Boffins teach cars to listen for the sound of a wet road The sound a tyre makes on a wet road could become part of the road safety arsenal, if a proposal submitted to an IEEE publication becomes widespread. Pre-published at ArXiv, the idea comes from a group of German and US boffins, who reckon that deep learning techniques can help cars detect not only that the road is wet, but how …
Funny that. This morning, on the way to work, I was thinking about just the same thing, self-driving cars on icy roads. I supposed all the current testing is done in California where the weather is pretty good all year around and therefore the cars don't get practice in different conditions. It would be a bit rubbish if the car just decides it won't drive if the temperature goes below 3C. I guess the long term goal would be for a self driving cars to be better at driving than any human, including rally drivers on an ice stage. Now that would be a commute!
"[...] self-driving cars on icy roads."
In Stockholm, Sweden they don't put salt on the roads because the temperatures drop too low for it to have any effect. The snow ploughs remove the loose snow - leaving a hard compressed ice layer that really needs studded tyres to get a grip. They do spread gravel - which in my experience can be counter-productive by acting like ball bearings under braking wheels.
I wonder how good AI will be at recognising that an elk is about to amble across the road in front of the traffic?
I wonder how good AI will be at recognising that an elk is about to amble across the road in front of the traffic?
As far as I recall that has only ever been a problem for Mercedes :)
If only there were some other way of measuring or predicting if the conditions are likely to be icy. I'm no scientist, but I'm told ice is usually quite cold. Maybe our top boffins could see if there's any pattern to how cold it has to be before ice forms and invent some form of... I don't know, thermal meter or something?
You could measure the temperature and if it's low enough for black ice to be a possibility then adopt the required driving style. If they're connected to the internet they could even download weather reports to refine the decision.
As for detecting it in real time, I'm pretty sure it's more reflective than a normal road surface so maybe that could be used to spot patches of the stuff?
"As for detecting it in real time, I'm pretty sure it's more reflective than a normal road surface so maybe that could be used to spot patches of the stuff?"
That's why it's called black ice. Black ice is transmissive, not reflective, so light passes through it, making it nigh invisible. Plus as the article notes, it could be happening at night on a rural unlit road. Either way, you can't rely on light to detect a patch of black ice ahead of time. About the only way to do it properly would be some kind of active forward 3D imaging system that can detect the smooth surface of ice even in pitch black. Something like that I think is still some time off due to the processing requirements and time constraints.
Refraction won't help much in pitch dark conditions since the light has to come from the car, and as noted, black ice practically doesn't reflect, meaning anything coming from the car isn't likely to come back to be detected. As for surface temperature, a cold road is one of the conditions needed to make black ice (which is actually a kind of ice glaze that freezes so quickly it doesn't cloud up).
A distance runner in Dundee writes: the bits of the path that are not white crunchy frost are ASSUMED to be icy until when forced to negotiate them they are found not to be. Note here in Dundee at this time of year and throughout most of the winter I frequently run on unlit paths/roads in pitch darkness relying only on a head torch (a Petzl Tikka Plus2, a very good model).
I once ran 17.8miles almost entirely on white crunchy frost at at least -10c. The icy bit was the flooded bit of the cycle path I couldn't avoid. I was first one through that morning, probably due to the cold. Usually a cyclist had broken the ice before I got there. This time I had to do it. Thick stuff in water half way up my calves. I high stepped it so my foot was coming down perpendicular to the ice.
Further technical note: my twin skin socks meant 200m down the path from the water my shoes were squelching, my leggings were wet but my feet were dry. Another 200m and they were warm again. Now THAT is tech.
"That's why it's called black ice. Black ice is transmissive, not reflective, so light passes through it, making it nigh invisible."
Driving back from Scotland at night some years ago I came across a minibus sideways across the road and a couple of cars who'd managed to stop without hitting it. It was a black Tarmac road and at best one could say it looked a bit wet. You could barely walk on it, it was so slippery. Being a very narrow road, about the same width as the length of the minibus, it took about 8 of us to push the minibus around to get it straightend up. We were struggling to get traction with our feet but we pushed that damned minibus sideways, not along it's normal axis of travel. Those wheels just slid sideways relatively easily.
I can tell you I took it very, VERY carefully past that bit of black ice and for the rest of the trip home!
Ah, no. Black ice is called black ice for a reason. No reflectivity. Having driven mountain passes daily for many years, you just assume it's there from temperature and few other clues, like cars off the road.
And when you hit it, it's too late do much except ride it out. The key is to assume it's there and drive accordingly.
Driving on black ice makes *less* noise than a normal road surface, so if the microphones pick up a reduction in road noise and you've got a temperature sensor that detects that the surface is below freezing, it's a pretty good indicator.
Also the steering gets "light", ie inputs become easier (because the wheels may be slipping) and, of course, at worst, there may be understeer.
All of these can checked by an autonomous car.
Not true. Every car that passed it altered the conditions of the road by driving through it. Furthermore, it's rare for two cars to be exactly alike in terms of physics when they pass over a wet road. What one car can pass safely another may not simply because they're lighter/heavier, have less downforce, balder tires, etc.
"Every car that passed it altered the conditions of the road by driving through it."
Yes - but not to the extent of making a dry road wet.
If you are driving *that* close to the edge of the physical capabilities of your car then get the hell off the road and onto a track where 'getting it wrong' doesn't have the potential to kill bystanders (yes marshals occasionally get killed, but they're made aware of that risk when they sign up)
"Yes - but not to the extent of making a dry road wet."
But YES to the extent that the parts of the road that are wetter and such differ from car to car as the previous car's tires displace the water what was there.
"If you are driving *that* close to the edge of the physical capabilities of your car then get the hell off the road and onto a track where 'getting it wrong' doesn't have the potential to kill bystanders"
Question: How do you KNOW you're close to the physical capabilities of your car at that given moment?
"Question: How do you KNOW you're close to the physical capabilities of your car at that given moment?"
Because I've done various amounts of driver training, including skid pan sessions. I also have eyes and ears, and choose to drive *well within* the capability of the car and road in front of me, it's not a race.
"I'd rather arrive 5 minutes late in this world than 50 years early in the next..."
Most mechanical devices will give you feedback as you approach the limit - for instance tyres start squirming (and the traction available actually increases up to a certain slip angle).
"Because I've done various amounts of driver training, including skid pan sessions. I also have eyes and ears, and choose to drive *well within* the capability of the car and road in front of me, it's not a race."
Thing is, the capability of the car changes over time, and what you THINK is the capability of the car may not be the ACTUAL capability of the car. Remember, the average person keeps a high opinion of oneself which leads to inherent overconfidence.
"Most mechanical devices will give you feedback as you approach the limit - for instance tyres start squirming (and the traction available actually increases up to a certain slip angle)."
And sometimes they don't warn you at all, like a sudden blowout, by which time it's already too late.
"As opposed to being a nice sunny day?"
A friend of mine is a New Orleans native. His description of what happens when they get a rare snow flurry is pretty much like what happens in the southern parts of the UK. The whole place grinds to a halt because the drivers have little to no experience at how to drive in even a light dusting of snow and the authorities don't have large numbers of gritters/ploughs on standby for a once in 10 year event.
Detecting a wet road alone isn't very useful,. Water on top of good tarmac is far less dangerous than water on tarmac which regularly gets diesel and oil dripped on it. A road that is just wet after a shower can be far slippier than the same road after a downpour that has washed all the oil off. Then there's rural wet roads with tractor mud (or worse) on them, the effect of worn tyres, etc.
Sounds like just another pointless gadget designed to lull the careless drivers into a false sense of security. How long will it be before insurance companies start increasing premiums for cars with excessive driver "aids"? They don't care about advertising and boffins, just the hard numbers.
Agree with Phil (is it me or are there a lot of rabid downvoters on this thread?)
How much better is it than say existing car sensors such as wheelspin/traction sensor, windscreen wiper sensor and humidity sensor?
I suspect that for 99.9% of scenarios a simple feature map would do the job just as well.
Seems like a total overkill solution looking for a big enough problem tbh.
Still at least they got a grant out of it.
"How much better is it than say existing car sensors such as wheelspin/traction sensor, windscreen wiper sensor and humidity sensor?"
Traction control gives you feedback only if you push the big pedal and push it hard. An autonomous vehicle is not going to do that. Heck, very few drivers do that. I'm a somewhat aggressive driver (well, my Saab is very easy to drive, so...) and with only 260 bhp I rarely see the traction control warning light flicker on. I can hear the tires squealing often enough though, but that is not enough to trigger the traction control warning light. (nor do the tyres squeal loudly when the road is wet)
The wiper sensor tells you if it is raining. But it tells you nothing about what is actually on the road. The road might be relatively dry still while the winshield gets splashed with water from a passing truck hitting a puddle. Or maybe it stopped raining 10 minutes ago and the road's drainage needs work and there is still plenty of water left.
As for a humidity sensor... Does your car have one? Mine doesn't. Again, it doesn't tell you what the road is like. That the atmosphere is humid gives you a small indication that something is up, but it is hardly conclusive.
I would also like to point out Volvo's trial project in Gothenburg. That is a summer-only project. They are not ready yet for a winter trial.
Somebody further up pointed out successful trials in Bavaria. I'd love to see some references. Particulary their winter driving experiences.
I don't see any solutions here, but given the number of companies dabbling in this area I guess I must be wrong. Where there is smoke there is fire, etc... It will be interesting to see what they can accomplish in ten year's time.
"A road that is just wet after a shower can be far slippier than the same road after a downpour that has washed all the oil off. "
In Pretoria, South Africa there is a long sunny winter with no rain. During that period the roads accumulate a layer of oil and rubber that has no particular effect on driving conditions. The first few days after the spring rains there are numerous car crashes until the now slippery coating is washed away.
I read the article, but I concede that I may have read too much into it. The creeping development of certain systems in cars troubles me. Developments like ABS and traction control are excellent improvements and have saved lives, for sure, but other systems (e.g. "lane awareness") creep into an area where the driver is partially abdicated from the need to drive, but is still required to be in control of the car. In short, driving becomes more like a video game experience IMHO with an increasing disconnect from the real world where hard, sharp and rough things can injure you, your passengers, and other road and pavement users.
I'm not advocating a return to 1920s seat-of-the-pants driving (when motor fatalities were higher, I know) but I believe driver awareness and defensive driving techniques are a better way to proceed.
Me too. You don't need lane awareness alerts. You don't need cameras to read the road speed signs.
You just need to open your eyes and look further than the end of your bonnet, something which it seems many drivers are spectacularly unable to do, so events occuring a hundred yards down the road appear as a complete surprise to them until they come into the 10 foot range.
I think it's because most people are thick as pigshit and frankly life in general is a series of completely unforeseen consequences as they lack the ability to view or plan further ahead.
Ah, yes, of course. Like the autonomous wipers and ice detectors and adaptive controls.
So an acoustic road surface profiler would be able to adjust the car's traction systems for wet surfaces, freshly tarred surfaces, loose chippings, toad crossings etc.
Now if only they'd find a system that tells the effing oversensitive ABS in my Prius that it's a pot hole causing the wheel rotation variation, not lack of traction. Losing the brakes on the approach to a fast roundabout which just happens to have a very rutted, potholed bit at a critical moment scares the bejesus out of me.
"It would be a bit rubbish if the car just decides it won't drive if the temperature goes below 3C."
Was it last winter when parts of the US were a bit colder than they had been in recent decades?
A little birdy tells me that what you speculate about did actually happen. Problems with a small number of low volume high value transport (often used by high net worth people), which refused to start because the temperature was implausibly cold (not 3C, well below that) according to the engine control unit, which diagnosed a temperature sensing problem and refused to even limp home.
>Problems with a small number of low volume high value transport (often used by high net worth people), which refused to start because the temperature was implausibly cold (not 3C, well below that) according to the engine control unit, which diagnosed a temperature sensing problem and refused to even limp home.
Link? Not that I don't believe you, but that seems, well, very unlikely. I'd like to see something that backs it up, before I believe that the software engineering teams behind the development of a high-value engine control system failed to anticipate that it might get cold outside.
"the software engineering teams behind the development of a high-value engine control system failed to anticipate that it might get cold outside."
Sorry, no link, people I know still work there.
Nobody failed to anticipate it might get cold. But like many others they failed to anticipate just *how* cold it might get last year in North America [1] and yes I should have been clearer about this.
The software team used what they thought were reasonably tight but reasonably plausible limits, as anyone might for an important parameter in a safety critical system. NB the software engineers wouldn't pick these numbers, they'd be given them by the engines/systems people.
Last year parts of the US and Canada went down to almost -40C, which has not been experienced in recent years. If the engines/systems people had said "what's the coldest we've ever seen, go 5 degrees colder for margin", in those conditions the control unit might well have said "all temperature sensors broken, please call for service".
I know I said "no link" but you might enjoy this one [2], not from America, not from 2014, but very related to what happens when the engine won't start because it's too cold out ('only' -35C in this case).
[1] https://en.wikipedia.org/wiki/Early_2014_North_American_cold_wave
[2] http://www.kaiserair.aero/press_4_23_05.html
Combined with a cloud sourced (ho ho ho) data #whenwillitstopraining, slurping of regional weather records and roadside atmospheric monitoring stations, geo-fencing around Manchester where it is always raining, slurping the fault reports from regional water supply companies, humidity sensors and rain sensors in the vehicle etc. I reckon they could push it up a bit further. Still doesn't account for Trevor and his regular Sunday morning car washing ritual, but hey, it's never perfect. Of course, if you've got these vehicles feeding BACK to a regional resource centre, then your accuracy would increase even further.
that roads are always dry and straight, and the sun is shining?
Despite all the progress being made (which is fantastic!), I think there's still a very long way to go until an autonomous car can outsmart a reasonably experienced driver. Tricky road conditions cause a lot of accidents. At least in Blighty you can encounter tricky road conditions even on dry, straight roads and excellent weather! (Some councils make sure of that by not spending any money where it's urgently needed.)
"Despite all the progress being made (which is fantastic!), I think there's still a very long way to go until an autonomous car can outsmart a reasonably experienced driver"
All they have to do is to slightly outsmart the average driver then as a whole the population will be slightly safer.
Why cannot the AI car do what the manually driven car does: have a look ahead at the road surface to determine whether it is wet, icy, oily, contains debris, has potholes, is flooded (especially useful this week in the UK), etc.
Listening is so present tense: so past best usefulness!
Best regards
The trick is like with us: accurately identifying hazardous conditions ahead of time when they're not obvious. For example, black ice which is difficult to see or trying to do it at night when ambient light is low. Accurate detection of road conditions ahead of time is still much a Work In Progress.
The question should not be "are the roads wet/icy", it should be "how much grip is available". It should be possible to determine this fairly accurately by continuously monitoring the cars actual movements relative to control unit's application of steering, brake and throttle.
The only thing that would catch it out is sudden changes in the level of grip available, which usually catches out human drivers too.
But monitoring the car's actual movements can be tricky, too, particularly if the car's not accelerating. It's the whole inertia thing, so it's just as hard gauging the car's actual-versus-desired movement as it is gauging the exterior conditions that could influence the car, but at least with the latter, you have plenty of options to use to figure things out.
And detecting sudden changes in those conditions that could affect grip is something they're trying to do to make the car superior to the human in terms of catching dangers ahead of time. Thing is, it's still a work in progress.
"it should be "how much grip is available"
it should be "how much grip is available ahead"
FTFY. The autonomous car needs to look ahead at the road conditions. It's a bit late if you're doing 70mph and don't react to the wet roads, flooding, black ice, whatever until the grip level changes.
Despite some of the incredible work being done in the field, I suspect we are still a long, long way from the truly autonomous car that can take on all the tasks of a human driver. Pootling around towns and cities as taxis etc might arrive sooner, maybe even between towns where there are motorways or equally decent roads, but coping with country driving in bad weather might well be a "challenge" for another some other decade in the future.
I have stated before that I am firmly of the view that "autonomous cars" won't be able to work properly without real time data being transmitted to them from the roadside for a variety of purposes. (OK; then they aren't strictly autonomous </pedant>)
If my belief is correct then "surface wet/icy/fuel spill" can be included in that data.
Having said that don't some cars automatically switch on the wipers if they detect rain; perhaps that's special rain that wets the glass but not the road, but it's going to be bloody scarey for the contents of the car if it keeps going in wet conditions without switching the wipers on, particularly if the car decides something "doesn't compute, Captain" and requires the driver to take over.
Brown trousers anyone?
"Having said that don't some cars automatically switch on the wipers if they detect rain; perhaps that's special rain that wets the glass but not the road, but it's going to be bloody scarey for the contents of the car if it keeps going in wet conditions without switching the wipers on"
Not forgetting the times you get to a stretch of road where it's recently been raining so you have wet conditions and no rain, so the automatic wipers don't trigger because there windscreen isn't wet.
...for actual intelligence and proper application of it. AVs offer a lot of potential good in society but it would be incomprehensible to allow the herds of get rich quick charlatans rush half baked products into the marketplace that will get people killed. Google has already demonstrated why poor programming can get people in AVs killed. AI may have a place but all AVs must have failsafe designs and safety modes to get a disabled AV off the roadway when it is hacked or has a systems failure.
I used to live in an apartment building on a busy street in the CA bay area. The cars were pretty noisy driving by, and yet it seemed they were particularly noisy on rainy days. It was the usual engine and tire+road noise with a high pitched "hissing" sound overlaid on it.
Perhaps they will design a sound sensor to detect this hiss and its intensity (more hiss = more moisture)...?
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
