Spring RTS Engine

I have to sort out whats going on with the attacking problems IK said about first

attacking is partially working but competetive performance is still low. The problem I'm encountering now is that even with an excellent config NTAI plays very badly, so much so that every AI but RAI can still defeat it most the time.

I put this down to a number of issues

1) Attack group behavior:

When an attack group is ready, it is selected and given move commands to the chosen location. the group proceeds there in a long thin line, at the speed of the fastest unit, meaning by the time the heavy units reach the battle all the fast units are already dead. units move in a straight line to their destination, ignoring counter attacks, defenses etc.

attack groups should always fight-command into battle, and keep reforming so they do not get spread out.

2) Threat behavior:

as we discussed before. picks the strongest location to attack rather than the weakest. travels in a straight line from factory to enemy base, resulting in all attacks occouring down the same vector, making it extremely easy to defend against.

3) Defensive behavior:

non-existant. If one of NTAIs con units is attacked out in the field, there is no response. every other AI tends to move units to defend itself where attacked. NTAI only defends via mass counter attacks, which only happens against large attacks. raids devestate NTAI without provoking a defensive response.

Ideally, units could be set to gaurd cons, units would respond to attacked cons, and con units could be given a tag that causes them to retreat if attacked, then resume their buildlist after finished retreating

4) Commander behavior

Simple enough: no dgun.

5) Idle attack groups

A group that has successfully arrived at its attack location and survived will idle there indefinitely. this is a huge problem when a strong attack force attacks a far off point and then sits there idle for the rest of the match.

Ok, an idea:

since reconstructing the entire attack system so its actually good is a huge job and out of the question, how about splitting attackers into several diffrent groups.

Attackers:

As now, except fight commands

Raiders:

Units that move into enemy metal extractors

Defenders:

Units that move in reaction to attacks, or gaurd con units etc

In this way, for example, I could put weasels as raiders, and all game NTAI will use weasels as intended. I can put all the fast tanks into Attackers and put the slow riot units into defence. Result: all unit types are used better, and it would be alot less effort (if more work for the config maker) than recoding your threat system.

There is no dgunning anymore?! Put in redmine!!!

Issues you put in redmine will be addressed, they have to be dealt with else I cant reach 100% on my shiny progress bar and make a release. Your making work for me here, everything ahs to eb documented even fi its only to make it known it exists.

Same goes for feature requests. Ill think about mex raiding in the meantime...

Remember, attack units can have task lists too...

your making work for me here, everything has to be documented even if its only to make it known it exists.

I'm doing work for you here by testing and bug-reporting for your releases then taking the trouble to give detailed feedback. Its no surprise to find that according to my profile this 90 page thread is my most active topic; that is because this thread is mostly comprised of us arguing.

Anyway, its documented right here, its not like it is going anywhere. copy-pasta done.

as to dgunning, yeah, looks broken- pity because your old "shooting down hawks" predictive dgunning was pretty fun.

attackers can have tasklists indeed. The addition of several tasks like:

Attack
Unit becomes an Attacker and is included in next attack group, regardless of whether the unit is by default attacker=1;

Mex Raid
Unit receives move order to an Enemy metal extractor

Defend
etc etc

would be pretty handy

I mean I have to redo all your posts as redmine tickets, and flag them according to type. Its a lot easier for the both of us if you entered them in there rather than here as it saves me a lot of organisational hassle and strange questions from my professors

Ideally, units could be set to gaurd cons, units would respond to attacked cons, and con units could be given a tag that causes them to retreat if attacked, then resume their buildlist after finished retreating

There is already a retreat behaviour which was built specifically for this purpose...

Dgunning behaviour tested and rejected

http://redmine.darkstars.co.uk/issues/156

Dgunners now target at 1.5x their attack range rather than 1.3, this should make it more apparent.

http://redmine.darkstars.co.uk/issues/158

AF wrote:

Ideally, units could be set to gaurd cons, units would respond to attacked cons, and con units could be given a tag that causes them to retreat if attacked, then resume their buildlist after finished retreating

There is already a retreat behaviour which was built specifically for this purpose...

yes but it is no use as part of a units tasklist. it needs to kick in only when the unit is attacked

I think fixing the 'attack groups idle after attack completed' issue is a high priority fix

1v0ry_k1ng wrote:

AF wrote:

Ideally, units could be set to gaurd cons, units would respond to attacked cons, and con units could be given a tag that causes them to retreat if attacked, then resume their buildlist after finished retreating

There is already a retreat behaviour which was built specifically for this purpose...

yes but it is no use as part of a units tasklist. it needs to kick in only when the unit is attacked

Noo it isnt just a task, it is also a behaviour!! See CRetreatBehaviour.h/cpp

I think that the threat matrix is empty preventing attacking because of the multiplier that reduces a threats value when added to the matrix as it gets further away from the AI. I've commented this out to see what it does.

AF wrote: Noo it isnt just a task, it is also a behaviour!! See CRetreatBehaviour.h/cpp

bear in mind I have no idea what you are talking about there-
I have seen no evidence of cons retreating when attacked, nor any tag that causes them to do so.
how is that behavior applied?

Ill check the exact logic in a moment.

You'll be glad to know I have chanegd atatcking slightly. NTai was dividing threat values by their distance from the AI base (defined as the nearest base point, which would include any position a factory/hub was built on and the start position ), this step ahs been removed so no distance scalign applied, and attackign works nicer now when targetting enemies.

Spring 0.8.5.1
BA 7.04, Archers Valley V5
Insta-crash on game start. Infolog empty, log files of AI are not even created - so everything indicates that the AI didn't even start to work.
(Tried same with RAI - works fine). Same version (your latest one) worked (well, somehow) before upgrade to 80.5

I do have infolog.

* Spring 0.8.5.1
* BA 7.04
* 2Lakes

I appreciate the bug reports, but afterall a new spring release, and my current code base doesn't match the public binary.

What'smore infolog.txt is useless to me. I have no idea how to work with mingw32 stack traces, and my attempts to meddle with addr2line or whatever the tool is called have ended up with hundreds of question mark characters.

As for posting them in this thread, it would help me a lot if you put them on my redmine tracker, because if you put them in there they have to eb dealt with, but if you post them here, then I might not catch ti all as it gets buried under new posts, and any I pick up on I have to retype into redmine. It saves the both of us masses of time and effort that would otherwise be spent debugging and coding..

http://redmine.darkstars.co.uk/projects/ntai/issues/new

What's more, if you put it in there, it sends me an email and I get it on my phone so I know within 5 minutes that a bug report or request has been filed.

So the moral of the post:

1) Bugs in redmine are 50x more likely to be fixed than bugs in forum thread

2) I'm working on 10.1 and Ill have it out soon but until then, hold tight!

if you made NTai compile with all boosts, we could put the new version in, and you could translate stack traces with a single line on the buildbot (in case the user used the NTai compiled with MinGW/by the buildbot).
as i got it, this will happen when you moved to the new interface.

AF wrote:What'smore infolog.txt is useless to me. I have no idea how to work with mingw32 stack traces, and my attempts to meddle with addr2line or whatever the tool is called have ended up with hundreds of question mark characters.

I don't understand one thing. What debug info (which is really valuable for you) does your AI proved then? What should i as a tester upload to your bugtracker? Distribute with AI binary at least PDB file to let people with VS.NET to give more detailed info.