Randomized pathfinding

Caydr · Post by **Caydr** » 24 Jun 2007, 20:47

Would minor randomization to the pathfinding be possible? Maybe so that one out of every 20 pathfinding decisions would be slightly different from another unit on the same path?

Not really important, I just thought it could help to eliminate the problem where every unit in an entire train will get held up by a single unit that's slower than all the others. Since they're all determined to follow the same route, they just keep bumping into each other.

With fewer collisions to gum up the works, pathfinding calculations might in my ignorant view possibly be sped up a little bit even.

LOrDo · Post by **LOrDo** » 25 Jun 2007, 07:28

This would be quite cool. And though I'm no coder, I think it may be quite complicated.

ironized · Post by **ironized** » 25 Jun 2007, 07:42

im not a coder either but i foresee it to be easy,
all you need to do is instead of it deciding on the quickest route, it decides on the 3 quickest routes then each unit picks one of the 3 randomly

SwiftSpear · Post by **SwiftSpear** » 25 Jun 2007, 09:42

ironized wrote:im not a coder either but i foresee it to be easy,
all you need to do is instead of it deciding on the quickest route, it decides on the 3 quickest routes then each unit picks one of the 3 randomly

The 3 quickest routes are likely to be a segment of 1-2 total pixles off of each other... There are potentially infinite possible routes and the routes at the top of the bell curve in the weighted function will be virtually identical to the naked eye.

ironized · Post by **ironized** » 25 Jun 2007, 12:32

well then the easiest the 100th easiest and the 200th easiest

Zpock · Post by **Zpock** » 25 Jun 2007, 13:04

I get annoyed that units moving close together constantly try to get around each other doing a lot of awkward turns. I'm thinking about trying some random small unit speed variations in the units script and see what that does.

For random pathing just make the unit go in the wrong direction by some varying random number from the planned path. Might just look like it's drunk tough.

SwiftSpear · Post by **SwiftSpear** » 25 Jun 2007, 13:19

ironized wrote:well then the easiest the 100th easiest and the 200th easiest

It doesn't list out every failed combination, it just finds the most efficient path at the top of the bell curve and ignores everything else. In order to find another path you need to prioritize something besides efficiency in the algorithm. It's impossible to find the 100th least efficient path. There will be two, one tied with the other at the opposite side of the bell curve, if not more.

You have a very very poor understanding of the mechanics of path finding AI. Spring's pathfinder is not a simulated trial and error navigation system.

hunterw · Post by **hunterw** » 25 Jun 2007, 16:09

does spring use anything randomly generated?

i think it might be the only game that does not use a random number generator. if it does where does it use it?

Guessmyname · Post by **Guessmyname** » 25 Jun 2007, 17:16

Wind Power and Direction? (That's about the only thing I can think of)

Zpock · Post by **Zpock** » 25 Jun 2007, 17:19

You can use random numbers in the cob script.

It would be awesome if effects could be affected by wind, like TA. Smoke moving in the winds direction and stuff.

Peet · Post by **Peet** » 25 Jun 2007, 17:36

Zpock wrote:It would be awesome if effects could be affected by wind, like TA. Smoke moving in the winds direction and stuff.

It does...look at a map with ludicrous wind.

Neddie · Post by **Neddie** » 25 Jun 2007, 17:46

Peet wrote:
Zpock wrote:It would be awesome if effects could be affected by wind, like TA. Smoke moving in the winds direction and stuff.
It does...look at a map with ludicrous wind.

Of course, the actual windmills are merely scripted animation.

Caydr · Post by **Caydr** » 27 Jun 2007, 03:44

All I notice is, when I push "B", there's this tree-like structure branching out in front of units as they pathfind. Can a unit not be told to just, rarely, pick a branch and go with it, then resume as normal when it reaches the next segment?

Argh · Post by **Argh** » 27 Jun 2007, 03:58

I actually talked to K-man about this, and proposed a crazy scheme involving using a pixel-map to tell steering AI about the probable outcomes of their decision-trees, to solve a lot of this mess, but I'm too stupid to go beyond a theory and actually make it work

imbaczek · Post by **imbaczek** » 27 Jun 2007, 10:35

You could insert a midgoal on some non-optimal branch. There are problems, though - when you absolutely, positively need the unit to follow the exact path.

rattle · Post by **rattle** » 27 Jun 2007, 22:49

Sometimes I wished spring could be gridbased, i.e. 2x2 elmos (or what ever it was called) = one square. Units would always move to the center of a square.

And this is totally off-topic.

Yeah good idea.

Spring RTS Engine

Randomized pathfinding

Randomized pathfinding