Tripamang
Naxxramas 1.0 Raider
- 5,545
- 34,430
Autonomous Systems
- Thread starter Blakkheim
- Start date
-
Guest, it's time once again for the massively important and exciting FoH Asshat Tournament!
Go here and give us your nominations!Who's been the biggest Asshat in the last year? Give us your worst ones!
Gravel
Mr. Poopybutthole
- 40,359
- 133,750
Don't worry, the fleet of autonomous snow plows and salt trucks which are 500% more efficient than human drivers will have the roads cleared before you even knew it snowed the night before.
Tuco
I got Tuco'd!
- 47,951
- 82,702
Autonomous snow plows is a really, really hard problem. A good plow driver moves snow with a 3 ton truck with surgical precision, and can only do so because he's plowed that area many times and knows where all the curbs, bumps and gaps are buried beneath the snow. They have to operate with an immense amount of muscle memory, intuition and learned patterns that are really hard to build autonomy for.
As an aside, there's a snowplow competition for students, the robots aren't really well suited to it and it's been a disaster everytime I've looked:
You really need a treaded robot for it.
As an aside, there's a snowplow competition for students, the robots aren't really well suited to it and it's been a disaster everytime I've looked:
You really need a treaded robot for it.
Cad
scientia potentia est
- 25,867
- 50,843
Unless they drive the course when there's no snow and program the exact speeds/when to make the turns etc into the GPS? With a GPS fix plus a couple of location pings to known objects it should be able to get its position to inches and be able to drive a fixed course almost without external input.
Cad
scientia potentia est
- 25,867
- 50,843
Yea, but it should know exactly where all the curbs bumps and gaps are because it has electronic memory of the road without snow, wouldn't it? Once it has an exact position and a map of obstacles it can adjust to the snow situation and miss the curbs pretty precisely can't it?
Tuco
I got Tuco'd!
- 47,951
- 82,702
Yes, you'd absolutely need a fairly accurate world model generated from pre-snow data. That's what Ford is doing to do accurate localization (figuring out where the vehicle is) after heavy snowfall when there are no lane markings.
Once you have that, you've got a optimization problem of trying to bring the current world to the state of the previous world by removal of snow through the plow. Lidar actually works pretty well on snowfall, which is good because color cameras are shit for undisturbed snow. So you can get a good sense of what the current state of the environment is from your perspective. You then plan out an optimal set of passes to repeatedly drive at the snow until the snow is cleared.
There are a number of non-trivial bits in that, including building an accurate world model before the snowfall, controlling the vehicle in dicey conditions, balancing between being too aggressive/clean and too cautious/dirty (where you leave too much snow). All while trying not to destroy the landscape or run into oncoming traffic.
It's a problem that will be solved, it's just harder than people think.
Once you have that, you've got a optimization problem of trying to bring the current world to the state of the previous world by removal of snow through the plow. Lidar actually works pretty well on snowfall, which is good because color cameras are shit for undisturbed snow. So you can get a good sense of what the current state of the environment is from your perspective. You then plan out an optimal set of passes to repeatedly drive at the snow until the snow is cleared.
There are a number of non-trivial bits in that, including building an accurate world model before the snowfall, controlling the vehicle in dicey conditions, balancing between being too aggressive/clean and too cautious/dirty (where you leave too much snow). All while trying not to destroy the landscape or run into oncoming traffic.
It's a problem that will be solved, it's just harder than people think.
TrollfaceDeux
Pronouns: zie/zhem/zer
- 19,577
- 3,743
the cost just simply doesn't justify the reward....shit like this is hard to measure its investment value over a period of 10 years, especially with global warming and unpredictable weathers..
TrollfaceDeux
Pronouns: zie/zhem/zer
- 19,577
- 3,743
Tripamang
Naxxramas 1.0 Raider
- 5,545
- 34,430
Another thing to consider is that GPS isn't very accurate, it's more or less a 3 meter bubble with current tech. I'm not sure what the new constellation of GPS satellites can achieve but you need really accurate clocks to get very accurate positions, because it's using the time it takes the signal to get you from multiple fixed locations to determine your location. I know some new atomic clocks are good enough to achieve sub millimeter accuracy, but it could be years before they get up into orbit.
You can use ground based signals to help improve accuracy but I'm not sure they work consistently enough to be reliable. Either way we'll need better tech before we can accurately pinpoint locations enough to make something like a snow plow ride right along the edge of the road based on maps when it can't see them. Tuco any idea if there is a sensing technology that will work through snow/ice but not through asphalt/concrete?
You can use ground based signals to help improve accuracy but I'm not sure they work consistently enough to be reliable. Either way we'll need better tech before we can accurately pinpoint locations enough to make something like a snow plow ride right along the edge of the road based on maps when it can't see them. Tuco any idea if there is a sensing technology that will work through snow/ice but not through asphalt/concrete?
Tuco
I got Tuco'd!
- 47,951
- 82,702
Yes. Certain types of Ground Penetrating Radar can find ground beneath snow.
MIT Lincoln Laboratory: News: Lincoln Laboratory demonstrates highly accurate vehicle localization under adverse weather conditions
Ground Penetrating Radar Equipment for Ice and Snow Surveys - GSSI
Ice and Snow - Sensors & Software
This is actually one of the things I can't talk about in detail, but I can say that it has its limitations, but I think will be pretty important for solving a lot of snow-related issues.
MIT Lincoln Laboratory: News: Lincoln Laboratory demonstrates highly accurate vehicle localization under adverse weather conditions
Ground Penetrating Radar Equipment for Ice and Snow Surveys - GSSI
Ice and Snow - Sensors & Software
This is actually one of the things I can't talk about in detail, but I can say that it has its limitations, but I think will be pretty important for solving a lot of snow-related issues.
Tuco
I got Tuco'd!
- 47,951
- 82,702
On and re: GPS, with current tech you can expect that 3 meter bubble, but that can be heavily reduced with different methods of what's called differential GPS correction. Most error is atmospheric, so if you know what the current error is in a given location you can offset based on that. Depending on your method you can reduce the error to be a few centimeters. This doesn't hold true in city where multi-path problems can happen, or heavy forest cover.
Trimble - GPS Tutorial - How Differential GPS works
But your overall point is correct, GPS will be accurate enough to have pre-recorded GPS paths to follow. You'll always need onboard perception to figure out what's around you and where you are.
Trimble - GPS Tutorial - How Differential GPS works
But your overall point is correct, GPS will be accurate enough to have pre-recorded GPS paths to follow. You'll always need onboard perception to figure out what's around you and where you are.
chaos
Buzzfeed Editor
- 17,324
- 4,839
I took my kids to a landfill for their girl scouts shit with their troop, it's the largest landfill in NOVA, was super interesting. But it got me thinking about the trucks they use and how they solve automating them. Because right now, they're paying shifts of guys to drive back and forth over piles of trash in order to compress it down, using these giant unwieldy trucks with these crazy monster truck tires made of steel. But the terrain is constantly changing, and there are TONS of pitfalls and obstacles that must be avoided while performing what is essentially a pretty simple job. I'm sure it's probably a simpler problem than I am imagining. At the time, it seemed kind of a comparable problem to the snow thing. When you can't rely on visual or GPS you get into some voodoo shit.
Tuco
I got Tuco'd!
- 47,951
- 82,702
landfill automation would be goofy. I don't know anything about landfills, but I imagine you'd have to carefully model deformable terrain while driving on it. No idea how much a landfill operator intuitively understands that by knowing generally what's in the garbage he's driving over.
BrutulTM
Good, bad, I'm the guy with the gun.
- 14,762
- 2,644
Trying to see the edge of a road with fresh snow on it is brutal, especially if it's overcast. All you see is white road and white ditch and white on the other side of the ditch. I don't know how something like lidar would do with that but human eyes are pretty shitty for that use in my experience. Here's a pic of a day when I had to drive 12 miles on a road no one had been on the day after an 8" snowfall. This was about 8 white-knuckle miles into the drive.... (spoilered because giant pic)
Sentagur
Low and to the left
- 3,825
- 7,937
I think you found the ditch.
Things can be crazy even in a city after a few inches of snow, when the road is split between the "I have snow tires so i can drive as fast as in the summer" and the " 5mph is perfectly reasonable speed" people.
Human mind is very adept at extrapolating stuff from an imperfect set of information as well as suppressing the unimportant stuff. That is why some of us can drive in snowy conditions. I dont know how long the machines will need before they can do that.
