RealBot and MS Visual Studio.NET

[V or 'Tex]

They do not move around because the bots create their own nodes. Thus when they 'wander about the spawn' they create dozens (or in my experience, hundreds) of nodes in a very tight area. These nodes are all interconnected.

The problem of course for playing with Realbot is that you must first play it with NO bots and simply walk around the map for at least 15 minutes from both spawns, to every important area, along every hidden path, etc.

I play Mindmaze2 with 7 bots (prior to the current version I was able to play with ~23) and it's still rediculously laggy.



^ That's a picture of Mindmaze2.

Before I ran a single bot through the map I walked everywhere. I criss-crossed the ground at least 60 times in two different directions, I circled around every pillar, box, and ramp. I went through each tower several times, and I did every possible route on the walkways.

Unfortunately, in hindsight I should have double-noded the walkways (walking along each EDGE to try and fit 2 nodes) so that they would have a higher chance of being selected. The AI very rarely goes to the upper levels, but after playing CT many times, if I switch to T the AI is smart enough to follow me up to the highest walkways and chase me through towers.

I am going to re-node it with that system in mind, because currently its rather rediculous that every game consists of a single ground rush, and possibly a few bots being smart enough to run up and pick off the others with altitude advantage.

[@$3.1415rin]

do you think it's yet time to optimize a bot for such crazy maps ?

Dear Stefan!

Originally Posted by stefanhendriks

I will rebuild my entire dll and then add bots on a map that has no nodes. Then i will simply walk around (bots shooting off)...

what kind of computer do you have in CPU terms? I have a P4 2.5 Ghz

Well, I have an AthlonXP 1800+ and 1024 MB RAM, that's enough. And - to be more precise - I'm telling the different behaviour my self-compiled DLL shows compared to the WIP made by you! So I just wonder why my DLL is so slow and the bots do not learn as I know from your WIP ones...

Still no ideas?

Thanks a lot...

[frashman]

maybe what vor'tex say is not so false .... i hadn't test this yet. But so far I know, only humans can create new waypoints?!

I observe your development for a while. I have another theory. I think if you run a new map, and if you run it with realbots from the beginning, they haven't any waypoints ... isnt it?

maybe they have their startpoints. now the only goals they see, is another startpoint. so they run from one startpoint to another ... they learn this path as a good path, because they don't die. If you as a human run to their startpoint, they truely kill you. so they learn it's a good position. after playing 10, 15 rounds you have set waypoints all over the map. so they could run routes. but they don't want to, because they have always "good" positions.

after more rounds, they die at their postions and "think" its a good idea to test another route, not only their startpoints and the bot begin to do, what it should do. run around and play against you.

can this be the problem?? if yes, i will try to solve it theoretical. to hack it in the code, i don't have to much knowlege about your sourcecode and no running cvs :-)

mfg Frashman

[stefanhendriks]

The node machine learns from everyone, including bots themselves. The bots do not learn what good or bad nodes are in that sense as fragmen sais. Its actually way more simpler:

Wherever you go, nodes will be plotted. Where nodes are plotted, they are connected to other nodes (sort of autowaypointing). Now, there are goals, these are assigned to node's. So, for instance if you run a map that is not 'noded' yet, you will have auto-added nodes at goals, ie bomb spot and spawn positions. Now these are the only nodes available at the moment.

When bots want to go to another goal (mostly enemy spawn point, bomb spot, etc), they find they cannot create a path to it. THis is where my new code kicks in, it counts that amount of times when a goal cannot be reached, and then it gives it a bad score... for now.

When humans walk around the map, more nodes get added, so the path finder will try these nodes too (thus making it slower, but yet, this is not the problem).

When you have reached the enemy spawn point and connected to one of the nodes, b ots will magicly start to move. THis happens to my DLL and i tested it on some other computers as well.

Why the dll of musicman is slower, i dunno. Or, why a self-compiled dll is slower, id on't know either. All i know is the compiler is different, i use MSVC 6.0...

Did you try the new DLL at the filebase btw?

[@$3.1415rin]

maybe it's so slow because you compiled it in the debug mode ?

[frashman]

I am sure, that the principle of good and bad nodes was integrated in Realbot AI, wasn't it?

If i understood it now right, Realbot walks totally randomly to his randomly set goals after all waypoints exists?? This means Realbot doesn't learn good sniper points or camppoints or points with good stratetical positions?

Perhaps i should start a new threat about this?

[stefanhendriks]

@ asp, i do not know how he compiles it. But my project space does not have a debug mode, i just made my own 'debug mode' sort off. Its sort of release mode with some debug elements? dunno how to call it rofl

@ frashman, realbot does not randomly move. It does not learn sniper spots, but it does learn camp spots and such. They learn much more then i just described, the 'theory' you pointed out is not valid though.

[frashman]

ok. i think i will install cvs and ms visual studio.net tomorrow.
if i get it to run and to compile, i write a short tutorial in this forum how to do this, for all interrest...

then i try to understood how your realbot does learn and want could be better if i have an idea.

But when i am learn for my examinations?? damm