Here's Johnny Ca$h, from germany. The first thing to say is: THANKS!
You can't imagine how happy i was, to find Warzone back, it's just awesome.
Well, but also i found something i dont like. At first the general Skirmish AI, second the Aivolution AI and third the little sweet bug, i found really annoying. When i'am moving a group of units, sometimes one to three of 'em dont react on the "move" command. But they react if i send 'em to the HQ or Repairstation or something else.
Okay, now back to the AIs. I dont know how much work is needed to create one, but the current AIs are bad. I'am a bit skilled in Warzone, not the best but a bit skilled and the AI didnt have any chance against me. On an 1vs7 they got a chance, if i make a few fails, but that's all (i'am allways playing against the highest skill for the AI). And sorry, but Aivolution is really bad. Not using all ressources, not building an advanced base, no evolution at all. First i thought that Aivolution is gathering ALL informations, but i checked the learningdata and for every player the learningfile never changes the size (103 kb).
I think the AIs are just to... Static. Not really reacting from a pool of reactions, just proceeding its "one way". Okay, it's a bit fun to defend against that mass of units, but really repetitive and once they got the Satelitelaser, it's over if i can't kill 'em all at once. So, i'am asking now for a better AI, an AI who really reacts and play more like a Human (cause they're really rare in Warzone) and changes it's strategy if one fails.
Thanks for your time and reading my post.
Best regards,
Johnny Ca$h.
Hey there!
Re: Hey there!
hello,
it's not hard, I developed one a few years ago.
Problem was that parallel to my AI they continued to improve the existing AI, and the result was that both had different advantages and disadvantages.
One thing that would triple the AI strength is to set all units to NEVER repair. That's one line in the code. That's a fact, not an opinion. It is very very very very rare that an AI makes good use of repairing a damaged unit by aborting a useless attack and sending it back later. It happens very very very very often that an attack that would win the whole game is aborted just because the attacking units are down to 66 % of their health, and it happens very very very very often that a unit returning for repair causes a traffic jam, EVEN MORE when the human makes "traps" just for that to happen (just a few defensive structures at the END of a canyon - not strong enough to kill anyone, just to cause 33 % damage).
A huge regression from 2.0 to 2.1 is that the AI when starting without bases and in a team sends ALL trucks to far away locations to build oil. That's the biggest mistake that has been made in that huge AI code change. In 2.0 the problem was that the AI did not build enough derricks, but that's not a good way to fix it at all. It sometimes takes 10 - 15 minutes before the AI builds the first factory and lab. The quickest fix would be to allow this oil-hungry behaviour a few minutes after the game started, not earlier. Or only allow it when there is at least one factory.
That are the two biggest problems (the second one only when starting in a team with no bases). They are easy to fix. More can be done, but those are the lowest hanging fruits and the highest priorities in one.
Some things are great since 2.1, for example that the AI finally does devastating VTOL attacks, great both in numbers and in target selection. One potential improvement there is to detect when the attacks are not strong enough to overcome AA and then to attack in bigger groups, but that's an example for a SMALL improvement with HUGE effort.
it's not hard, I developed one a few years ago.
Problem was that parallel to my AI they continued to improve the existing AI, and the result was that both had different advantages and disadvantages.
One thing that would triple the AI strength is to set all units to NEVER repair. That's one line in the code. That's a fact, not an opinion. It is very very very very rare that an AI makes good use of repairing a damaged unit by aborting a useless attack and sending it back later. It happens very very very very often that an attack that would win the whole game is aborted just because the attacking units are down to 66 % of their health, and it happens very very very very often that a unit returning for repair causes a traffic jam, EVEN MORE when the human makes "traps" just for that to happen (just a few defensive structures at the END of a canyon - not strong enough to kill anyone, just to cause 33 % damage).
A huge regression from 2.0 to 2.1 is that the AI when starting without bases and in a team sends ALL trucks to far away locations to build oil. That's the biggest mistake that has been made in that huge AI code change. In 2.0 the problem was that the AI did not build enough derricks, but that's not a good way to fix it at all. It sometimes takes 10 - 15 minutes before the AI builds the first factory and lab. The quickest fix would be to allow this oil-hungry behaviour a few minutes after the game started, not earlier. Or only allow it when there is at least one factory.
That are the two biggest problems (the second one only when starting in a team with no bases). They are easy to fix. More can be done, but those are the lowest hanging fruits and the highest priorities in one.
Some things are great since 2.1, for example that the AI finally does devastating VTOL attacks, great both in numbers and in target selection. One potential improvement there is to detect when the attacks are not strong enough to overcome AA and then to attack in bigger groups, but that's an example for a SMALL improvement with HUGE effort.
-
Johnny Cash
- New user
- Posts: 4
- Joined: 07 Oct 2009, 15:42
Re: Hey there!
At first, thanks for the answer. Just coming back and find one, makes me happy, so i've something to write! : )
Well, back to topic, i can agree only in parts. I never noticed devastating VTOL attacks. And that the numbers per group changes, i've never noticed too. Most times, the general AI, attacks with maximum 8 VTOLs with medium or small bodys. Only sometimes i saw more with heavy tanks. Both spotted in lategame. I would prefer not only small changes, that the AI got a pool with more actions, reactions and buildorders. A bit more dynamic behavor would awesome. I've never noticed that the AI is dynamic in any way.
A nearly perfect BO should be a priority. In every game i'am able to build a whole base with many units, without defence and could ravage the AIs in one tour. Not only a perfect BO, also it needs an improved way to defend itself. Everytime the AI builds the base like a circle, instead of build the base inside of some critical defensepoints with some or less defencepoints on the second or third line. At least that means that an enemy can easily take out the critical defense and rushes in. AA seems to be unaviable, cause VTOL send in the right angle lets them bomb all mainbuildings away.
Also the attacks are not only failing on that repairthing, they're in general absolutely weak. If a player is able to build Archangels, Hellstorms and mobile AA, he ownes the game. I've also never noticed a Multisensortank with some Archangels in its back or combined attacks to clear the enemy defence with heavy, medium and light weapons and bodys. And its horrible, that the AI sends the units right on the attacks, the result is in trafficjam, cause the pathfindingsystem seems a bit worthless, cause i'am so much faster when i'am direct the units on my own.
Another problem is, that the AI isn't "preparing" an attack, i mean that the AI build like 50-70 units and engages then. Instead they're attacking continously. Well, i will try now do discribe my imagination of a good AI.
At the beginning, the AI should analyse the situation (i know that the AI doesnt think like that below
):
How many oilspots are aviable?
How many ways are existing to my base?
Where is the best place to build an optimal covered base?
How close or far is the enemy?
Could i rush him down?
Wich units are most effective?
Wich researchtrees i'am going to use?
Can i interupt the enemy?
And more, but at the moment i dont know anymore for the beginning. The next ist the midgame. The AI needs just too long to reach it. I think i wrote all reasons above.
Should i expand or will small frontbases do the job?
Wich weapons, bodys, propulsion and wich kind of defence (mobile or stationary) is the enemy using?
Can i fit holes in my defence or am i able to increase the overall defense of my base?
What's the matter of life? (sorry, can't hold the joke anymore...)
How to counter attacks? (depends on previous questions ect.)
Where are his weaknesses?
In the lategame the AI should optimize everything again. But thats everything my little brain spits out at the moment. If there's something to add, i will write it to the specific "list" (marked with *added*). Thanks for your time again.
Best regards,
Johnny Ca$h
Well, back to topic, i can agree only in parts. I never noticed devastating VTOL attacks. And that the numbers per group changes, i've never noticed too. Most times, the general AI, attacks with maximum 8 VTOLs with medium or small bodys. Only sometimes i saw more with heavy tanks. Both spotted in lategame. I would prefer not only small changes, that the AI got a pool with more actions, reactions and buildorders. A bit more dynamic behavor would awesome. I've never noticed that the AI is dynamic in any way.
A nearly perfect BO should be a priority. In every game i'am able to build a whole base with many units, without defence and could ravage the AIs in one tour. Not only a perfect BO, also it needs an improved way to defend itself. Everytime the AI builds the base like a circle, instead of build the base inside of some critical defensepoints with some or less defencepoints on the second or third line. At least that means that an enemy can easily take out the critical defense and rushes in. AA seems to be unaviable, cause VTOL send in the right angle lets them bomb all mainbuildings away.
Also the attacks are not only failing on that repairthing, they're in general absolutely weak. If a player is able to build Archangels, Hellstorms and mobile AA, he ownes the game. I've also never noticed a Multisensortank with some Archangels in its back or combined attacks to clear the enemy defence with heavy, medium and light weapons and bodys. And its horrible, that the AI sends the units right on the attacks, the result is in trafficjam, cause the pathfindingsystem seems a bit worthless, cause i'am so much faster when i'am direct the units on my own.
Another problem is, that the AI isn't "preparing" an attack, i mean that the AI build like 50-70 units and engages then. Instead they're attacking continously. Well, i will try now do discribe my imagination of a good AI.
At the beginning, the AI should analyse the situation (i know that the AI doesnt think like that below
):How many oilspots are aviable?
How many ways are existing to my base?
Where is the best place to build an optimal covered base?
How close or far is the enemy?
Could i rush him down?
Wich units are most effective?
Wich researchtrees i'am going to use?
Can i interupt the enemy?
And more, but at the moment i dont know anymore for the beginning. The next ist the midgame. The AI needs just too long to reach it. I think i wrote all reasons above.
Should i expand or will small frontbases do the job?
Wich weapons, bodys, propulsion and wich kind of defence (mobile or stationary) is the enemy using?
Can i fit holes in my defence or am i able to increase the overall defense of my base?
What's the matter of life? (sorry, can't hold the joke anymore...)
How to counter attacks? (depends on previous questions ect.)
Where are his weaknesses?
In the lategame the AI should optimize everything again. But thats everything my little brain spits out at the moment. If there's something to add, i will write it to the specific "list" (marked with *added*). Thanks for your time again.
Best regards,
Johnny Ca$h
Re: Hey there!
There is currently nobody working on improving the current AI, as far as I know. If you want to help out with this, we would be very happy. I very strongly recommend to start by posting smaller changes, instead of starting with huge rewrites.
-
Johnny Cash
- New user
- Posts: 4
- Joined: 07 Oct 2009, 15:42
Re: Hey there!
Sounds cool. At the moment i've no plan how, but i will try. Maybe it's a good way to waste some hours. 
Re: Hey there!
The AI before 2.1, which I studied in great detail and reviewed completely, did that. It attacked in groups, and periodically it did an "all out attack" in which basically all units, maybe even scouts and base defenders, all attacked one target.Johnny Cash wrote: Another problem is, that the AI isn't "preparing" an attack, i mean that the AI build like 50-70 units and engages then. Instead they're attacking continously. Well, i will try now do discribe my imagination of a good AI.