and/or vehicle. Informing systems passively provide information; warning systems are also passive, in the sense that they cannot exert control over the vehicle, but these systems convey their information by actively drawing attention when pertinent. Intervening systems are like warning systems in that they act in case of emergency, but these systems do exert control over the vehicle. Finally, comfort-enhancing systems also control the vehicle, albeit at the driver’s request, that is, when there is no immediate need to intervene to ensure safety.
Another interesting thing this study does is to introduce some new categorization ideas, based around the concepts of control and urgency:
Informing systems are just like passive nav systems and things like that: information without related actions. Warning systems like automatic collision detection have lots of urgency, but don’t necessarily take control. An intervening system is something like automatic emergency braking: lots of urgency, lots of control. And finally, cruise control and adaptive cruise do a lot of controlling, but minimal urgency.
This quadrant system allows them to have charts like these:
The paper isn’t all that long, and I suggest giving it at least a quick read; I think we’re at a point where the technology for automated and assisted driving systems is getting quite advanced, and is rapidly passing our own overall thought about what sort of systems actually make sense, and how these systems should be implemented.
Comfort-enhancing systems are, of course, good and desirable, but we need to really consider the safety losses that they can bring, at least in their current states. We also need to be realistic about who we, as humans are, and not turn away from our failings because they will come back to bite us in the ass. If we can feel like we’re getting away with doing something like driving while paying less attention, we will happily do that, and not think about safety at all. We’re kind of idiots that way.
The more of these studies we can see, the better as far as I’m concerned. This is a drum we need to keep hammering upon, because carmakers will always just give us what sells unless forced to change by outside forces. It would be nice if, for once, we could actually try to do the smart thing from the get-go.
I’m really liking Jason’s common sense articles on tech in cars. We do not have to accept what the manufacturers are selling. Carry on!
My eye-opening experience with LKA happened while riding shotgun with my then-80 year old Dad. He had been raving about how great his new Prius was because, “It practically drives itself!”. We did the 40 minute drive from his place to my sister’s, pinballing from one side of the lane to the other the entire way there and back. He thought it was all great because the car was handling everything, but if something unexpected had happened – say, a hay bale dropping off of a truck in front of him – we’d have been toast. Needless to say, we got him out of that car shortly thereafter. But it was scary how the LKA basically allowed someone who shouldn’t have been on the road at all to pretty much appear as though they could drive. Terrible stuff.
And! that! is! a! really! interesting! study! Well! worth! the! read!
Just wondering when sitting behind the wheel with LKA on just what do people do rather than driving themselves?
I’m not some big fancy writer on a famous car emag. However, if it was up to me I’d put the definitions of the abbreviations before or next to the first use of them rather than at to he end. A whole lot less scrolling that way.
I seem to remember these level 2s having a requirement to make sure the driver was looking ahead or paying attention or had their hands on the wheel or something. Is that no longer a thing? Otherwise how did Cyberdud wreck the truck?
It makes sense as one encourages checking out and the other assists someone who is only half checked out, but still (sort of) operating the vehicle. Personally, I hate all these systems and I treat every drive like a sortie (as someone with ADD, driving is one of the few things I’m actually able to maintain long term focu—shit, what kind of bird just flew by my window?), so I only notice these systems when they (not infrequently) malfunction and their unexpected alerts/intervention raise my stress level for no damn reason and require me to shout a long string of curses and creatively gruesome threats to the entire lineages of every asshole who works on the things.
While on a boring highway drive, I turned on LCA in my wife’s 2024 Subaru Impreza just to see what it would do. It took about 60 seconds for her to look over and say “why the hell are you suddenly driving like an idiot – do you want me to barf?” All it did was drift from one side of the lane to the other, never holding a straight line. I suppose that’s better than my Camry, though, which consistently picks trying to drag itself along a Jersey barrier over kissing the yellow line, especially in the fast lane. It’s like it has a death wish. Both systems are permanently turned off…
I once got pulled over in Eureka CA for weaving within my lane. When the cop said that I replied so you are pulling me over for driving legally? He laughed and said I guess so. It is a great story involving hard of hearing cops, a 1 on 8 cop fight, and 6 hours of Barney Miller followed up by calling up my boss to explain why I would not make it into work after he had already been contacted by the newsroom of the newspaper we both worked informing him of the arrest and asking if they should publish it per normal operating procedures. The answer was yes.
I only experience with lane keeping assist is in my wife’s Subaru cross tek when I visit her in Brooklyn. With 100 years of infrastructure like old trolley tracks, seams in the pavement after excavations and the cryptic messages that ConEd workers leave in the street randomly popping up, it’s like the car has a severe case of ADD and drives like a dog chasing squirrels. God only knows what it would do in Manhattan, where the avenues randomly go from four lanes to five lanes wide and back again.
As a human driver, I just adhere to the general rule that it’s probably a good idea to have an equal amount of space on each side of the car and maybe if it’s less than say 8 inches and closing, maybe it’s time to hit the brakes.
Of course, if I drove like that in my part of California, I’d probably get shot.
My wife has a 2021 Honda Pilot.
ACC works just fine. I really like it on long trips.
AEB is turned off after the stupid car would slam on the brakes for no reason. This happened multiple times and almost caused a few accidents. It still alerts to brake when it detects oncoming traffic while going around a curve or for any other reason it can find. We fucking hate this ‘feature’ and it scares the crap out of my wife every time it ‘alerts’.
LKS is hilarious. On very well marked roads, LKS will snake between the white lines. Seriously, it just snakes the entire time.
If the road is anything other than perfect, LKS goes insane. It can’t figure out where to go. This POS is also turned off and never used.
I drove a 2022 Honda Accord and AEB and LKS were just as bad.
I realized recently that LKA induces within myself a sense that I might be able to devote not quite as much energy to the task than as without such a feature. This is of course a ridiculous and stupid thought, but probably the exact thing Jason has been highlighting for years. The LKA system in my car isn’t very good either, so I have no real trust in it, yet here I am having such thoughts. I know it’s borne of fatigue because I like doing a good job driving, but I still caught myself thinking it. If it can happen to me, it can happen to others.
I’m beginning to think the only safe way to use these systems is in a totally adversarial manner wherein you take it as a challenge to find them failing and derive enjoyment from catching those failures. Anything else is essentially Russian Roulette.
So ask yourself, how long can I play something that requires complete attention without a break? That’s essentially the situation with these assist systems. Would you willingly take a basketball or hockey pick to the face? Would you play football blindfolded? Of course not, but you might trust a driver assist system?
Back in college I got to test out some pre-production ACC vehicles. The “adversarial manner” driving was quite amusing, testing out what would happen with situations such as two ACC enabled cars following a 3rd car half in each lane, or switching lanes half a car length behind another car (we confirmed that the radar wouldn’t pick up a car less than 6 feet ahead). One thing I decided wasn’t worth the risk testing was a situation where I was in the right lane of the interstate on a curve, the car in front of me that ACC was following was exiting straight, and everyone was quickly coming to a stop just ahead, so I hit the brakes instead of waiting to see if the car would figure it out. It was a fun day of experiments, but not something I’d actually want as a consumer.
Yes, the more that we can objectively show the stupidity and safety tradeoff for these conveniences, the better.
I can’t remember where I read it, or whether it’s all anecdotal, but I saw where ACC and LKA reduced fatigue by about 10% over very specific, long drives, allowing people to drive a little more before getting the perception that their fatigue is making them dangerous…but this ironically shows that they’re probably becoming more dangerous along the way.
I’m still a little amazed when I see people get on the interstate for just a couple miles and go straight to cruise control, like it’s required. I suspect the future for these aids will be sections of highway where they’re allowed, but most public roads where they’re not. And enforcement will be a huge PITA.
The exclamation points and second mention in the brackets make me think this study was written by Chappel Roan.
I’m Rick James, B*TCH! (B*TCH!)
If you read the report, however, it’ll note that ACC (+8%) is ‘safer’ than basic CC (+12%). Implying that the increased automation of the ACC is arguably safer.
My car’s primitive LKA, and AEB seem to work well enough for me to consider them a reasonable safety backup. And the data supports that theory showing LKA , and AEB provides a 19.1%, and 10.7% reduction, respectively, in crash rate.
Not “safer” – less unsafe. And, with the context of this paper, I’m not sure those two data points tell us there is a linear relationship here.
No.
No.
Nope!
Golden Age Of Enthusiast Cars? the Dump asks. People can be enthusiastic about all manner of absurd things. I’d argue that Engagement is The critical aspect of an enthusiast car. Also that peak average Driver competence is long past, definitely pre connected car, possibly pre internet, since we had to amuse ourselves by building and having fun with gocarts and minibikes which was the style at the time.
They didn’t have any white onions, because of the war. The only thing you could get was those big yellow ones…
Wait, does somebody make a bugeye front clip for NA Miata‘s?
Requires passenger compartment filling robot purchase.
I mean, someone clearly should
Very unimpressed with the LKA system in my Escape. It’s doesn’t seem to work but about 50% of the time. It seems to engage mostly when I’m clearly changing lanes. When I drift over a bit, sometime it will gently nudge me so I stay in the lane, but sometimes it doesn’t, yet it doesn’t shake the steering wheel like when I’m actually changing lanes. Yes, putting on the turn signal does let you change lanes without the shaking, but if I’m the only person on the road I don’t see the need to do that.
You’re the only person you SEE on the road. If you knew you were going to block another driver you wouldn’t change lanes in the first place – it’s exactly when you DON’T know that another vehicle is a factor that signalling becomes important. And, given what you’ve explained about your driving habits, I’m less than confident that are maintaining an active and high-frequency scan of the lanes around you.
Don’t make this a decision, because you will inevitably be wrong. Just signal every time you change lanes, exactly the same way.
I hate the ACC system in my wife’s RAV4. I’m cruising down the interstate, monitoring the road for debris and pot holes, scanning the ditches and roadside for deer, waiting for the car in front of me to get just a bit closer so I can go around… and after a bit I finally realize it never got closer and I’m doing 65 in an 80 zone. Lots of cars are now flying past me and I can’t get in the other lane safely to pass the slow vehicle.
Not all new car features are improvements.
Yep – ACC is clearly programmed by lawyers and the ghost of my wife’s dead grandmother, both of whom demand at least 10 (very large) car lengths between you and the next vehicle.
I tried. I really tried to make it through this post, but my mind kept wandering and I dropped my phone. What were you saying?
First, there’s plenty of space on that chart for the acronym’s expansions. I don’t care about elks and printed circuit boards.
Next, it’s interesting what this study means for the bar for acceptable un-safety for convenience. You’re not getting cruise control out of people’s hands. If any new system comes in and credibly says “as safe as cruise control” you can sell quite a bit of risk to people. Which is roughly where we’re going.
Finally, I still think we’re progressing towards more safety. I think driving is getting harder. I think commutes in congested traffic are worse – partly because of housing issues, partly because we’re not all doing it 5x a week so we’re less practiced – and I think we’re expected to be responsive to our phones even when driving.
It’s difficult for many – especially under 25 – to ignore their phones. Having to ignore them while driving comes at a real cognitive load cost.
We’re also kinda expected to be perfect navigators because of our phones. I used to often do plenty of u-turns when I was navigating pre-gps. It was normal, nobody said anything, I didn’t think much about it. I didn’t mind skipping an exit because the merge was too unpleasant. I could figure out a u-turn via the next stop or something else. That kind of flexibility isn’t really as implicit, I think, now when you’re accustomed to perfect routing every time.
I’m looking forward to self-driving cars, I really am. When I’m in my truck, I just want to get places. I’m happy to let it drive itself. I have motorcycles for enjoyment.
The flip side to “expected to be perfect navigators because of our phones” is I swear I see more people swerving across multiple lanes of traffic to get off the freeway at the last minute than i used to-which I would theorize is caused by people planning ahead less bc they’re just waiting for directions from their phone. Of course the irony is thanks to having GPS missing your exit is less of a big deal than ever and certainly not worth getting in a wreck over.
There’s parts of human behaviour called cognitive load an minimum engagement. Cognitive load is the required mental effort to perform a task, be that solving a problem, focusing on something, or even simply staying awake. Minimum engagement is the required level of cognitive load for the brain to properly dedicate resources to a given task.
Driver assist systems reduce cognitive load to almost nothing. As a result the task of driving often does not reach the threshold for minimum engagement. Thus the brain tries to occupy itself by generating another task that does reach the threshold for minimum engagement.
Often the fastest and most accessible way to do this is with interactive media. Watching video feeds on your phone, playing your Nintendo Switch, so on and so forth. When you don’t have access to these there’s a yearning for them because it’s stimulation you don’t currently have. Your brain is doing something else rather than the task at hand.
In recent years as multiple things have failed the meet the threshold for minimum engagement because the cognitive load has been offset to somewhere else, there have been calls to reintroduce tedium and complexity. Psychologists are doing studies that indicate (but do not have definite results yet) doing so increases attention span and prevents neuron atrophy.
In short automation beyond a certain level drives you more and more towards being Frito Pendejo levels of stupid.
I had to look up Frito Pendejo… I forgot he was the guy from the documentary Idiocracy.
Wasn’t he one of President Dwayne Elizondo Mountain Dew Camacho’s cabinet members?
He was the lawyer of the main character who screwed up so spectacularly that he got his client imprisoned. He later became vice president when Camacho ended his term because the main character wanted someone so stupid as to be ineffective in meddling.
Ah yes, Dax Shepherd’s character… I actually totally forgot his name.
100% agree on all points! I used to work as a quality assurance evaluator and a major part of the job was performing over-the-shoulder evaluations on people performing a tedious, repetitive job. As taxing as you might think performing a tedious, repetitive task is, you should try watching someone do a tedious, repetitive task. It was so much more mentally taxing being the evaluator than it ever was for me being the person being evaluate. When you’re the person performing the task you know what you’re about to do and can take little mental breaks because you’re the one in control; when you’re the one monitoring things, you never know what the other person is about to do or where they might do something wrong and you may need to intervene (in case of imminent safety concerns) or take detailed mental notes (as to not tip the other party off with a lot of furious scribbling).
I look at driving in much the same way. Until the manufacturers are willing to take liability for accidents for FSD crashes, they pose an increased risk. Everything short of that which removes engagement (and associated focus) from the driver is a negative since the human brain will just find something else to occupy itself…
The other day my funtodrive2020hondacrv slammed on the brakes seemingly out of nowhere and scared the shit out of everybody involved. I guess it doesn’t make sense to turn off safety features, but if they’re classified as driver aids, I say the less drivers, the better, and I’d disable it if I could. I don’t let my passenger co-drive, so why would I want the computer to take control?
This is an interesting juxtaposition with Mercedes’ piece on seat belts from yesterday.
It’s great that there are now studies that quantify it, but it’s really common sense to think that systems that require less attention will be given less attention, potentially resulting in worse results. Combine this with the very rudimentary warning systems that sound the alarm because of a particularly dark shadow, or a tight turn with walls on the outside. It really makes me skeptical of mandates for these technologies.