AI observations

AI observations (Full Version)

All Forums >> [New Releases from Matrix Games] >> Strategic Command Series >> Strategic Command WWII: World at War

Message

Hartmann -> AI observations (7/24/2019 11:41:55 PM)
I’m playing my first game of WAW where I only control Japan. This gives me a privileged view of the German and Italian AI. Here’s some things I noticed (until December 41) which might be useful for devs to improve on it: - German AI (Europe, land): Poland and Fall Gelb is ok. AI later tries to prepare for the Balkans and Barbarossa simultaneously where it seems to have some decision issues (moving some stuff back and forth). In my game, war on Greece was declared late and Germans had to fight Greece while Barbarossa started. One particular thing of note is that they (and Italy) committed too much stuff to Greece and even still moved armies south when Athens was already taken so that it would have been wise to already begin relocating troops to the Ukraine. Barbarossa until winter was a hazardous adventure. While steadily advancing, Germans haphazardly attacked fortified positions (instead of going around them and picking them off later when they would have been out of supply). They also preferred using tanks against cities (which I think is not recommended in this game?). Finally, they failed to make sure to guard conquered cities against counterattacks and so they lost valuable time having to reconquer Odessa which a lone Russian unit coming along the Black Sea coast took back totally unopposed while the Germans were already busy getting to the Dnepr. - Italian AI (land): They neither bolstered Abyssinia nor Lybia before going to war which I think is a big mistake. - German AI (Africa, land): This was an utter desaster. When the Africa corps arrived, Tobruk was about to fall. Germans dashed forward in a most uncoordinated way (HQ first!) and were picked off by the British easily one by one. Rommel HQ survived, though, managed to fall back behind the line of its units and at one point even was in port ready to transport out, but for some reason they didn’t and thus were doomed. - German/Italian (naval): At first they seem to have ok scripts – before the fall of France, Germany kept their fleet in the Baltic sea, Italy theirs in the Adriatic. But once France fell, they went absolutely yolo, losing most of their ships in short time. There’s one special thing I noticed which really should be addressed: Germany still had a couple of CA when Barbarossa started. They went to the Baltic, sunk the Russian ships, but one CA was badly mauled by the Russian sub (from 10 SP down to 2). But instead of fleeing, the German ships now tried to “surround” the sub. They then were mauled again and again by that lone Russian sub – none of their ships were actually sunk, but they must have lost like 30 SP by that “strategy” until now. I will continue to use this thread to report further observation about AI behaviour (doing Allies next game).

Mercutio -> RE: AI observations (7/25/2019 12:05:55 AM)
Nice feedback. I never have the AI control any forces on my side. I find the AI does a decent job when they are against you though, well until 42-43. When you give other nations the AI does that include research? Just curious.

Hartmann -> RE: AI observations (7/25/2019 12:30:11 AM)
quote: ORIGINAL: Mercutio Nice feedback. I never have the AI control any forces on my side. I find the AI does a decent job when they are against you though, well until 42-43. When you give other nations the AI does that include research? Just curious. Yeah, that does include research. I could interfere (the game does allow the user to do that for AI controlled nations on the same side), but I leave them totally alone. When I played WIE, the only challenge remaining after a while was to only control a single country so that's why I decided to start out that way with WAW right off the bat.

Hartmann -> RE: AI observations (7/25/2019 1:22:33 AM)
Graph shows a whooping MPP loss spike of 1800 MPP for Germany in the winter strike turn! Same turn Russians only lost about 700 (while they peaked at 2600 during a turn at the beginning of Barbarossa). Interestingly, the Germans seem not at all fazed by the losses and instead of replenishing their forces and preparing for spring 42 they still attack throughout the winter - which leaves many units at strength 4-5.

Hartmann -> RE: AI observations (7/25/2019 5:57:10 PM)
German AI observations (continuation): The good thing: Already when it became clear that the war in Africa would be lost, German engineers began building defensive fortresses in Italy. They also put three corps there. The bad thing: In Russia, Germany didn't prepare in any way for a 1942 spring or summer offensive. As they attacked in bad conditions all through the winter, their strength points are depleted, plus their forces aren't positioned adequately. For example, there are three (!) artillery units still near Brest-Litowks where they can do zilch. Panzer spearheads are still all over the place with strength points down and sever supply problems - some are even in danger of getting annihilated. We're still in April, but there is no dicernible "Schwerpunkt" setup anywhere for summer ...

wevilc -> RE: AI observations (7/25/2019 6:50:51 PM)
Well, the AI’s conduct of the US Pacific campaign is much worse than that! They either show up at Java or off the coast of Japan and in either case walk right into disaster., time after time There is no preparatory conquest of the outlying islands. The US needs naval bases with decent supply. Fighting scross the Pacific from bases back home is hopeless.

Hartmann -> RE: AI observations (7/25/2019 7:36:54 PM)
quote: ORIGINAL: wevilc Well, the AI’s conduct of the US Pacific campaign is much worse than that! They either show up at Java or off the coast of Japan and in either case walk right into disaster., time after time There is no preparatory conquest of the outlying islands. The US needs naval bases with decent supply. Fighting scross the Pacific from bases back home is hopeless. Too bad, but I was beginning to suspect something like that as programming a decent naval strategy seems quite a bit harder for some reason. I will probably go back to WIE soon as with the patches since release the allied AI naval invasion performance (and the AI's performance in general) was much improved. When I came back to WIE recently (hadn't played the game since release), I was very pleasantly surprised about that. I began with a quite sloppy Axis campaign with a belated Sea Lion in 42, and (after initial success) the Allies managed to mount counterpressure and eventually I had to evacuate. They then did a quite cohesive landing in the Normandie later too.

otumfuo2 -> RE: AI observations (7/27/2019 12:57:39 AM)
One more issue with Italian AI. It's easy to get the Italians to send out their fleet while France is still unconquered. If the Allies have moved the whole French fleet to the Med, along with 2-3 British ships (DD,CA,and BB or CC), then it's easy to wipe out the Italian fleet with the French taking the bulk of the losses. Then the Allies blockade all the North African ports and Afrika Corps deploys in Europe. With Afrika Corps out of the picture, it's easy for the Allies to conquer all of North Africa, and that ultimately makes it possible to get back into Europe earlier. It seems unlikely that the real Italians would be so cavalier with their fleet.

wevilc -> RE: AI observations (7/27/2019 1:08:07 AM)
quote: ORIGINAL: Hartmann quote: ORIGINAL: wevilc Well, the AI’s conduct of the US Pacific campaign is much worse than that! They either show up at Java or off the coast of Japan and in either case walk right into disaster., time after time There is no preparatory conquest of the outlying islands. The US needs naval bases with decent supply. Fighting scross the Pacific from bases back home is hopeless. Too bad, but I was beginning to suspect something like that as programming a decent naval strategy seems quite a bit harder for some reason. I will probably go back to WIE soon as with the patches since release the allied AI naval invasion performance (and the AI's performance in general) was much improved. When I came back to WIE recently (hadn't played the game since release), I was very pleasantly surprised about that. I began with a quite sloppy Axis campaign with a belated Sea Lion in 42, and (after initial success) the Allies managed to mount counterpressure and eventually I had to evacuate. They then did a quite cohesive landing in the Normandie later too. What is W1E?

Mercutio -> RE: AI observations (7/27/2019 1:16:56 AM)
War in Europe

wevilc -> RE: AI observations (7/27/2019 9:42:58 AM)
quote: ORIGINAL: Mercutio War in Europe Thx

Hartmann -> RE: AI observations (7/30/2019 5:03:02 PM)
Further Observations: 1)German AI couldn't take Sevastopol because they didn't make use of the fact that two adjacent units reduce a resource (and therefore supply) by 1/level. Edit: I was wrong insofar as the rule does not apply to fortresses. Still, the AI should probably be either more determined when attacking Sevastopol or leave it alone - otherwise it's just a MPP sink. 2)German AI really should be reprogrammed to not attack towns with Panzer units - ESPECIALLY not if those are significantly understrength. They lost 2 panzers at Narva and Novgorod because of this (as the Russians just mopped them up on their turn). 3) Allied AI did a rather half-hearted invasion of southern Italy from Lybia with two units. They are now trying to back this up with like 5 additional transports, but it may be too late to save those two advance units which are starving. 4) Allied AI took Rabaul from me even though I had a special forces unit there - nice. :)

Hartmann -> RE: AI observations (7/30/2019 9:55:07 PM)
It seems to be a thing that Germans always attack through winter - they did that in 41/42 as well as 42/43. Russians do that too, though not as extensive and with slightly better results. I also noticed (somewhat belatetly) that the behavior of the Fins in this campaign is strange. They joined as soon as Barbarossa started, but failed to take their forward fortified positions, which allowed the Soviets to move there. Since then the Fins stayed at Helsinki and vicinity. As the Soviets chose to leave them alone, the theater north of Leningrad was quite a non-factor thus far. There was a strange thing happening at Java: I moved my paratroopers next to Batang where the Dutch had an air unit stationed in the city. Right next to the air unit was an infantry unit. Now what they did was this: They attacked my paras and then moved the air unit away (to save them). In this, the AI didn't realize that this would free up the city so that I could occupy it next move (putting DIE out of the picture). What the AI should have done was to move the air unit away first, and then move the infantry into Batang.

Mercutio -> RE: AI observations (7/30/2019 10:54:32 PM)
hmm, did you invade next to it or farther away and move in? When I invade there is a unit of infantry there, not a plane. The critic is still valid though.

Hartmann -> RE: AI observations (7/31/2019 9:53:37 AM)
quote: ORIGINAL: Mercutio hmm, did you invade next to it or farther away and move in? When I invade there is a unit of infantry there, not a plane. The critic is still valid though. I invaded belatedly in Winter 42. I used paras from Borneo to get to Sumatra. I took Palembang and the nearby oilfield. It then was revealed that there were fighters in a city further north. The infantry unit that is in the capital in Java at the start of the game moved towards Palembang when I moved my unit north, and it seems that the fighter unit moved to the capital later. Anyway, when I realized that the game assumes a "land connection" between Sumatra and Java, I moved south and that's when I encountered the situation at the capital. There was another AI quirk involving planes a bit later. I had captured O'ahu (Hawaii) some time ago and decided to get the other island too (because of abstraction I don't exactly know which island the second one is supposed to be). It had a special forces unit in its city. Anyway, when I had landed my first amphib unit next to it (second one was to follow next turn), the AI decided on its turn that it would be a good idea to transfer planes (coming from I don't know where) to the island (just one hex away from my forces). Of course I obliterated them on my next turn.

Hartmann -> RE: AI observations (8/2/2019 5:19:37 PM)
It's March 43 in my game, and the first time I think I see the German AI preferring to consolidate their position in Russia instead of attacking without a plan. Is there a script that tells them when to go on the defensive? US has occupied Peleliu unopposed and also tried to take back Hawaii. For this, they sent 4 LR amphibs which were totally unescorted! Big mistake as my 3 battleships and 3 carriers at Hawaii sank them all. I think the unit which took Peleliu also hadn't any naval escorts. I sank their Pearl fleet in December 41, but haven't sunk any of their carriers as of yet, so they should have enough naval assets to escort their invasions. If my observations are correct (and not a fluke), then US naval AI definitely should be improved to give Japan players more of a challenge.

Hartmann -> RE: AI observations (8/2/2019 8:35:25 PM)
I realized that the Germans have advanced tanks level 4, but only one unit is actually upgraded - all others are only 3 or 2. Despite having enough MPP, they seem to not come around to that because they won't take away their tanks from the frontlines. May 43: Allies have pulled off a successful early invasion of France - I like that. Let's see how the German AI handles this situation. They should urgently pull back their forces in southern Russia behind the Dnjepr, but I fear the AI will not realize that and be crushed in a few turns. In this game, the various shortcomings of the German AI led to two desastrous developments: First, in southern Russia, they never got anywhere near the Caucasus or Stalingrad (thus missing out on valuable resources). Second, in Africa, the incompetence of the AI in handling the Africa corps led to its early destruction. (Italy's AI had a hand in this too as they never transferred any units to Africa before their entry into the war.) It will be interesting to see how my Japan fares when being "last man standing".

Mercutio -> RE: AI observations (8/2/2019 10:04:44 PM)
To be fair, the Japanese/US Pacific theater is nothing like the War in Europe. There are so many possibilities and such vast expanses it would be hard to remotely script it. I doubt there even is a plan of "retake Pearl Harbor" Now one improvement I would recommend is the AI tends to focus on things it can damage the most. So I have seen carriers attacking subs and destroyers, then getting their rear ends handed to them on my turn. One thing I have noticed is if you are beating the ai it stops researching or does very little. Especially with planes. UK and US research fine. USSR routinely in my current game is having fighters die in droves and then reinforces them instead of research and upgrade. This leads to more air loses and more MPPs down the drain. It also needs to ground its planes at times IMO. That is just one example, but you get the idea. Upgrades are better in the long run than more fodder for me to kill. Also amphibious invasions are tough enough. However the AI will blindly send units from the UK to France and get annihilated. Then try again. I saw it coming when all of the sudden the US and UK navies both jammed the channel. I was winter in Russia, so I just operated all my bombers over and had a couple in the queue. I also had something like 9 level 4 subs and was booking the get naval weapons 2, which I got not long after. My subs and bombers crippled and sank the capital ships. Then while the destroyers and carriers tried to sink my subs (good luck with a few units at level 2 or lower) my planes and capital ships took out the destroyers. I then started picking off amphibious units. The AI landed anyway. It should have probably reconned and seen it probably wouldn't go well. Anyway that got wiped out, although I think 1 unit managed to land in the SW and take out a damaged fighter with the its armor. However it now had zero supply, so it was blown away next turn. All in all I think the AI lost 7-8 ground units landing in a couple of waves. They went through my subs (go figure) and landed. There were probably 6 or so more sunk in the channel. Maybe there should be a free recon x hexes based on your air range rating? Most recon wasn't send a wave of bombers over there and see what happens. It was scouted ahead of time. Perhaps you get to reveal one area per air unit? Just throwing out ideas. Anyway, there is a reason the Allies waited for D-Day and game with overwhelming force. The AI should gain air superiority before it rushes the channel IMO. I don't want to sound negative, it is a very well made game. I, like I believe Hartmann is, am trying to suggest how to make it better.

Hartmann -> RE: AI observations (8/3/2019 12:25:11 AM)
Yeah, the whole thread is meant to provide clues about what could be improved about the AI's behavior (and not meant to put the game down or something). Also, I have to say that when I think back to games like "Storm across Europe", "Clash of Steel" etc ... AI programming for WW2 grand strategy games came a long way since then. But of course it should not stop here. :)

Mercutio -> RE: AI observations (8/3/2019 12:48:52 AM)
Personally I think the easiest country to play is Japan, which it should not be IMO. Italy just sucks, period. Germany needs to be played correctly to hit that tipping point. There is just too much going on with Japan, US, Australia to program that remotely easily. Hell, I am a programmer and it isn't this complicated with a multimillion dollar program. Usually programming is "Have an algorithm to do X". In simplistic terms, and it is not simple, think of Excel and all its options. At least they are fixed. AI vs a human? Pfft.

nnason -> RE: AI observations (8/3/2019 11:33:05 AM)
I have said before but the developers should have fielded a SC3 WiE and a WitP and had a way to join them into one game with two scales and two timelines and two AIs. The mechanics would for both theatres but the scales and timelines would be better served if they were kept separate. This would have allowed a scale for the West and the East that better approximated the difficulties each theatre had. And a less complicated AI. For a game as ambitious as SC WaW the AI programming is just well too difficult. There are just too many variables. I have a longer post on how to join. A joined game would only be played by us die-hard wargamers. Maybe a project for the future. Both games are great but we should remember WiE has had years of gamer input and great developer support.

Mercutio -> RE: AI observations (8/3/2019 12:45:35 PM)
Both games have great support. Just spitballing ideas here for hopefully discussion to help the developers. For the US, I would recommend the AI work on systematically taking back islands (including Pearl Harbor if needed) Taking back Midway if it was lost frees up the attacks on Pearl Harbor supply Then the Islands in the Gilberts and Marshalls Wake Guadalcanal - prevents Australian supply raids Rabaul You get the idea. It doesn't have to even be in that order every time. Instead of a script to say "Do this now" have one that is a logic tree to follow. A series of if this, then do that. Boolean rules for those steps. Something like naval power of X and and special forces Y then attack the target(s) That way it forces Japan to cover more area. Offers bases for maritime bombers to scout and sink any Japanese counterattack. Actually I would also like these islands to have an area of reveal, based on long range aircraft research, as they used scout planes extensively. Really FOW should be reduced as planes get better range. This reveal range can be reduced by enemy fighters perhaps. Trying to take the DEI early is destined to fail. Trying to invade Japan is likely suicide if these other areas and the Philippines, Guam, Saipan, etc. are not taken first. I think Australia should have cities with more supply. Maybe even 8 to help its defense. If Japan takes a city it is reduced to 5.

Hartmann -> RE: AI observations (8/3/2019 4:01:47 PM)
It's June 43 and I'm disappointed how the German AI has handled the situation arising last turn. They should have realized that they HAVE to counter the Allied invasion of France and should have operated troops from the Eastern front to France. But they didn't even operate one single unit! Also, they did not realize they have to assume defensive positions in Russia. They still in vain attacked against vastly superior forces and weakened themselves to a degree that in the next turn they lost another couple of panzers. In fact, this turn was their demise as next turn they will be overrun on all fronts. On the other hand (not having played WAW before), I was surprised by the American/British naval operations at Java. I KNEW they were coming, but I thought they'd come with amphib landings. Instead they sent a big fleet of 3 battleships, 2 heavy cruisers, 4 destroyers and 2 carriers. (I spotted them mainly with my subs first and later in battle. There MAY be further naval units hidden behind those I spotted.) I had one carrier, 1 escort carrier, 1 battleship, 2 heavy cruisers, and 2 subs there - which I thought would suffice. Man, was I wrong. All ships were at full strenght, execpt my 2 cruisers were at 8. I have naval weaponry 2, advanced fighters 3 and long range aircraft 2. They got the same upgrades except they only have advanced fighters 2. Using my subs and my carrier, I still could just sink one carrier on my turn (I was the one to spot and to strike first). As their other carrier (providing escorts) was down to aircraft strength 5 by then (mine was down to aircraft strength 3), I thought I'd survive next turn easily (even though I was at naval/tactical and could not adjust this to cap for next turn, but my escort carrier was providing full cap). But then, on their turn, their carier sank mine in two strikes and their 3 battleships and 2 heavy cruisers sunk both my heavy cruisers and damaged my escort carrier badly. Next turn I was not allowed to attack their "hidden" remaining carrier with my subs, even though I knew where it was and inched towards it - when I moved next to it, my strike allowance went down from 1 to 0. I'm in full retreat now (with my battleship and escort carrier) and hastily try to get the bulk of my fleet over from Pearl (though it still has to be replenished).

Hartmann -> RE: AI observations (8/3/2019 8:21:11 PM)
July 43 and the German AI still doesn't think it necessary to defend France. What is it with them? Is it because it's 43 and they only have scripts regarding D-Day in 44? Or does it somehow have to do with Spain being in the Axis tripping them up? Actually, they DID move one unit to France and that's Spanish HQ Franco - which doesn't seem to lend supply to the lone German corps in Paris, though. Because meeting with no opposition, the Allies have already liberated about half of France since they landed last turn. In Russia, the German AI seems to act like under a "stand and die" order - they still needlessly throw their strength points away by futile attacks on their turn. They lost a whooping 11 land units this turn, mostly on the Russian front and got their last tanks surrounded. Meanwhile the Allied navy controls the waves in Indonesia, but don't seem to follow up with amphibs as of yet - except for one single unit they landed in Sumatra last turn already. This COULD be good AI planning - making sure that naval superiority is achieved before following up with an amphibious invasion. Let's see.

Hartmann -> RE: AI observations (8/4/2019 4:31:55 PM)
In my last posts, I noted that the German AI seems not to be willing to counter the Allied invasion of France. Well, it's now August 43 and they did something. HOWEVER, they immediately transferred troops to the Sigfried line, i.e. preparing to defend Germany instead of France. Which I think is a strange decision as the Allies haven't even taken Paris and it would have been a short and defensible front from their coastal gun at Le Havre over Paris down to the border of Vichy. Hadn't they hesitated this long, they even could have driven the Allies back into the sea. In Russia, there's still that "stand and die" attitude, they lost another tank and now only have one left. In the Pacific, nothing changed: Allies took one island (Mataram) unopposed and prowl/control the Javanese sea. But for now there doesn't seem to be a big invasion at hand. While this is not an AAR: Preparations of my naval counter-offensive take a while because I got too few big harbours for upgrading and replenishing all my ships (plus I just got long range aircraft level 3, so I want my carriers to have that). I set an amphib transport at sea to take back the Marianas as I don't want them to station bombers there.

Hartmann -> RE: AI observations (8/5/2019 8:21:33 PM)
German AI does a competent job manning the Sigfried line, but a horrible job retreating and forming a cohesive defensive line in Russia. Allied AI invading France did a strategically sound thing in declaring war on Vichy - which provides them with more room to maneuver (and better access to Axis Spain). One thing I noticed in this context was that Vichy doesn't have a garrison or corps in its capital, only in the South. This led to immediate surrender because US troops could just move in same turn. Maybe Vichy should get a corps on Vichy itself. Allied AI in the Pacific finally invaded Java with three units, so PROBABLY they first triedto make sure that they have naval superiority before committing troops (which of course is great). I just wonder why they didn't do the same when trying to reconquer Hawaii (where they sent 4 unescorted amphibs to their doom). While I still have two carriers upgrading in Japanese harbours, I'm now about to engage the Allied fleet next turn with the bulk of my fleet (which I cruised to somewhere just north of Saigon). I also packed two armies on transports which trail the main fleet. I plan to get them to port in Sumatra (hopefully before the Allies take over).

ThunderLizard11 -> RE: AI observations (8/6/2019 1:58:02 PM)
I haven't played SP since the last big patch so wondering if US naval AI has been improved. I found in 1.02 that the US AI would send out their fleet to the Solomon's and just sit there. It was way too easy to wipe out the CVs. etc. Has this improved in 1.03? Unescorted amphibs is an issue from WiE as well which hopefully devs can address.

Hartmann -> RE: AI observations (8/6/2019 3:29:19 PM)
I have not enough experience with WAW to say whether Allied naval AI was improved. It seems they try to escort scripted invasions, but not unscripted ones. On the other hand, I played WIE at release for a time, then departed, and then came back to it recently. I found the AI was noticeably improved overall since release and US/UK escorted and handled their D-Day invasion well enough.

Hartmann -> RE: AI observations (8/9/2019 6:15:19 PM)
As 1944 rolls around and Germany is losing, it reall shows that the German AI (and by extrapolation the AI in general) needs reworking for when it's on the defensive. First, it needs to realize when it's time to switch to defending instead of still trying allout attack even though the tides have noticeably turned. Second, it needs to learn how to fall back in an orderly fashion behind well defendable (and shortened) defensive lines - behind rivers or on mountain ridges. Thirdly, when the enemy breaks the defensive line, pouring through it in force, the AI needs to try and plug the gap instead of clinging to forward positions and getting surrounded. Fourthly, the AI needs to pull their aircraft and HQ away from positions where the enemy broke the defensive lines. ############## If the AI could be made to learn these things, it would be MUCH improved.

taffjones -> RE: AI observations (8/9/2019 9:27:23 PM)
Hi Hartmann Is that a observation of the tactics of the AI Or AH the Furher [:D] (Sorry this reply is intended as a joke) I'm sure Hubert and Bill are working on it as they have on WiE [:)]

Page: [1] 2 next > >>

Forum Software © ASPPlayground.NET Advanced Edition 2.4.5 ANSI

1.59375