AI question for DAVE

[john connor]

I've been playing Beda Fomm and got bored with the AI doing nothing (enemy AI, I mean), so started to try to remedy this by tinkering with the scenario objectives and stats (results over in the AAR section under my Beda Fomm AAR there). But doing so reminded me of the whole lengthy process of tinkering with my Prokhorovka scenario to try to get the AI to do certain things and I realised that part of my frustration was that I just do not know what the AI can or cannot do, what it is meant to be able to do.

So I wondered if it would be possible for Dave to answer a few general questions about it, to inform scenario design? Only if you have time. Maybe this kind of question could be a break from bug-squashing. ....

This is what I would like the Axis (Italian) AI to be able to do in the Beda Fomm scenario:
1. Advance to its objectives (exit points).
2. When it encounters resistance, probe around to try to find a way through. When it finds a weak spot, mass a concentration of force against it and plot a concerted attack along that axis.
3. When the attack succeeds follow through to the exit points and exit.

So my questions are:
1. CAN the AI do this? Is it actually designed to do this? (Because if not then it's pointless me trying (in scenmaker) to mess around with objectives etc to achieve this).
2. If the AI does probe, find a weakness and try to push through, if it then encounters decent resistance so that it fails to break through, in a scenario of decent length, CAN it then pull back, re-plan, start again probing for other weak spots? Is it meant to do this?
3. Most crucially, because I hardly ever see it happen - is the AI designed to concentrate force and plan significant attacks with a concentration of forces? In many scenarios now I have seen the scenario virtually halt without the AI putting in attacks, once it has run into an initial resistance. But maybe it's not even programmed to do this? If so, that would be good to know. So, in Race for Bastogne, for example, (about which I did an AAR over in Matrix) the Allied AI at end of scenario had lots of forces available to plan an attack in a concerted fashion, but it never happened. Attacks happened, but they were always small (more like probes). Certainly never larger than Bn level efforts. In Spearhead v Reich the whole thing similarly ground to a halt and when I surrender to take a look I find the Axis AI (in that case) has masses of forces, but is not mounting any significant attacks. Mostly I see this kind of thing in larger scenarios.

I ask these questions not to criticise the AI (which is still, without doubt, the best out there) but to try to work out better what it is MEANT to be able to do and how we are meant to try to get it to do that via the objectives and stat tools available in scenmaker.

So I've just messed around with Beda Fomm, for example, to try to get a more 'active' AI. I gave it as Axis a choice of multiple exit objectives, all linked, and maxed out to 100% all the unit and leader stats for the Axis. The result so far is that this succeeds in making it consider choices, makes it probe around for weak spots, but it still doesn't seem then to step back, pause, plan and put in significant planned attacks where you see anything larger than a Bn involved. Is the enemy AI capable of mounting regiment sized attacks, for example?

4. In scenmaker, how do I maximise the chances of the AI putting together concentrated attacks, using the significant forces at its disposal? Do I put as many points as possible on killing the enemy, with as few points as possible on objectives, for example? How do I give the AI leeway to plan attacks? Is there a way?

Thanks!

Peter

 

[Wiggum]

I would like to add a general question:

How good is the AI in the offensive ?
In most wargames you play as the attacker most of the time just because the AI sucks in anything other then static defense.
How good is the AI at counterattacking and how good is it at the offensive after all ?

 

[john connor]

I think the AI has changed on the offensive since I first played this game - say over a 4 year period. I just played Manhay again as Allies, for example, and it was extremely easy to occupy the main objectives, which were contested by small attacks by the AI (Manhay is a bit like a meeting engagement, but the Allies are closest to the key objectives, so the Axis as AI has to attack constantly) and once those attacks ran aground nothing happened. It was very disappointing. I am finding this in almost every scenario now. Have a look at the last pages of the AAR I've been doing on Beda Fomm, Wiggum, (over in the AAR section) if you want to see how frustrated I've become with being unable to get the AI to attack. It does attack, but only ever, imho, to single Bn strength (and then not often - more often it's just single units sent forward, more like Probes than attacks), even if much more is available. Once this beta process is settled down I was planning to put pressure on Dave about this, in the nicest possible way, because it discourages me from playing, especially larger scenarios, or anything where the AI has to attack. There are problems with the AI in defence too, and Dave has already said he's open to developing this, when there's time. Mostly, I find myself wishing we could script attacks and defence in scenmaker.

Most often, however, when I raise these issues, others are quick to point out that the AI is alive and attacking. So maybe it's at least partly my interpretation of what's going on. After all, it's all read through the FOW screen. Above small probes and the occasional Bn level attack, I haven't seen anyone posting proof of coordinated AI attacks though. I would love to be proved wrong , and await the pictures....

Meanwhile, with all the bugs to iron out in CO2, I doubt Dave will find time to answer the above post in any detail. But maybe later in the year.

Peter

 

[Dave 'Arjuna' O'Connor]

Meanwhile, with all the bugs to iron out in CO2, I doubt Dave will find time to answer the above post in any detail. But maybe later in the year.

Peter

Alas that's true Peter. There is little enough time in the day at the moment just focussing on ironing out the bugs. Once I have done so I am more than willing to engage with the AI discussion.

 

[Daz]

No need to reply to this Dave, I know your busy.

I have my own theory on why the AI may be having problems pulling off a concentrated attack:
There is quite a serious bug, that has been in the game for a while but as a player I know how to work arround it.

If an attack gets bogged down instead of the AI doing a sensible replan or bunkering down, the attack turns into a normal move to the attack marker in road column formation, which is the worst possible action to this particular problem.
As a player I can counter this by keeping an eye out for it and giving a new order when it happens.
If the AI is not able to escape from the bugged action, this could be causing problems.

Another problem could well be the inability of the AI to push through an attack once one of the units in the attack formation takes to many casualties and goes into the retrat recovery mode.
I have seen it happen on numeorus occasions where one units info icon will turn orange, then the whole formation instead of pressing the attack with the uninjured units, it will lock up instead. Often with the result of the uninjured units being left in the open waiting for the retreat recovering unit to get its act together.

I think the AI needs to be much quicker in dropping the retreat recovering unit from the formation, and allowing the other units to continue to push the attack home.

 

[john connor]

You know what, Daz - you may be onto something there. Because how else would I have such a strong perception that it had changed? It must be something like that - a bug, or several of them. Lock ups, and the bug you mention - all these could cripple the AI as they cripple my own ordered attacks, but like you, I watch out and correct.

 

[Kurt]

Only been playing CO for a year but i have noticed the same thing happening in my games , I have had to micro-manage all attacks like a meddling fuhrer

 

[Daz]

Another possible problem is the length of time and the randomness of refuelling.
In the Beda Fomm scenario, I observed units dropping out of formation to refuel, but after the refuelling event (which took hours) the unit never re-joined its formation, but stayed frozen to the spot.

This may just be an issue from the players side of the game, but if it is effecting the AI enemy side as well, it could be another thing to look into.

On a side note I personally hate this randomness, and would much rather be able to order a refuelling event at a time that was convenient.
Some form of manual override would be lovely, as well as some indication as to how much more you will get if you select max fuel resupply in the Edit Tasks Dialogue.
This would be especially useful to fill up your tanks before a deep penetration or an attempted encirclement.

 

[simovitch]

Those Italians trying to escape through Beda Fomm aren't the best soldiers for the AI to try and formulate a fluid strategy with, and the on-map boss is fairly overloaded as well. Try tinkering with troop quality and see what happens. I didn't get much of a result either during development of this scenario.
The AI has a tough time mastering the Army size forces, but these Italians didn't have much luck historically either...

 

[Dave 'Arjuna' O'Connor]

I do recall sometime ago changing the code to ensure that assaulting units did not keep pressing the attack if a certain percentage of the assaulters are reacting - ie retreating or recovering. This probably needs to be tweaked.

 

[Daz]

Those Italians trying to escape through Beda Fomm aren't the best soldiers for the AI to try and formulate a fluid strategy with, and the on-map boss is fairly overloaded as well. Try tinkering with troop quality and see what happens. I didn't get much of a result either during development of this scenario.
The AI has a tough time mastering the Army size forces, but these Italians didn't have much luck historically either...

Funny you should mention that
Peter has been doing that exact same thing with reasonably encouraging results.
Look at his Beda Fomm AAR in the AAR section to see more.

 

[Daz]

I do recall sometime ago changing the code to ensure that assaulting units did not keep pressing the attack if a certain percentage of the assaulters are reacting - ie retreating or recovering. This probably needs to be tweaked.

Thanks for the reply Dave.
I think it would be a good idea if you get a chance.

 

[Daz]

Another thing that I think is locking up the AI just as it does one of our attacks is the stubborn HQ problem.
Dave is aware of that and looking into it when he gets time.

 

[Iconoclast]

Another possible problem is the length of time and the randomness of refuelling.
In the Beda Fomm scenario, I observed units dropping out of formation to refuel, but after the refuelling event (which took hours) the unit never re-joined its formation, but stayed frozen to the spot.

Out of curiosity - Has the refuelling problem been taken care of? I can't recall. Just played Beda Fomm and stopped after my units would just not refuel.

Cheers

 

[Daz]

The Beda Fomm scenario is quite an intriguing situation as far as resupply goes, if you look into it in depth.
The Regimental Base, " Raggruppamento Bergonzoli Base" starts the scenario as the only supply for what is probably a Brigade size formation, all of which have fuel hungry vehicles, and are having to travel great distances.
It starts the scenario badly stocked and is working on the foot terrain tables, using horse and cart primarily as a means of transporting its supplies.
In order to get your units any timely resupplies you have to move it as close as you can to them, as the trip up and down the road at horse and cart speed would take many hours per journey.
As it is moving at walking pace it takes a long time to move the 20 or so Km down the main road.
All the time it is moving it can not resupply.
The lack of user feedback and user interaction as to what is happening with supply, is still a major issue in game, in my opinion.

To summarize.
The early part of the Beda Fomm scenario is very much about trying to make the most out of a very bad tactical and supply situation, for the Italian forces.
Over vast distances with inadequate resupply vehicles.

Warning. Spoiler below this point.

I come to the conclusion that the best tactics is to push through what you can to the exit point as fast as possible the first night but instead of exiting them, make them form a defensive perimeter around the exit objective. This will split the Allies effort as they now have two different areas of operation with enemy on both sides of their lines, albeit with one of them in a pocket cut off from supplies.
The pocket around the exit objective only have to hold out for a couple of days though, but remember to set ammo usage to min.

Then muster your forces over the first few days to build up a healthy supply base and attacking force. Allow the Allies to beat themselves up against your defensive line, then switch to the offensive nightfall of day 2, after your last lot of reinforcements arrive and try to push through the weakest points in their line towards the exit.
If you don't make it there with your attacking force, remember to exit what you can of the defensive perimeter around the exit objective.

Also remember that half your points are for destroying the enemy in this scenario, so rather than feeding your men into the grinder piecemeal, let them come to your grinder until you have enough well supplied units in place to crush them.

 

[Iconoclast]

Thank you Daz for this in depth explanation! But I haven't read the spoiler ;-)

 

[john connor]

Read it! It's good stuff!

 

[Daz]

I love your blind faith in me Peter

I haven't actually completed the scenario using the tactics I mentioned above.
I have only played it all the way through the once and that ended in a draw, as it was before I realised the critical nature of the supply situation, and tried to conduct a wide flanking move through the desert, where I promptly ran out of fuel.

My second attempt is the screenshot above where I intended to use the tactics I mentioned, but never finished it because I think it was when we had the problem with the Intel bug and I got frustrated and abandoned it.
Also the bases for some reason just wouldn't move down the road. They just got stuck about half way, that was when they were drawing fuel and basics from the base above them instead of their own supply, so maybe that is fixed now.

I might have another crack at it to see if my plan outlined above will work

Would love to hear how you get on with your game Iconoclast.

 

[john connor]

Oh. I thought you'd played it through like that, Daz. Sorry. It sounded like a great plan.

Peter

 

[Daz]

Well I think its a great plan
Guess we will have to try it and find out