33 replies to this topic

[LarkinVB]

I noticed two big and some small problems which do need attention.

1. We must find a better way to detect a threat on the path of a unit. I painfully watched some DoWpro replays (using 1.75 core AI) showing this problem. The AI was locked in its base and did send squad after squad to some distance point walking right through about 2000 points of enemies.
The AI looked real stupid. I know the Relic code can cause crashes but perhaps there is a better way to handle this. It is happening more often on maps with curved pathing which are not that rare.

2. There is still this annoying ork/CC back and forth movement. Happens very frequently and shouldn't be ignored. I watch them attacking up to nearly touching the enemy, then retreating a short distance just to return. Sometimes they finally fight, somtimes the retreat, all the time they loose troopers for nothing.
Together with 1. this is a real show stopper.

3. Attacking squads should gather more consequently. I watched a leading squad followed by some bigger group. Often the leading squad was killed before the group did hit. This more important for orks.

4. Jump/teleport squads jump-attacked the enemy just to immediatly retreat on foot through the enemy lines without fighting their target at all.

5. Farseer should dance healthy commanders as she will loose fighting close combat. I will change that.

[Finaldeath]

Alter the AI for the sorceror in addition to the farseer, even with the 1.5 patch notes (1000 HP now), he's dead weak (low powered attack weapon), so dancing for him should be a prioritory against high health enemy commanders.

I'm slowly working on my attachment code still, if I get anything workable, I'll pass the code around to test, sound good?

[ArkhanTheBlack]

1. We must find a better way to detect a threat on the path of a unit. I painfully watched some DoWpro replays (using 1.75 core AI) showing this problem. The AI was locked in its base and did send squad after squad to some distance point walking right through about 2000 points of enemies.
The AI looked real stupid. I know the Relic code can cause crashes but perhaps there is a better way to handle this. It is happening more often on maps with curved pathing which are not that rare.

Yes, though I'd try to solve this a bit different. If the squad meets a superior force, it'll retreat. Even if HasThreatOnPath() would work, it can't detect forces that cross the path after the check and it doesn't return an accurate value about the strength of the threat.

2. There is still this annoying ork/CC back and forth movement. Happens very frequently and shouldn't be ignored. I watch them attacking up to nearly touching the enemy, then retreating a short distance just to return. Sometimes they finally fight, somtimes the retreat, all the time they loose troopers for nothing.
Together with 1. this is a real show stopper.

I've seen that too. I'll think about how to coordinate the whole problem of multiple demands resulting in chaotic movement changes. It would be good to know which demands collide most often. I'll probably make a list about which situations can occur and how to react on them.

3. Attacking squads should gather more consequently. I watched a leading squad followed by some bigger group. Often the leading squad was killed before the group did hit. This more important for orks.

I've a crazy idea about an 'army formation' algorithm, but I don't know if it's possible since I'd need more terrain info. It's a shame that the GetPathingDistance() method is buggy. It would be very usefull for that. Maybe I really try to call it with a pcall command, but I don't think that I can avoid a crash with it.

4. Jump/teleport squads jump-attacked the enemy just to immediatly retreat on foot through the enemy lines without fighting their target at all.

Interesting! It's a while since I saw this the last time. It's probably part of the point 2 movement problem.

5. Farseer should dance healthy commanders as she will loose fighting close combat. I will change that.

Good idea! I've seen that quiet often, but I did not really 'realize' the problem.

I'm slowly working on my attachment code still, if I get anything workable, I'll pass the code around to test, sound good?

Fine! I personally won't code too much this month, so you've even a free playground for some weeks. I'll probably start serious coding again in may.

[LarkinVB]

It would be good to know which demands collide most often. I'll probably make a list about which situations can occur and how to react on them.

Perhaps it is a good idea to log squad moves to file. Since we use the move wrapper function we could log move commands by text. If there are different commands to the same squad within a small timeframe there is most likely a problem. I will tackle this and can give you some detailled feedback when you are back to work

Edited by LarkinVB, 10 April 2006 - 04:45 PM.

[LarkinVB]

One more suggestion. I tried to use values like 50 for ork banner addon_1 and 25 for ork banner addon_2. What I expected was to get 50% of all banners with twin bolters and 25% of those with rockets as addon_1 is a prereq for addon_2. I did not get a single one with rockets. It seems your code does assume that 100% banners will have addon_1 before addon_2 is requested. This is not the case with orks banners. Often you don't want all banners to get addon_1 at all. Can you fix that ?

[Finaldeath]

Good catch on the banners, since these "turrets", while a good early-mid game defense, are too expensive to waste resources when some are never ever in combat

LP's should always be tier 1 first, turrets shouldn't be though.

[thudo]

For Ork Banners why not just push back their addons to higher tiers? I'm game for that ALTHOUGH the forthcoming v1.5 patch, at least unofficially, looks like they are gonna buff em again. Could be a new revitalization for DA ORK BOYZ!

Lets just hope Relic listens to the community as balancing 5 factions is such a fragile act! Esssh!

[Finaldeath]

Thats just a different suggestion Thud that mearly implies all banners need upgrading to teir 1 (for some reason) but later in the game...huh?

Anyway, perhaps do both - less upgrading to teir 1 for unused turrets, and do it later then it currently is.

[thudo]

Orks need all the cashola they can get BUT that will likely all change if Relic shows more ORK LOVE in May. April Patches bring Grot Byatches!

[Finaldeath]

I am now totally confused and I think thuds gone mad.

I'll still look into leader attachments, god forbid I have anything to do with Orks though!

[thudo]

Oh blaaah! Yer no fun..

My point was how the factions were to be treated in 1.5? This remains unclear.. it will once again change the arguments.

[LarkinVB]

Stop cluttering my thread with 1.5 guesses. I like Arkhan to lokk into the banner addon code and you drown him with chatter

[thudo]

Why you! I'm the admin here and.. Oh poo. Okay.. I hijack less.

[ArkhanTheBlack]

One more suggestion. I tried to use values like 50 for ork banner addon_1 and 25 for ork banner addon_2. What I expected was to get 50% of all banners with twin bolters and 25% of those with rockets as addon_1 is a prereq for addon_2. I did not get a single one with rockets. It seems your code does assume that 100% banners will have addon_1 before addon_2 is requested. This is not the case with orks banners. Often you don't want all banners to get addon_1 at all. Can you fix that ?

The problem with second addons is that the build controller assumes that the addon is already done if he can't add it to the building stack of the addon building. But since he can't add the addon 2 to buildings with a missing addon 1 he accidently thinks that the addon 2 is already done. And the problem indeed vanishes as soon as 100 % of the buildings are upgraded with addon 1.
At the moment addons have no information about other required addons. To fix that I'd have to expand the buildcontroller. How important is this for you? At the moment I'd just add it to my ToDo list, but if it's really urgent then I'll try to add it this week.

[LarkinVB]

Take your time. Nothing is really important. Relax, have fun and come back as soon as you like to.

[ThetaOrion]

5. Farseer should dance healthy commanders as she will loose fighting close combat. I will change that.

--

I have never once won as the 1.75 Eldar at HARD setting on Mountain Trails or Soul Forge or anything after about B13 or thereabouts. The Eldar just can't keep up in terms of research and production against the Chaos and the Orks, and SM AI allies aren't enough to keep me alive either as the Eldar. I can't take or hold territory either as the Eldar at the HARD setting. In Plain Vanilla, the Banshees chewed through the Chaos Spam, and were practically indestructible -- you could definitely replenish them faster than the enemy could destroy them. Now in 1.75 of AI Mod, the Banshees melt away and are completely impotent against Chaos and Orks, even with a Farseer attached.

Of course, this FarSeer thing is just a small part of the problem. She's dead in a matter of seconds whether she is attached to something or not, and units that she is attached to don't seem to get any benefit from the attachment.

The Eldar haven't felt right or been right or had any teeth in the opening tiers for a very long time. Your base never lasts long enough to get Fire Prisms.

I can win as the Eldar on SoulForge at the STANDARD setting no problem. The Eldar can take and hold ground at the STANDARD setting of the AI Skirmish Mod and they produce as fast as anyone else.

But, there's just no easing into the HARD setting with the Eldar, no training or experience makes any difference. I lose no matter what I try with the Eldar and 1.75 of the AI Skirmish Mod at the HARD setting. They just can't keep up in terms of production and research at the HARD setting.

Larkin got his wish, many unbeatable scenarios at the HARD setting of the AI Skirmish Mod.

[ThetaOrion]

While trying to think of all the ways that Thudo can call me an idiot for posting my observations, I thought of another observation that I have made in the past.

In the past, with 1.41 Plain Vanilla, I noticed that the Enemy AI seemed to get a slight production advantage over the human and the Allied AI. The human and the Allied AI were slightly handicapped in terms of production at the HARD setting of Plain Vanilla. It's one of those things that only seemed to show up with a human in the mix.

I think this phenomenon is showing up or even being amplified in the 1.75 version of the AI Skirmish Mod at the HARD setting. Certainly no need for the HARDER or INSANE settings on these big maps with lots of AI involved. The HARD setting produces plenty of unbeatable scenarios.

But, this phenomenon doesn't explain why the Eldar Banshees and Eldar Seer Council have gone wimpy against the Chaos Spam. In Plain Vanilla 1.41, I could take a squad of Banshees or the Seer Council right through Chaos Territory and they could stand up to the spam, and with a FarSeer attached, they could harrass and be replenished faster than they were destroyed at the HARD setting and sometimes last in enemy territory the whole game. Not so with 1.75 of the AI Skirmish Mod. FarSeer attached or not, these Eldar units die faster than they can be replenished.

When you lose every encounter between your Eldar and the incoming Orks and Chaos and not a single unit you produce can hold at the HARD setting of 1.75, then you try pure research, and once again, your Eldar base is overrun by tanks and hoards of Ork and Chaos troops, because you don't have a single unit or tank in your arsenal, and you haven't even begun to finish the research, and that fails too. They out-research you and out-produce you two to one at the HARD setting, or so it seems, and all of your Eldar troops, what few you have, melt away like dew in a hot summer wind.

Of course, there's no way I can predict all they ways that Thudo can think of to tell me that my input doesn't apply and all the ways he can call me an idiot, but despite his namecalling when I mention trends that I'm seeing that he doesn't like to hear about, I do tend to notice trends. I have noticed a trend since about B13 or thereabouts when I first mentioned it -- I can't win as the Eldar at the HARD setting of the AI Skirmish Mod. It might be possible that I just haven't found the key to victory yet, but so far everything I have tried hasn't worked for me, whereas it worked perfectly fine for me in the past and got me the occasional victory. Dozens of games as the Eldar at the HARD setting over the past month with not a single victory, I count as a noticeable trend. Yet I keep trying to find a way to win. It's just that at any moment of the game, the enemy AI has twice as much as you do, and it's all a lot stronger and does a lot more damage, and your Eldar troops just melt away. Soon, they have tanks and outnumber you ten to one while you have nothing, maybe a tank or a set of troops, and that's it. It looks and feels like a trend to me, despite what anyone else might try to say to debunk it.

My question is, "Will that trend hold true for the other factions in the game as well?" I haven't played the other factions enough yet to be able to say one way or the other.

[ThetaOrion]

Well, I continued to study it today.

1.41 Imbalance Showing through:
In the past with the 1.75 betas, I got ninety percent of my wins as the Space Marines or as Chaos, so I haven't been playing them much lately. I'm more interested in the balance or the win statistics of the other factions. But, I have been seeing signs of the 1.41 imbalances shining through in the 1.75 AI Skirmish Mod.

The Eldar Situation:
As mentioned, I haven't won as the Eldar yet since around 1.65 B12 or before. My theories for that:

• Humans and their AI allies receive a production handicap at the HARD setting.
• Something has been left undone or undercooked with the Eldar that was there in 1.41 PV and is not there now.
• The game plays completely differently with a human in the mix, and most of the testing and balancing was done AI vs AI, which isn't even remotely the same as when there is a human in the mix.
• Or, the AI unites and goes after the human player first.

We like to blame Relic for all the glitches and imbalances in the game. The problem with this is that I don't remember the 1.41 PV Eldar being so wimpy. In 1.41 PV, the Eldar could take on the Chaos' superior numbers and hold them and even win against them. Not so in 1.75 AI Skirmish Mod. I have tried the tank route, the relocation route, the troop route, and the research route, and no matter what route I take, my production can't keep up with Chaos, and my troops and tanks never win a single skirmish. The expensive Eldar troops and tanks just simply can't stand one to one against the enemy and win, at the HARD setting of the AI skirmish Mod. And, in every game I play as the Eldar, the enemy seems to outnumber me at least two to one, right from the start.

The Ork Situation:
In the game or two that I tried on Soul Forge in the past month with the Orks, I ran into the same problems. As the human player, I seem to have a production handicap, and I just can't win a single skirmish or a single battle, because one for one, the enemy units were stronger and they had more of them, and Waagh Banners were no protection. But, in this situation, it is in line with what I saw from the 1.41 PV Orks when I played them as a human player. And, in the AI Skirmish Mod, the 1.75 AI Mod Orks really shine as an AI enemy, especially if you are going up against them as the Eldar. Alas, I haven't played enough games recently to know and feel for sure how the Orks stand in the 1.75 lineup. The 1.75 trends aren't as firm or conclusive for me where the Orks are concerned. I don't know if it is possible for a human player to win as the Orks on Soul Forge at the HARD setting with 1.75 of the AI Skirmish Mod or not. I do get the sense, though, that the Orks and the Chaos do much better as AI enemies than the seem to be doing as AI allies.

The IG Situation:
Back in the Betas, I was having a hard time winning Soul Forge as the Space Marines because my Eldar and IG allies proved rather worthless against Chaos on that map. But, I have won Soul Forge at the HARD setting in the last month with the Imperial Guard, with SM and Eldar AI allies. I tried them again tonight, the Imperial Guard at the HARD setting, and I lasted at least twice as long as I do when playing as the Eldar. You really have to concentrate on keeping your research and your troops and tanks going all the time, but at least you have enough production to win. You also have to make better use of heavy cover than I did, because the IG excel when it comes to using heavy cover. Anyway, when I finally got overrun, I noticed another trend that has been somewhat ongoing ever since Beta 7. The enemy AI at some point or another unite and come directly after the human player's base. That has been happening over and over again with the Eldar, and it happened again with the Imperial Guard tonight, which means that as the human player, you have to build most of your base in allied territory if you want to postpone your demise, which I didn't do, because the AI will come to your base first and attack it first. From what I saw, it's still possible to win as the IG on Soul Forge at the HARD setting of 1.75.

Summary:
Anyway, it is my belief that not all of the problems are tied to an inheritance of Relic's imbalances. There are a couple of other 'imbalances' that have since been introduced in 1.75 that tend to work against the human player in the game. And, as some of us know, when a human concludes that he has no hope and no chance of ever winning, then he eventually stops trying. After a month of it, I'm pretty much to the point with the 1.75 Eldar at the HARD setting that I have no desire to try it further, because I am now pretty much of the belief that this human is always going to lose when he plays as the Eldar, whatever the cause or the reason for the losses.

So, I go and try to get a win as the Eldar at the STANDARD setting of 1.75, and the AI allies are running them down and beating them up faster than I can get to the enemy. Anyway, Thud's dream of progressively training one's way through the different difficulty settings from EASY, through STANDARD, to HARD mastery isn't quite reaching reality or fruition for me, where the 1.75 Eldar are concerned. Nothing in the STANDARD setting or in months of previous Eldar play prepared me for what I actually experience while playing the 1.75 Eldar at the HARD setting, and I still haven't found the key to victory.

As the rest of you start racking up victories as the Eldar on the larger maps at the HARD setting of 1.75, be sure to upload your replays here so that I can witness your superior mastery and skills and learn from your example. I don't have what it takes.

Edited by ThetaOrion, 12 April 2006 - 06:50 AM.

[Finaldeath]

As the rest of you start racking up victories as the Eldar on the larger maps at the HARD setting of 1.75, be sure to upload your replays here so that I can witness your superior mastery and skills and learn from your example. I don't have what it takes.

You're on I'll play a 2v2v2v2 with lotsa eldar and see who comes out victorious. (bets on me! ).

I am willing to go test this "resource handicap". I generally don't see anything that resembles it since the AI is more competant with regards to keeping troops alive (well...sometimes).

So, I'm going to look through a replay saved game, and check the timings of resources and rates and stuff.


I'll also, after doing commander AI, look into resources and reinforcing troops. For some armies, such as Orks, they need to definatly reserve money to upgrade nobz and weapons (on melee troops - I see them often as not on shootas where the nob is out of range! would be better to get big gunz for them), since sometimes I see the AI suffer in this regard. We'll see how I get on at that, currently you can only reserve money to build new things


Also, on the patch, it might be worthwhile waiting for the next release to go out and co-incide with the patch, which might alter some parts of the AI (eg: turret building, tier 1 troop usage, landspeeder and vyper usage! and a few additional restrictions or new things for armies) more dramatically then other patches.

Behaviour can still change in our alpha beta changes I am sure but there are some dramatic build order changes that could take place (or at least tweaks) with 1.5.

[thudo]

Also, on the patch, it might be worthwhile waiting for the next release to go out and co-incide with the patch, which might alter some parts of the AI (eg: turret building, tier 1 troop usage, landspeeder and vyper usage! and a few additional restrictions or new things for armies) more dramatically then other patches.

Yeop! Although the AI devs are having a well deserved rest. Even the SMARTEST of brains needs some time off. Looking quite forward to 1.5 in May. We'll have to make some obvious alterations to the buildorders and stats but thats expected with these Relic releases.