Korean Competition Shows Weather Still a Challenge for Autonomous Cars

http://spectrum.ieee.org/cars-that-think/transportation/advanced-cars/japan-competition-shows-weather-still-a-challenge-for-autonomous-cars

by markmassie • 12 years ago

61 82 12 years ago

spectrum.ieee.org

therobot243• 12 years ago

a guy in our group did clear path detection for vehicles using only vision - his biggest hurdles were shadows and rain

his thesis was awesome though, he created shadow and rain drop removal while also mitigating the effects of 'smeared headlights' from a wet road at night

yason• 12 years ago

Then the same software could produce an enhanced image of the road to a human driver, too, right?

TheLoneWolfling5• 12 years ago

Probably.

I always thought that an interesting intermediate for cars would be to have a HUD displaying wireframes around detected objects. The motivation being that it can help, but doesn't have nearly the potential for harm as a full self-driving vehicle.

Of course, then you end up with the same problem that we have with aircraft - people become reliant on fallible assists.

therobot243• 12 years ago

i suppose, but it was just PhD research, nothing enterprise

agumonkey2• 12 years ago

I take it that it's not openly published, but just in case, feel free to drop the url or at least this guy's name.

therobot243• 12 years ago

it was funded by GM so i'm sure they own the patents, but the phd and masters thesis documents are published somewhere (at the very least the CMU library - Tony Qi Wu, he graduated in 2012-2013)

TheLoneWolfling5• 12 years ago

I would like to hear more.

csense• 12 years ago

I'd imagine it's more challenging when the road is covered in snow and the snowplows haven't gotten to it yet.

infogulch• 12 years ago

That's hard for humans to navigate, let alone self-driving cars. Besides driving very slowly, drivers will often estimate the lane position based on the shape of the snow and the quantity/size of lanes and shoulder they expect. If the snow is not too deep, rumble strips [1] can be very helpful.

[1] https://en.wikipedia.org/wiki/Rumble_strip

TheLoneWolfling5• 12 years ago

In our area, there are a couple roads where they have ~2.5m thin poles with reflectors on top sticking out of the ground at the edge of the road shoulders.

Very helpful, especially for snowplough drivers.

maxerickson4• 12 years ago

There's research into providing enhanced information to snowplow drivers and also automating snowplows:

https://www.fhwa.dot.gov/publications/publicroads/01janfeb/s...

But you can stick a lot more sensor gear on a snowplow than a cheap sedan.

rdtsc• 12 years ago

One of the worst combination of weather related road conditions I've been in -- fog and ice, at night, in the mountains.

A self driving car wouldn't have to work very hard here to do better than me.

* Check for weather conditions at altitude (not just start and destination points).

* Refuse to leave the house using that route if weather conditions are bad.

Oh well, lesson learned. On that topic does anyone know of maps software that does that for you -- i.e. check weather not just traffic?

snowwrestler5• 12 years ago

Ha, yes, maybe self-driving cars would be safer if they also functioned as safety monitors.

"Take me to the ski resort, HAL."

"I'm sorry Dave, I can't allow you to do that. It's snowing along the way."

"Yeah, I know--that's the idea. I want to make some fresh tracks."

"Driving through the snow would be more dangerous than driving on dry pavement. In addition, I calculate that skiing is 4.76 times more likely to result in injury than reading a book. Please return to your couch; I will download a new book to your Kindle."

Incidentally, this was a major theme of Asimov's Robots/Foundation series--that intelligent robots programmed to prevent humans from coming to harm would become nannies of humanity in order to execute that command.

qwerta• 12 years ago

I would be more worried about insurance companies. They will get way more data and ways to sneak out of paying damages.

knowtheory• 12 years ago

This would have been a much more interesting origin story for the Matrix, esp to play off of what machines would optimize for regarding human longevity.

csours• 12 years ago

I think I got bored of Foundation before it got to the good stuff. Perhaps I should try again now that I'm older.

ableal• 12 years ago

intelligent robots programmed to prevent humans from coming to harm would become nannies of humanity

Jack Williamson went to town on that. "With Folded Hands", a story later expanded into novel(s). Before Asimov, if I'm not mistaken.

deelowe• 12 years ago

The problem with the mountains though is that these conditions can manifest themselves in a matter of minutes. Last time I drove through the mountains, it was sunny on the way up, but as we were leaving it was already overcast and within 10 minutes we were in thick fog and a near blizzard. Also, roads completely iced over 1/2 way through the 30 minute drive back down the mountain. I'm not sure any AI would have done better than I did in this case.

maaaats3• 12 years ago

This is very true. I grew up at a "mountain-pass", e.g. the last civilization before 100km of mountains. Even if the sun shines and the forecasts say the weather will stay nice, one should bring blankets, hot drinks etc., as the weather can change in a couple of minutes and then you're stuck on the mountain.

tokenadult• 12 years ago

On that topic does anyone know of maps software that does that for you -- i.e. check weather not just traffic?

When I use Google Maps to check routes, on my desktop or on my Nexus phone, it provides National Weather Service weather advisories along with my route directions. (I live in Minnesota, where those are presumably more frequent than they are in most parts of California.)

abdias• 12 years ago

What will snow do?...

snowwrestler5• 12 years ago

On HN I frequently see comments to the effect that self-driving cars will be commercially available in a decade or two, and that they will be safer than human-driven cars. Personally, I question whether either is true, and I strongly doubt that both can be true at once.

Humans have a lot of shortcomings--emotions, sleepiness, distraction, the effects of drugs and alcohol--but humans have some amazing strengths too. We've got cognitive abilities in recognizing patterns and making decisions that were honed by millions of years of evolution, and directly trained by, on average, decades of actually driving cars. I think it's easy for technology-minded folks to overestimate the impact of the weaknesses, and underestimate the deep complexity of the strengths.

We barely have fully-autonomous self-flying planes today, and I don't know of any research showing that they are safer than human-piloted planes. And flying is a far less complicated domain to automate than driving.

rimantas2• 12 years ago

Those coments are mostly extrapolating results gotten in a place with a nice weather, and every inch of the road mapped.

dan00• 12 years ago

I think it's easy for technology-minded folks to overestimate the impact of the weaknesses, and underestimate the deep complexity of the strengths.

I'm also really astounded when I'm reading the answers about self driving cars here on HN.

If you have ever worked on algorithms that are mostly based on heuristics, then you know all of these ugly edge cases, and this uncertain feeling, that you most likely don't have covered all edge cases.

jusben13694• 12 years ago

I think the risk here is that you're aiming the bar for 100% perfection (very high) when the accurate bar to compare it against is human drivers (far from 100%) So self driving cars just need to be better than your average human driver. That's a much lower bar from the parking lot fender bender to the prevalence of speed in fatalities etc.

EDIT: Changed to "100% perfection" from 0% to remove any confusion.

jfoutz2• 12 years ago

There are some cases where we want amazing drivers, for example ambulances. But for an ordinary day to day commute, people shouldn't have to be at the top of their game. I think your point boils down to variance. Human drivers have a high variance, we all can be very good or very bad somewhat randomly.

Your point about research is really interesting. I think a large majority of cars actually following the speed limit, looking before changing lanes, and using turn signals will have a positive effect on overall safety, even if they aren't anywhere near as good at dealing with extraordinary situations. It would be nice to know what % of "good drivers" is required to minimize those extraordinary situations, if any at all.

We've all seen that one car weaving in and out of traffic. Perhaps they do indeed make us safer because they force us to pay attention. But they do seem to create a whole bunch of extraordinary situations for dozens or hundreds of drivers to deal with. It only takes the one weak driver (i'm assuming weaker than self driving here) to panic and then the freeway is a parking lot for an hour. That research would be interesting.

mavhc• 12 years ago

Ambulances may kill more people than they save by driving at high speed http://usatoday30.usatoday.com/news/nation/2002/03/21/usat-a...

0x5f3759df-i• 12 years ago

No Humans really aren't that good at driving cars, 33,561 people were killed just in 2012 in the US[1].

[1] https://en.wikipedia.org/wiki/List_of_motor_vehicle_deaths_i...

TheLoneWolfling5• 12 years ago

...Which, as this [1] commenter points out, works out to a per-mile success rate of 99.999925%.

(Or, alternatively, works out to on average one injury or death per 1.33 million miles traveled.)

Or, to put that number in perspective, there are 435,000 deaths per year caused by tobacco in the US - 13 times that number. [2]

[1] https://news.ycombinator.com/item?id=8700342

[2] https://en.wikipedia.org/wiki/List_of_preventable_causes_of_...

josefresco2• 12 years ago

Just drove about 8 hours and experienced heavy fog along the way. Visibility was reduced to maybe a hundred yards. The entire time I wasn't thinking "self-driving cars would fail here" but rather that a self driving car with it's multitude of sensors would probably do much better considering what it could "see" and also the reduction in "eye fatigue" as driving in heavy fog can wreak havoc on your eyes and perception of movement (not a problem for a computer).

rayiner2• 12 years ago

Funny. I just had my Uber driver take me on a 50 minute trip from the Kennedy Center to Union Station, a trip that usually takes 15 minutes, because he blindly followed the GPS straight through downtown. The entire time I was thinking: "it's 2014 and we still haven't gotten basic route-finding right; self-driving cars are a long way off."

TwiztidK• 12 years ago

I think a network of self-driving cars might actually help with that. All the cars can communicate where they are going and could agree to take paths that reduce traffic and get everyone to their destination faster.

EDIT: Also, these features should be coming soon thanks to devices like Automatic and apps like Waze.

afterburner• 12 years ago

Which app was he using?

josefresco2• 12 years ago

Downtown city driving is certainly a high bar. Even veteran taxi drivers make mistakes or bad decisions when choosing a route. But certainly it seems like a a solvable problem, given that most apps/GPS units can track, and aggregate data to find the optimal route.

I was impressed on my trip that Google Maps actually recommended alternate routes (on the fly), with time estimates based on traffic. I was able to cut off a dozen or so minutes based on it's suggestions.

Also very impressed with the foresight that an app like Waze offers. During my 8 hour trip I navigated for 4-6 hours and was easily able to warn my driver about hazards (cops, wind, accidents, construction). The kicker is that Waze is "powered" by human reporting - which suggests the ideal solution is a hybrid of sorts.

AshFurrow• 12 years ago

To be fair, wet roads confuse human-driven cars, too.

bane• 12 years ago

One of the interesting things we're finding as we start to coax machines to do human jobs is that humans tend to not accept a machine unless it's much better at the job than a human. For example, on many computer vision or NLP tasks, the current state of the art is about as good as most humans.

The problem is that when they're wrong, they're wrong in ways that humans wouldn't be, and the wrongness appears to be "broken" to us and we overweigh these incorrect responses because they're both wrong and not human in their wrongness.

rimantas2• 12 years ago

Are computers able to recognize a cat in a picture and tell it apart from dog?

laurencerowe• 12 years ago

To be fair, wet roads confuse human-driven cars, too.

Especially in California.

abc_lisper• 12 years ago

Won't infrared sensors work well here?

waps2• 12 years ago

Water in the air has a large heatmass. So rain leaves a "shadow" in the infrared. Worse: it can make things invisible in infrared sometimes too.

Rain/fog distorts an image because there's droplets in the air that function as random lenses. This will affect infrared images just like it affects visible spectrum images. Only things with wavelengths larger than the droplets will be able to see through fog.

zmanian• 12 years ago

I've seen the Google self driving car on the roads on Mountain View and Los Altos since the current storms started. Clearly they are taking advantage of the weather conditions for data.

lallysingh• 12 years ago

This article mentioned cameras, not the laser scanners used by the Google cars (http://www.theguardian.com/technology/2014/may/28/google-sel...)

waps2• 12 years ago

Water is a lens. It doesn't just screw with cameras but with lasers too.

aidenn0• 12 years ago

Google uses both.

maaaats3• 12 years ago

Which companies had cars here? I've been impressed with some European car makers and their self-driving cars, especially Volvo. They seem to make cars that handles true driving conditions, while American makers have idealized conditions, especially the Google car.

maxxxxx• 12 years ago

Once you live in California you start to forget that there are weather conditions other than sunshine and warm. It happened to me!

hamitron• 12 years ago

Seriously, as a californian seeing rain for the first time in a year I doubted the sun would ever return

001sky• 12 years ago

The lack of skill, experience, and equipment conspire to make it much more dangerous than the objective risk. That's why there are so many hydro-planing wrecks out after heavy rain. And why californians need chains to drive in conditions that are normal on the east coast.

maaaats3• 12 years ago

Then you should try northern Norway. Months without sun, then midnight sun the whole summer! It's wicked.

jusben13694• 12 years ago

And my earliest iPhone's were pretty poor at getting a GPS signal and drained my battery like a madman. Within 3 years that had all changed dramatically. So I'm not sure how much weight I place on current results in an area advancing so rapidly.

icebraining• 12 years ago

Yes, but your iPhone wasn't exactly the first mobile device with GPS; having a smartphone with GPS, Wifi, etc and decent battery life was a solved problem, even if Apple couldn't do it at the time.

jusben13694• 12 years ago

I guess my point is very smart technology companies focused on solving problems usually experience rapid improvement in the first few years in key areas of performance. So I expect with all of the attention/investment and competition around self driving cars the advancements over the next few years will be dramatic.

ChuckMcM2• 12 years ago

I was thinking about this when driving in the recent rain here in California. A lot of data 'goes away' and sometimes I relied on memory of what was there rather than actively see what was there to navigate. I haven't seen any discussion on whether or not auto-cars would have some sort of equivalent to a 'muscle memory' of where they have been driving to help them over the rough patches.

In the test case, if the car recognized "oh this is the course from yesterday, even though it doesn't quite look like it, its in the same GPS coordinates" and then try to match larger easier to spot landmarks with its previous experience. That would certainly help it out.

TheLoneWolfling5• 12 years ago

And then you get a case where something is added or removed unexpectedly and the detection is marginal, and the car makes the wrong decision because "it was/wasn't there last time".

glenra• 12 years ago

Sure, but the goal isn't "perfect", it's merely to do better than human drivers. Human drivers make that mistake constantly, so the case you're describing is one where the cars will do MUCH BETTER than people because cars are better at sharing information over time. So we go from a world in which every driver independently has to update their own personal worldview and is thus individually surprised by the addition of a new sign to one in which a few cars are surprised and have to slow down and update but within a few days every subsequent car through the intersection automatically knows the new data.

danielweber4• 12 years ago

"The wrong decision" is very unlikely to be "drive into a ditch."

If a stop sign disappears, the car won't figure "gee, I guess I can plow on through."

If a stop sign shows up, the car is very likely to still notice it, and if not it still will be paying attention to what other cars are doing. It won't plow into the car in front of it. It will still notice what cars on side streets are doing.

danielweber4• 12 years ago

Google's strategy is to map the hell out of everything and rely on multiple inputs. It's not a dead-reckoning every single time. They know where lane markers and traffic lights are supposed to be.

brandon272• 12 years ago

Living in a cold, snowy climate, it seems like self driving cars will never be possible here.

Winter snow removal equipment scrapes the lines off the roads requiring re-application every summer in my area, and a common sight in the cities is the "re-positioning" of lanes during the winter. Drivers carve out their own lanes in the snow, so a 4 lane road might have only 3 lanes in the winter. I find this challenging to negotiate in the winter as no one seems sure whether they should follow the carved out path or be in the actual lane, assuming they know where it is under the packed snow. How could a self-driving car deal with that?

Not to mention inconsistent, icy road surfaces with unpredictable traction and the springtime muck of snow/gravel/salt that makes keeping the windshield clean, much less exterior sensors and cameras.

maxerickson4• 12 years ago

Electronic stability control is already required equipment on new vehicles in the U.S. (and around the world: http://en.wikipedia.org/wiki/Electronic_stability_control#Re...), because computers using sensors are better at measuring traction than humans are at estimating it.

nogridbag2• 12 years ago

It's also the first thing I disable when driving in the snow in a RWD car, since the last thing I want when trying to make it up a snowy mountain road is for the computer to cut all power to the driving wheels.

robomartin• 12 years ago

One of the most important things I learned very early in my CS schooling was that data (or problem) representation is, by far, the most important step to solving a problem computationally. Your computing platform, compiler, language, editor, memory size, speed, etc. cannot compensate for bad representation.

While I recognize the amount of work and smarts that has gone into self driving cars it is my humble opinion that the typical approach being taken is fundamentally flawed.

The right approach, again, in my humble opinion, is to change the representation of the problem: Roads have to be smart. Cars smarter, yes, but roads need to be modified for robots. Change the problem you are trying to solve and the solution might just become massively simpler.

Embedding technology into roads in order to facilitate robotic navigation would solve tons of problems. Private enterprise can take care of parking lots and private garages, and they will, when enough self-driving cars are around to make it worth their while to install the technology.

When my kids were younger they had huge problems opening doors around the house that were equipped with spherical handles. If their hands were wet after washing them, even worst. So, we changed the problem: We installed lever type door handles and even the youngest kid with hands full of soap could open the door now.

But, hey, what do I know.

gwern• 12 years ago

Embedding technology into roads in order to facilitate robotic navigation would solve tons of problems. Private enterprise can take care of parking lots and private garages, and they will, when enough self-driving cars are around to make it worth their while to install the technology.

It is easier to have some brainiacs work very hard on making very smart cars than it is to change local government and infrastructure everywhere.

If you ever make a plan which begins, 'let's convince every county, state, and federal agency in the USA to change their road construction methods to add this novel expensive feature...' you can probably drop it then and there as a stinker.

Korean Competition Shows Weather Still a Challenge for Autonomous Cars

Related Stories