Re: [glob2-devel] Pheramonal-Directional Algorithm

[Andrew Sayers]

On Sun, Nov 06, 2005 at 03:25:01AM -0500, Andrii Zvorygin wrote:
>    for those that dont' know what pheramonal-direction algorithm is look at
>    Alleyways, and pathfinding and Lokadin's posts mainly
> 
>      - I think your pheromone-based algorithm can work, but I'm afraid it
>      will take
>      even more memory than the actual gradients. Could you do some estimation
>      for
>      a typical map? It would be nice because it would make a starting point
>      to
>      discuss the algorithm in more details and it would help us to be sure
>      that we
>      have all understood it.
> 
>    note: I haven't actually had a chance to look through the specifics of the
>    current wavefront but I mean there are only so many ways you can do a
>    wavefront

I did some calculations about how much memory wavefronts take up before:
        http://lists.gnu.org/archive/html/glob2-devel/2005-09/msg00032.html

Your details are mostly right, except you don't need to store
information about type explicitly - it's reasonable to assume that the
pheromone trail associated with your defence tower leads to the defence
tower ;)

Also, why use two bytes for the amount pheromone?  Surely you don't need
more than 256 levels of scent?

Also, pheromones do have to be recalculated, as the scent trails expire.
However, that's a much simpler calculation than propagating gradients,
and can be done much less often.

>      - I've been working a bit with ant biologists those last two years. Of
>      course,
>      ants uses pheromones, but:
>              - ant solve congestion (up to a certain extent of course) by
>      climbing on each
>      others, they live in 3d not 2d.
> 
>    unfortunatly i don't think we can't do that :(

I agree, that would look silly, but allowing globs to take resources
from one another would have much the same effect in some situations.

        - Andrew