Information Needed for a Ground Vehicle

[0111narwhalz]

I've been wondering this for a while: What information do you need to pilot a ground vehicle? (When I say ground vehicle, I mean something serious. Not just an automobile. More like something you'd see on a frontier world.) Obviously you need navigation, cameras in several directions, and some way to talk to your autopilot. Speed, grade, and perhaps...altitude? could be put into a single screen. Communications could take its own screen as well. Are there any other must-have things?

In other words, how do you configure your RasterProp MFD's?

EDIT: Refining the request:

1: The autopilot is a glorified cruise control, and little more. It'll keep you going in the same direction, at (largely) the same speed, but not much more than that. Think MechJeb's rover autopilot.

2: The terrain information is not immediately evident. Think topographical map, you have to pay extra to get higher quality.

3: The terrain tends to be craggy, or full of gullies, ruts, and ridges. You can't really trust an AI to navigate this, like you can't trust an AI to rock-crawl.

4: Although you can go outside for a bit (hull breaches won't kill you), you really don't want to. Think high desert or Antarctica; you want that climate-controlled cabin.

5: Weather could be a concern. This includes solar weather, as well as simple rain and stuff. Might wanna keep an eye on that.

6: The vessel won't really be doing much SCIENCE! on the way; it's more of a freight sort of thing.

Edited November 9, 2015 by 0111narwhalz
Refined Thingy

[SomeGuy12]

Depends on the game, the physics model, and so on.

Are you talking about in a realistic simulation style game or in real life?

Because if you're talking about a game, you ultimately need a single piece of information : whether a given piece of terrain (you can divide it into 2d squares) has too much steepness/large elevation changes for the drive system of the vehicle in question to traverse it.

You subdivide the 2d squares into squares that are about 1/4 to 1/16 the total "footprint" of your vehicle. Your drive system for the autopilot either :

1. Proceeds directly in the direction of the waypoint until it encounters a set of squares the vehicle cannot traverse

2. Uses A* to compute a path to the destination from start to finish and follows that path

#2 is better for AI controlled vehicles, I think, because I find it immersion breaking to have AI controlled vehicles in a game get "stuck". I think it is a superior solution to have the AI controlled vehicles never even move if there's no valid path to their target than have them 'running against a wall" fruitlessly trying to get there.

#1, you could do limited A*. You have a certain "bubble" around the vehicle covered by sensors. Once you encounter non traversable terrain, you try to find a path to the far edge of the bubble in the direction of your desired waypoint. If no path exists (say the rover encounters a mountain range blocking the way, the autopilot shuts down the vehicle and sounds an alarm)

[0111narwhalz]

I think you might be missing the point a bit. I'm not asking what you need to give an AI; I want what you'd configure your truck's cabin monitors for. Assume your "autopilot" is effectively a glorified cruise control; it doesn't really pathfind or anything. Like you said, a pathfinding AI would simply need a topo-map. So, because that's boring, let's not use one of those.

And let's call it real life for sake of argument.

[Rath]

Piloting, or all the other stuff. If you have a ground sampler you would need controls/diagnostics on one panel, data on another.

- - - Updated - - -

what about storm/radiation/environment data.

[SomeGuy12]

I think you might be missing the point a bit. I'm not asking what you need to give an AI; I want what you'd configure your truck's cabin monitors for. Assume your "autopilot" is effectively a glorified cruise control; it doesn't really pathfind or anything. Like you said, a pathfinding AI would simply need a topo-map. So, because that's boring, let's not use one of those.

And let's call it real life for sake of argument.

A reasonable setup would be similar to the data visualization displays the Carnegie Mellon team won when they won the Darpa challenge...for a rough road, autonomous vehicle, very similar to what you are asking for.

I can't find an image handy, but essentially it's just a simple gradient map. One color - we can use white - represents "goodness." It's the combination of a bunch of algorithms working together to guess at how "drivable" the terrain in front of the vehicle is. You see black areas representing "dangerous" terrain" and white areas representing drivable terrain, and the autopilot tries to find a route through the terrain that will stick to areas that are "good enough".

The sensor data is combination of lidar and cameras - on the CM vehicle, lidar was only short range, so the vehicle would guess at what terrain features were at a distance, optically. When the vehicle gets close to that same features, it scans it with lidar, and estimates, based on the real world shape of the object, how "drivable" it is. (yes I think this mean the vehicle would be equally afraid of tumbleweeds as boulders because both have a solid lidar return, but only one actually impedes the progress of the vehicle)

It stores in memory information about the object it saw, and gradually builds up a database, so as it sees new objects, it learns - in realtime - what they are shaped like and can now recognize them from a great distance.

For a futurisitic vehicle you probably could make the lidar a lot brighter and thus boost the scan range. I don't know what you can do to reduce the "tumbleweed" problem - you don't want your autonomous vehicle to be afraid of foil wrappers and plastic bags and cardboard cutouts but I know of no single sensor systems that can recognize all of these things reliably.

One more thing - any reasonably competent engineer for a system like this is going to demand overall mapping data. Having your vehicle drive around blind, only able to see what is immediately around it, is a terrible way to make sure you even reach your destination. On a planet with enough atmosphere you'd launch a propeller driven flying drone that would ascend to high altitudes and map enough of the terrain to find a route to the destination, completing the map before you even depart. If there's not enough atmosphere, you'd use a satellite map.

So if you had 2 basic autopilot screens, one would be from the perspective of the front of the vehicle, facing towards the direction of travel, and it would show the autopilot's estimation of the safest route to travel between terrain obstacles. The other screen would be from above, and would have a line on it showing the planned route from your present location to the destination, and the remaining distance and ETA.

Edited November 9, 2015 by SomeGuy12

[Rath]

Is this vechile made for shirtsleve planets, or mars type stuff?

[0111narwhalz]

SomeGuy, you're really missing the point. I'm not trying to supply an AI with information. I'm trying to supply a pilot with information. Some things to note:

1: The autopilot is a glorified cruise control, and little more. It'll keep you going in the same direction, at (largely) the same speed, but not much more than that. Think MechJeb's rover autopilot.

2: The terrain information is not immediately evident. Think topographical map, you have to pay extra to get higher quality.

3: The terrain tends to be craggy, or full of gullies, ruts, and ridges. You can't really trust an AI to navigate this, like you can't trust an AI to rock-crawl.

4: Although you can go outside for a bit (hull breaches won't kill you), you really don't want to. Think high desert or Antarctica; you want that climate-controlled cabin.

5: Weather could be a concern. This includes solar weather, as well as simple rain and stuff. Might wanna keep an eye on that.

6: The vessel won't really be doing much SCIENCE! on the way; it's more of a freight sort of thing.

I'll update the OP with this stuff.