Death by GPS
arstechnica.com
arstechnica.com
TIGER classes pretty much any rural "road" uniformly - class A41, if you're interested. That might be a paved two-lane road, it might be a forest track. Just as often, it's a drainage ditch or a non-existent path or other such nonsense. It's wholly unreliable.
You can see this in OpenStreetMap easily. Navigate to the backwoods and look at the road types. It's predominantly highway=residential, the road type that A41 was mapped to in the TIGER import. highway=residential is usually meant for city streets and suburban residential roads. Not this.
So: help! I've genuinely spent weeks on this but it's a massive job. Paved road? Reclassify it as highway=tertiary (wider with centreline) or highway=unclassified (narrower, no centreline). Gravel or dirt graded road? highway=unclassified, surface=gravel (or =dirt). Rough, not recommended for most cars? highway=track. Genuinely a paved residential road? Keep the highway tag as it is, but delete the tiger:reviewed=no tag. Nothing there at all? Just delete it.
(My particular interest in this is bicycle routing. Bike routing usually prefers the smallest paved roads in the grid. It breaks if those "paved roads" are impassable desert tracks in reality. I take a very conservative view for my site, http://cycle.travel/map, precisely because I'm anxious about situations such as that described in the Ars article.)
But I can see now why navigation breaks so horribly when all you have is an essentially uncategorised road network.
I'd love to help, but how? Is the only option for people to go driving all these "roads" in person to see which ones are actually roads, or is there another source of data that can be used to compare against?
I've found one "master" secret to not getting lost: Get rid of all that 3D garbage and put the map in overhead view, north up. It takes a tiny bit of practice getting used to the dynamic meaning of Left and Right but you build up a geospatial awareness and can usually tell when the GPS is getting stupid.
Totally agree.
I think the other thing that helps in an unfamiliar area is to pre-scout the route.. basically just look at the map around where you're going, while you're stopped. Honestly I don't always do this, but I should and it usually helps.
> can usually tell when the GPS is getting stupid.
The key to this is how you respond.
I used to just decide to go the way I think makes more sense, then often find out there was a reason: a one-way or closed street, no left-hand turns intersection, only one on-ramp to the highway (in the opposite direction I'm trying to go), construction, etc, and then I'd have to loop around.
Now I just become more vigilant. I always pay attention to the road names (and not just "take next left"), but I also really keep an eye out for extra signage (like signs showing the route to get to a highway) or closures.
One of the problems with current GPS tech and the maps is they're really damn good -- but still not perfect. The fact they work so well most of the time makes people complacent.
It took me quite a few months to relearn what you are practicing every day but now I'd say that this is a must on every curriculum. After I moved to a third world country (and did my re-learning) I understood that in case of a zombie-apocalypse (used as a place-holder for your "no tech works anymore"-scenario) people from less developed countries are so much more likely to cope than 99% of the 1st world population.
Google maps also doesn't seem to know whether roads are gravel or tarmac, and since they both have a 100 km/h speed limit, it will direct me down a shorter gravel road, that will actually take longer because I'm not suicidal to go 100 km/h down a windy gravel road.
Heh, you say that as if I didn't know! ;)
I'm sure you've noticed that in computer UIs, all the "3D" effects are based on the sun being at the top. Same with geography -- it's easier for me to visualize the terrain if the sun is at the top. That means South-up in the northern hemisphere, and North-up in the southern hemisphere.
Always fun to watch that West Wing clip.
You can see some photos from the trip here:
https://www.flickr.com/photos/defender90/albums/721576543876...
I also did a detailed write-up with photos:
http://www.expeditionportal.com/forum/threads/146545-The-Owy...
I assume we can make it work with computer vision for much the same reason humans don't routinely (though sometimes) drive off cliffs.
Here's another story: Driving across Guatemala on the Panamerican highway during hurricane Stan[0] (although we didn't know how devastating it was going to be at the time), we saw half of the highway that we had just passed collapse behind us down the side of mountain. We were one of the last 2 or 3 cars to drive west-bound that day. On that drive, we also saw full grown trees sliding down the mountainside, still rooted and standing vertically.
I eventually stumbled upon a sign, in German, that was big and red, and I knew I was on the border between Hungary and Austria. I wasn't sure what the sign said, and at the time I didn't realize the two countries had an open border. I looked around for police like cars, and then quickly crossed the border, eventually reaching my destination just inside Austria.
As it turns out the sign just said "no vehicles over two metric tons", and that somehow I had gotten of the highway which had a very nice border crossing.
But a few years earlier and that mistake would have gotten me arrested!
Interesting that the sign was in German while you were on the Hungarian side.
I find GPS to be invaluable in countries where I can't even decipher their alphabet (Sanskrit, Chinese) and like you, I end up trusting them a lot more than I probably should!
It was at the border to Austria, technically on the Austrian side. It was just standard street sign seen across Austria that says no heavy vehicles.
I'm waiting for the day that someone tells their iCar to "Drive me Home" and they wake up stranded 7,000 feet up "Mt. Home". It's not the GPS that is the problem. GPS does exactly what it was designed to do. The problem isn't even in the mapping software. The problem is that the human is too ready to give up decision-making to a machine. That should inform many autodrive enthusiasts who think humans cannot be trusted with their own navigation.
The implications of that statement, however, are not pleasant.
When the car primary control is the GPS that better change.
They were summarily purchased by Garmin and their devices are no longer for sale. The only remnant of their awesome Lane Assist Pro feature is in $30 GPS navigation apps.
It's really too bad.
But hey, if you're a clever person who knows how to code, please file a PR for this feature for Maps.me: https://github.com/mapsme/omim
It's:
1. aviate (physically fly the plane, or more fundamentally, fly the wing) 2. navigate (know where you're going/situational awareness) 3. communicate (tell whoever needs to know)
Route 1: Turn left here, right here, right here, left here, highway here, odd roads here. 30 minutes.
Route 2: Get back on highway, drive a while, Interstate, drive a while, odd roads. 33 minutes.
Thing is, I recognized each street in route 1: The Hood(s) with about 300 cops a shift who can't keep up with the amount of stuff that goes down. The other was a straight, safe shot. I thought, "Is Google really trying to send me through all the worst areas of my city... areas where you have to stop right next to potential carjackers and killers... just to save three minutes?"
Now I know why people coming through think my city is just a shithole. It's a mix of great, good, OK, bad, and ugly but Google keeps sending people through The Hood. No wonder they think it's nothing but poverty and crime. So, then I got the idea for a movie or TV show where there's a serial killer that, SAW-style, makes people kill themselves. By Google Maps instructions they follow blindly into the worst areas. You might see it on TV but you heard it here first.
Yeah, don't trust the GPS's until they know the difference between an armed robbery and an Interstate. At the least, they need to realize we won't take that kind of risk unless we're talking about saving 30min to hours, the interview is too important for Interstate traffic, or she was one bad-ass lady. Not over three minutes, though. ;)
C has very few intersections. But it has a 1 km stretch which takes 25 mins to cross if traffic is bad. Google maps always shows C to be the fastest. But local knowledge will tell you either A or B will be faster at peak traffic times.
sounds very similar to the beginning to Bonfire of the Vanities
I was driving on a reasonably remote highway in Victoria Australia. My car was running low on petrol. As I approached an exit there was a sign that said last service for 90km or something like this. I did some mental maths and decided I could make it another 90km so pushed on. The fuel warning light came on about half way through the 90km. When I reached the exit the pump was out of order and station closed (tiny rural station). Now I was forced either to turn back or push on to next station. I decided to push on and ended up running out of petrol I had to call roadside assistance. If I hadn't seen the sign saying the next service was 90 km away I would have pulled in then and there.
1. Toll roads - You can either choose to avoid, or choose to take. But I want to avoid toll roads when it will cost me > $x / saved hour. Or how about some optimization between fuel use and the value of my time? Paying $2 to save 30 minutes makes sense, buy not $15 to save 2 minutes. Where's the in-between setting/dial?
2. High-Occupancy Vehicle lanes. The highway may make a lot of sense with 2 people, but not while being a solo driver.
3. Traffic levels and when to wake up. I wish Waze could integrate with my alarm clock to wake me up extra early when major roads are closed, and I'd be majorly late for work without giving me an extra margin.
4. I'm not a trucker, nor an SUV driver, but something to account for high-height vehicles could be useful for many.
This exact issue causes a fair number of trucks to hit this one specific bridge [0] in Durham, NC. Blindly following a GPS meant for cars when you have a tall vehicle is a recipe for disaster.
[1] http://championtrucklines.com/latest-news/11foot8-bridge-cra...
[2] http://chapelboro.com/news/news-around-town/durham-can-opene...
[3] http://www.thetruckersreport.com/truckingindustryforum/threa...
> Vehicles will be limited by regulation and an overhead bar on entry to 2m (6.6ft, 79 inches). The space to the ceiling itself will provide 57cm (22 inches) for signage and other overhead gear.
I think the Montague Street Bridge and the Durham railway bridge claim so many trucks because they're just high enough (3m and 3.56m respectively) to trick inexperienced high vehicle drivers.
The 2m one would be obvious you can't make it through.
These days I have a dry erase board for directions in lieu of a gps and phone. I only need to look at it every now and then, and as long as my notes are correct, it's just about as easy to get there. If I get lost, pulling over and asking someone isn't that difficult to do, either. Not sure I'll go back. :)
The main thing this says for autonomous cars is: don't do that. If the routing algorithm has a bias towards picking highways and major roads and roads you've used before, this problem should virtually never come up.
Surprisingly, my neighbor revently described to me how she almost died in the same area by following GPS, just last winter, almost 9 years later. They blindly went up a mountain road in the snow as they were trying to drive to Portland from the Bay Area, and they went to the point where they could barely turn their car around. Had they attempted to stop later, they would have gotten stuck and it would have turned into a similar story.
Unfortunately, GPS needs a setting labelled "follow the safest, biggest roads".
I feel for James Kim, but why is that a reason for the road not to have been open? I make a hobby of exploring back roads. I sometimes use a GPS, but almost always for locating myself and getting a view of my surroundings, rarely for directions. I don't like to see bad things happen to people, but I'm sincerely puzzled when the reaction is to try and prevent them happening through prohibitions. When there are no roads left to close people will still get themselves into trouble by not observing enough, not thinking enough, or putting their faith in untrustworthy sources.
Perfect is the enemy of good. Sure, closing dangerous roads won't prevent all deaths, but if it has a good lives saved:cost ratio then it's worth doing.
The problem isn't so much GPS--although I imagine doing what a GPS "tells you to" can make the problem more severe. It's that in large areas of the Western US in particular there are a wide range of road conditions once you get off major paved roads.
ADD: As for closing off roads, there are a huge number of minor roads in the West that are more or less drivable depending upon weather, vehicle type, driver skill, and other factors. If you drive off paved roads in those areas, it's important to be careful and often to obtain more info than what's on the map.
Unfortunately you are vividly remembering some of the early speculation, and not what actually happened. James Kim died following a paper map - not a GPS or an online map, both of which were postulated but not true.
I turned around and several small dogs were in the road behind me. No big deal I thought, I walked past them as they barked at the intruder.
I continued in the most promising direction, which turned out to also be a dead end. Behind me the small dogs had been joined by an enormous fierce dog so angry his hair was standing on end. And the only path out was past him.
I stayed calm with my eyes on the exit, walking wide of dogs who were drooling and livid.
Better lucky than smart, I guess. I've never been cavalier about following Google maps since.
At the bottom of a fairly steep hill there was a huge unchained full-grown Rottweiler.
When we arrived at the flat area he attacked.
We had to hold our boards out to defend ourselves.
A women driver started honking and driving towards the dog causing him to run away.
We rode towards the main road, and I've never used Google maps in that tiny town again. Likely because I haven't been there since that day.
This seems to be an extension of what my family called "the Jeep principle". A jeep, and now a GPS, enables you to get stuck further from help.
Yes, it raises your performance. It doesn't magically turn icy, snowy Massachusetts roads into Arizona highway.
The result is that lots and lots of high-performance trucks end up in ditches and smashed into trees.
I saw a similar incident in Michigan. Big snow comes up suddenly, work lets out early. Someone with a new four-wheel drive vehicle becomes impatient waiting for clogged traffic to clear, and decides to go over a berm. Wasn't all that high of a berm. Vehicle sat there for three weeks. Kind as of a reminder to all.
One time, a driver chose to circle around after two blocks instead of turning directly right at an intersection, because Waze said it had a lot of traffic and therefore chose that route. If he had bothered to turn his head right, he'd see that the road was empty.
Some good friends moved to the North Shore near Boston about 10 years ago. We visit 3-4 times a year. They are big GPS people, and know nothing about navigating the region. If we go somewhere, can get us there more sensibly just about any time. It mind blowing to me. I always drive and/or walk around any new place I visit for more than a day. I like to know where I am.
They don't know, and in NYC in particular many are fresh off the boat, the yellow cab drivers were no different only slightly better connected to obtain a yellow cab.
Getting a route that follows many small roads usually only happens if you tell the GPS to find the shortest route (and really the shortest one), or by using its bicycle or pedestrian routing.
It also requires the user to ignore any signs along the way, putting way more trust in a bunch of males that's hard to keep updated properly than is healthy. (For topographical maps in Germany (I guess road maps have similar problems) it is sometimes normal for certain regions not to receive an update in a decade or two, thus sometimes showing very outdated features. OSM these days is luckily either very up to date or simply very sparse, depending on the region and how many people with local knowledge are contributing.)
Most digital US maps started with it and low traffic areaas are not a priority to update to something more sensible.
Furthermore, some visitors not familiar with such areas tend to assume that, if a road is on the map, that means it's fine to take in their passenger car.
I've been to the mall with my friends and their children, and there is no pre-planning, since they rely on the phones to meetup. However, when the battery dies, they're lost, or they rely on a friend's phone or charging station being available. There's an implicit reliance on redundancy, and blind faith somebody has a gadget.
I always taught my children to pre-plan. I have taken them around the world at young ages, and I think they have become better planners and thinkers when they are taught 'the old ways':)
I have traveled a lot, and I have lived in other countries, some remote places, and I have been lulled by technology too. Fortunately, not life-threatening situations, but I can easily see it happening.
A few days ago while driving in East Java, we followed the GPS into a sugar cane field. The road narrowed, was muddy, and started to diminish to almost just field. The road was on the GPS, and valid, but the van almost got stuck in the mud from the day's rain. This has happened to me about 2 to 3 times in 7 years, because smaller and smaller roads have made their way to the maps as mapping increases, and the GPS uses a traffic/length of route algorithm that sometimes has you turning off and back on to main roads to make time. I never would have followed that last turn-off several years ago.
I usually set the GPS to 'main roads and highways', but in some less-developed places this will put you at the mercy of weather, the vehicle's capabilities, and hoping that the routes have not changed since they were mapped. I also agree with other posters here in leaving it in overhead view with North as up. On a highway, no, but on twisty roads, yes. It is the only way you maintain orientation and a sense of the actual location you are in.
Source: was hit a few weeks ago by a car that did a sudden lane change with no signal right before an exit by a guy who was trying to follow his GPS...
http://greenash.net.au/thoughts/2010/08/boycott-gps/
Seriously, folks. Turn off the GPS, and start observing the world around you. And if you really do need route navigation, do it the old-fashioned way: study a map! (Printed or online, doesn't matter).
This line seems to suggest that the GPS wasn't ultimately at fault?
My GPS has options for avoiding tolls and the like, but I've not seen one giving the ability to define the capability of the car. That said, there is huge variability in quality of unpaved roads, especially when impacted by weather.
There's lots to the story that isnt the fault of Magellan or their data providers, starting with the fact that they ignored the GPS that told them to take US 93. And ID 51 isnt "a few miles to the west" of US 93, it's got to be a good hundred miles or more. It sounds like they had a paper map, which even the most undetailed of which would have told them the two roads are separated by wilderness areas, meaning there were probably no roads to get from one to the other. All of this before they got to the confusing crossroads. Had they stayed on 51 instead of trying to outsmart the GPS algorithm, they would have eventually hit I-84, arrived in Vegas a little late, and would laugh later about how they should listen to the GPS next time.
As another commenter pointed out, had they simply entered "destination='Las Vegas'" and blindly followed it (as the article ironically seems to argue against), they would have been fine.
EDIT: As for GPS options, the BMW-branded Garmin for my motorcycle has an "offroad" option. Because middle-aged men riding 600lb. "dirt bikes" like to pretend they're mighty adventurers. Better know what you're doing, because that thing will lead you down roads that aren't passable the majority of the year (or passable at all on an R1200GS, in some cases). What you propose isn't practical because one good rain and the road can go from "okay if you don't mind dings in your minivan" to "you brought a lifted Jeep, right?" very quickly. Or as a more recent example from my other comment in this thread, that road i bailed on this weekend needs snow tires and preferably 4wd right now due to snow. I'll bet in just a few weeks you can take a lowered Civic over the pass.
Observe your surroundings. If you begin to start to feel the slightest discomfort with the route you're on, consider turning back. Please reference "sunk cost fallacy". The time you lose backtracking is nothing in comparison to the time lost if you're dead. Or to be less dramatic, it might take a day for a tow truck to get to you, if they can get to you at all.
And always be prepared to turn back. Hell, I do it all the time; did it just this weekend, in fact. Wanted to go over Stampede Pass in the camper van with the dogs, camp, do some hiking, see the ghost town of Lester. Big plans! Warm spring, mid-May, should be good. Nope, turned back by snow. Might have been able to make it, but on those tires? Meh, go camp somewhere else, no fun to be had digging your van out of the snow ten feet at a time. More so on the bike: point A to point B on forest roads, map says it should go all the way through. Sometimes I'm turned back by gated roads. Sometimes, like the stories in this articles, the road turns to two dirt tracks separated by tall grass, to trail, to what once was a narrow trail and is now overgrown with saplings. And yet I still catch myself thinking, "park the bike, walk ahead a few hundred feet, see if it gets any better". Despite being somewhere no tow truck will come, with a 600lb. bike (heavy, for the uninitiated) ridden by a scrawny marathon runner who's going to have a shitload of fun picking said bike up when he drops it. Don't be a dumbass like me; turn around.
If you are well and truly lost, stay where you are. I'm not talking the "oh, we missed our turn" kind of lost. I'm talking, "without backtracking, I have no real idea where I am in relation to a major road or decent-sized town...or even a house" lost. If you find yourself hoping the road gets better, hoping this road will take you where you want to go, and hoping it will just all work out if you keep going, STOP. Maybe not until Search and Rescue finds you, but at least stop for an hour and get your bearings. Good bearings, so that you can point at a map (if you had one) and say with 90% confidence, "that's where we're at". Figure out your plan, 'cuz the current one ain't working. And do not leave your vehicle. Time and again I read these stories of "GPS took them down a snow-covered forest road, and they got stuck" that end with someone dying because they walked to get help. You're obviously already lost, you're not going to unlost by walking around ill-equipped.
Call in to work, call the friends you're supposed to meet, call off that client meeting, because you're going to be late or not make it at all. Seriously, just give up those thoughts right now. The time for "making up time" is over. You fucked up, and now you're late. Continuing on is not going to undo that fact. For one, you have insufficient information to conclude that you stand any chance of arriving on time. You don't even know where you are, how do you know how long it will take to arrive? So go ahead and backtrack, because you're plans are shot. Or go on over that mountain pass in mid-Novemember, the GPS says there's a road, and you might still make it. Just kidding, no you won't. You'll get stuck, try to walk to get help, and freeze to death on the way.
That's what my pamphlet might look like anyway. I often wonder how folks get themselves in these situations, and the best theory I have is that we have enough everyday convenience that we forget the parts of the U. S and Canada (let alone the world) that are still very remote and very dangerous for the unprepared. Many of us live where there's always a gas station, always a cell signal, always some landmark to give us guidance as to where we are. Road signs to tell us where to go, guardrails to keep us falling off a cliff. As the folks in this article found out, the western U. S. is a place where it is very easy to get yourself into trouble if you take all of those assumptions with you down a dirt road. Gas might be 100 miles away (whoops, that station closed ten years ago, hope you have enough to make it 50 more miles), you probably won't have a cell signal, and once you're off U. S. and state highways all bets are off as to road conditions or if the road even exists anymore.
After all this typing, it just occurred to me how some of these tragedies could be avoided, and in one sentence: if you're not a local, not experienced in off-pavement driving (especially in the mountains), and are not equipped to spend a few nights out there if you absolutely had to, then the instant the road turns to dirt, turn around; no exceptions. For the folks that drive off bridges, sorry, my only advice is to turn in your keys.
It probably wouldn't be very popular...
I regularly see tourists here in Australia do very odd things and sometimes I can't help but feel a little xenophobic.
Really, is that so hard? I see a lot of comments about what is going to happen when robotic cars do this. Well, every roboticist I know, and I know a big bunch of them, is well acquainted with the concept of having a cost-map overlay on the world map. And it is not only OK, but common, to have infinite cost assigned to many nodes -- "Don't go there robot, you will not be a happy robot."
So why does the GPS database have poorly assigned cost values on certain road segments? Perhaps because QA de-prioritizes little-traveled roads?
But more to the point -- I'll get out of roboticist mode and into seasoned backwoods traveler mode for a moment -- don't go where you are not prepared to go. Look at a paper map, ask a local at the gas station about conditions, and if the road ends up bruising your kidneys, turn back.
This would fix a lot of different types of problems. First: issues like the ones raised in the article. The killer "roads" have never had drivers successfully complete them, so they should have infinite cost.
It would handle construction. If the traffic in a route suddenly falls to zero, or suddenly goes from zero to lots, the cost functions can be updated to reroute.
It would handle dangerous neighborhoods. Route 31 was the most direct route between my first apartment and Pennsylvania. I once followed MapQuest directions on this route, and it turns out that it takes you through the part of Trenton that makes it one of the top-20 most dangerous cities in the nation (at the time). Nobody really drives this way, so it could route people via safer roads.
The problem here is that the companies that already have good databases are probably going to keep them secret, in case they figure out a way to turn that data into more ad revenue
There's a big distinction between a roboticist and someone programming GPS route determination. When you build a robot and throw some sensors at it and ask "what was your last position, your last turn, your last accelerometer and your last shift sensor reading," well, you get it. That kind of data is readily available. If you ask a lot of people for that information today, they uncheck the "share with ..." button before continuing. Or even more commonly, there's only one data point. If there's not enough data to do anything intelligent with it, then for practical purposes there's no data.
I'm not saying we should share more, but I'm trying to highlight a key difference and one that can make solving this problem more complicated than just throwing a relatively simple and common algorithm at it.
The rest of my comment is centered on what the real problem is: a lack of data. That will hurt any and every algorithm. If I have to make a decision/prediction and you give me a single or two pieces of input data, whatever process I employ is likely to return something poor.
Yeah we have algorithms for RL, but actually there are still many theoretical things we cannot prove and in practice do not work well. I am not intimately familiar, but one example is value iteration with a non-linear value function approximation.
I still maintain that RL is not well-understood, and I think evidence for that is that we don't see many examples of RL in our phones, cameras, search engines, etc. (which is not the case for supervised learning).
That the gps took them there sucks, but you need to be responsible for looking at the road and deciding if your vehicle can do it. Particularly in the mountains in winter. And be extra careful when you're approaching the half-way mark on gas, ie you can't get back to where you last saw a human. Going past that point means you're all in.
I drive to Tahoe all the time and the number of accidents you see because morons decide to zoom over mountain passes in their crappy civic with bald tires passing people in awd suvs with ski stickers on them (ie people who regularly drive in snow)...