I’d think that dual-frequency GPS and/or newer constellations like the EU’s Galileo system would be accurate enough for most any purpose- they stand to provide 30 cm accuracy instead of today’s 5 meter accuracy. In other words, 5 meters is good enough to usually determine which road you’re on, but 30 cm accuracy is good enough to tell you what lane you’re in.
I think self driving cars will be immensely popular when they get the kinks worked out, but what I haven’t been sold on is the necessity and inevitability of ride-sharing with them. I’m not at all convinced that suddenly the predominant mode is going to be ride-shared self driving cars. There are too many upsides to owning one’s own car- you can keep it as clean as you want, you can ensure it’s availble when YOU need it, you can trick it out how you like, you can temporarily store stuff in it, etc…
I do agree that the ability to say… drive me to work, drop me off, return home, pick up the wife and kids, take them to school, etc… and then come back and get me at the end of the day with one car will be convenient- no need for parking, no need for two cars, etc… One car can do it all, and if you can hook up a trailer, most people probably won’t ever need the robotic equivalent of a pickup either.
Once again, it is not you that these ride sharing services would be attractive to. It is people that have difficulties maintaining a car, or who don’t have a car at all, or who are not very good drivers and would rather not drive their car. Basically, the people who use Uber or Lyft or other livery services right now, along with those who, as the service becomes more robust, do not bother replacing their aging car, but rather just sign up to have a car come get them.
Public transit sucks, and even the best I’ve seen still is rather inconvenient. For someone like me in the burbs, public transit is essentially non-existent. I could take a bus to work, but the walk to and from the bus stops is nearly half the distance to work anyway. The bus also only goes the directions I need it to a couple times a day, severely limiting the hours that I would be able to work.
Your problem is that you make the unwarranted assumption that there would be something “sudden” about the transition. It will take quite some time, and I doubt that there will ever be 100% of cars are fleet cars. I’d be quite surprised it if were more than 90%, with 75% more around the point of “full saturation”.
I like that ability too, but without the need of paying to own and maintain a car.
One of the cool things about using a service would be that you can always get the car you need. If I just need to get to work, anything with a seat in it is fine. If I am going shopping, I may want a big trunk. If I am going furniture or building materials shopping, I may want an even bigger vehicle. If I am going camping, I probably want something that has a decent capacity, probably has 4wd, and maybe is also a bit of a “beater” car, so that no one cares too much if we get it muddy.
When I bought my last car, I had reservations about it, as it was the first car that I owned that wasn’t at least a hatchback. It has substantially less cargo capacity than any other car I’ve owned. This has actually been an issue a couple of times, when I wanted to take something somewhere, and found that I could not. However, the idea of buying a car with the requirements that it be able to take care of tasks that it may only have to do a few times over its lifetime is not an efficient way of allocating resources.
It’s visual cues primarily, as well as LIDAR cues. (from geometry). The algorithm takes into account all camera angles available, all LIDAR geometry landmarks, and at any given time there are hundreds to thousands of separate points being “triangulated” off at at a minimum. (this overwhelms the bad ones - mistaken points that actually belong somewhere else, dirty sensors, etc). With high accuracy sensors it is accurate to within inches and stays that way, being constantly updated with each motion. Particle filters is the localization algorithm, though since you also want to be constantly generating new waypoints - adding to the map as you go - you’d use SLAM.
So like I said, if you know exactly where you are, and you know which areas are to be cleared of snow, and you know the pavement depth/can determine it with load measurements (current to the servos) from the bulldozer blade scraping, and you know about how much snow is remaining, you can solve the problem.
Every action the machine takes updates the grid of discrete elements that are to be cleared of snow. Each action, prior to the action, the machine sends the situation to a simulation algorithm that predicts the outcome of the action. After the action, the simulation algorithm has a parameter weight update so that it’s predictions will more closely align with reality the next time.
So it’s not random, and even if snow does weird things, the algorithm can eventually determine the sequence of actions that will clear the snow with minimum cost.
But if it fails to work, the machine isn’t going to wander off, as long as it can see the snow is not satisfactory, it’s going to keep at it.
And when it returns to the docking bay and has a strong data connection, all these experiences can be sorted by order of surprise (greater error between predicted and actual) and sent to a cloud of servers, and pooled with every other bulldozing machine deployed using compatible software.
And the pooled experiences can be used to generate a weight update, which gets downloaded as a patch file, so the next time the bulldozer/snowplow fleet leaves for action, they all benefit from the collective experiences of all of them.
This is why even extremely tricky scenarios are solvable - very quickly these kinds of machines would have thousands of years of collective experience. If one of them found even by “accident” (when you don’t know what to do you take a random safe move) the way to solve the tricky situation on your driveway, the one clearing your snow can use that method right away.
Seriously? We can’t maintain the millions of miles of roads (& bridges) we have now, & you want to add something to them? Not only add something but add something powered? Are you proposing burying power to them? If so, have you seen the costs per mile for burying electrical service? This would probably be less as it’s lower power, but even at 1/10 of electric service cost you are in the neighborhood of $100m/mile. Even if they’re individually solar powered you’ll need a repair budget as thing like falling trees, freeze/thaw pothole cycles, & snowplowing will damage some percentage of them.
I just wanted to say this has been the most popular thread I’ve ever created, and I’ve been on the SDMB for almost 20 years. Lizard has still got his mojo!
Uh, yeah seriously. You could have things like cell towers every so many miles that pump out info on local road conditions that could be updated by local authorities. You could have materials embedded in curbs or the centre of roadways that could give information that wouldn’t require much power.
SamualA’s post upthread was a great overview of how these distributed machines can learn from one another and improve on a real-time basis. People seem to have lost track of that.
I’m not sure that we will require beacons to aid autonomous vehicles, but even if we did, they could be implemented over many years if they are found useful.
Responses like that it’s totally improbable for x reason were the same arguments that people against cars in 1918 would have posed. And because roads and highways were found to be useful, they were built and maintained. If it’s shown that autonomous vehicles save thousands of lives and billions of dollars, you can bet that a network of beacons will be built and maintained. However, some may disagree that these benefits will ever materialize. I guess we’ll see.
