An AI observation

[Tom Stearns]

Have played Eindoven and Nijmeagan scenarios twice each. Once each for each side. When I play as the Allies, it seems that the German AI attacks piecemeal with his formations. When I play as the German the Allies seem to coordinate formations better. When I play as the German I hold back formations until they are complete then attack. Having done this as the German I barely lost in the Eindoven scenario and achieved a marginal victory in the Nijmeagan scenario.

I'm still learning nuances of the game, but was just wondering if anyone else noticed this trend.

[Arjuna]

Tom,

There are a number of factors at play. Coordinated attacks are more likely where there is sufficient time "perceived to be available". Remember that those green formations of the German 406th Div take forever to mount a complex attack. These formations are also not likely to maintain an attack once underway. A few routers ( and let's face it most of their unit do ) and their HQs are likely to call of the attack and bunker down. This has a large bearing.

Another major factor is the timing of arrivals. The Allies by and large arrive in Bde sized groups. The Germans in Bns. Hence it is more than likely that they are thrown in piecemeal, just as they were in reality I might add. When we review the strategic AI I'll look at this aspect. I agree that they may be better served holding back for a bit and then attacking in Bde strengths.

[Black Cat]

Just re-read " It Never Snows In September" by Robert Kershaw to get the "Feel" for the German viewpoint while playing HTTR and Dave is quite right, they( Div. 406 ) were thrown in as they arrived, as were Von Der Heydte`s FJ Rgt. 6 against XXX Corps near Neerpelt.

I would wonder about making the Germans any stronger and/or more organized then they are now. The Balance seems right on IMHO.

[jcj]

I recently finished the Arnhem campaign where the British are given the Airlanding Brigade plus the hurried up drop schedules and better dropzones. I believe this is a scenario where the British should be able to hold their own. Without too much monkeying around I was able to score a decisive victory against the AI which played as the British. I was careful not to commit my own German troops into positions I knew the British would come from because of my knowledge of history.

In this particular scenario the British AI did not adequetly construct a line and did not coordinate its attacks. As it was attacking toward the bridge no provision was made to defend either its left or right flanks though it had more than adequate forces. The city is broken down into two sectors in my mind. The first is the approach from Oosterbeek and the second is the approach from the airfield. In the middle there are some woods which form a natural division between these two sectors. There is also the approach from the East but this sort of blends with the approach from the airfield. The AI attacked toward the bridge from the airfield approach both next to the woods and farther to the east. The AI decided to launch these two attacks of brigade strength each at separate times.

Because the AI decided not to protect its flank facing toward Oosterbeek, or better yet attack the bridge from this approach I was able to roll up his second brigade stength attack with an attack against his flank. Later I was able to attack from the other flank as well. Meanwhile the AI held a large portion of the Airlanding Brigade in reserve very nearby. The AI launched the third untouched airborne brigade against my line later to no avail. While this was happening the Polish brigade stuttered between attacking the rail bridge and the main bridge, finally deciding on hanging out somewhere in the middle.

In certain circumstances it is wise to deviate from the creation of continuous lines. But having thought about my own actions as a virtual commander, it occurs to me that the first thing I do is create some type of line. There may be gaps in this line, but in some way those gaps are at least screened with forces earmarked to cover those gaps should they need to be covered.

I think the AI would be much improved my managing its lines more effectively. In my own mind I always draw some type of virtual line, and then assign forces to it. Sometimes just one or two units at a critical roadblock will be all that I can manage, othertimes maybe I can assign a whole brigade. But, there is always some type of line, when there is not a line the goal should be to use forces to create a new line either in front of or behind their current position. While all this talk of lines may seem a little conservative, I think it is stepping stone to having a better AI.

I will complement the AI on one manuever that while doomed to failure was certainly bold and creative. The AI managed to sneak an entire brigade from XXX corps over a bridge and then proceeded to move toward the main Arnhem bridge from the Oosterbeek approach. I say doomed because there were no paratroopers left and only a small amount of units from the Airlanding Brigade. But if the attack would have gone better for the paratroopers, the arrival of these forces would have been quite unsettling.

[Tom Stearns]

I formed the 406 between Boxtel and Best running from the RR across through the woods just north of the 101st DZ. After all units had arrived I launched a complex attack against the DZ and over ran it. I took the VP objectives along the main highway NE of the canal. If not for arrival of 30 corps this attack by the 406th would have won the game. In fact I will try scenario again and try to hold less ground after the attack and hold the 406th together. Like you said, they easily routed when they faced 30 Corps.

I know when developing the AI it is always a question between adherence to historical response or allowing AI to create alternative history. In this game it seems to come close to striking the balance between the two. The piecemeal attack just seems to fritter away any chance the German's have of any success against the Americans and 30 Corps. Arnhem is a whole other story I'm sure, as I've not played it yet.

BTW my original post wasn't a criticism of the AI that required it be looked at further. I was just wondering if anyone else had noticed the tendency and how they felt about it. Also curious about similar results when holding Germans back a bit. Maybe I'll be accused of playing German's ahistorical. Which would actually be funny because my war game friends say I'm predictable in that I tend to follow historical strategies.

[Arjuna]

Great discussion guys. Thanks.

Robs talk of maintaining lines is a classic suggestion though, more easliy suggested than implemented I'm afriad to say. It's another classic situation of what is easy and intuitive for the human brain is not so for the AI. In fact to implement the notion of front lines at this level of simulation is quite difficult. oops got to go back in a while.

[Arjuna]

There are a number of issues relating to the maintenance of a front line. First, when do you do so. If you did it all the time you would never be able to simulate those long power drives by armoured spearheads. If you allow a force not to, then under what conditions?

Second, how do you implement it? Do you do it as a series of objectives? Are the locations for these fixed or do they need to be fluid, ie constantly adjusted relative to the force ( ie like flank guards ). How do you determine where these should be without generating a plethora of route calcs ( which will place a huge load on the CPU? How do you avoid diluting the force allocation to the "real" victory objectives. If you use priorities, then what priority do you assign to the objectives for the purposes of force allocation? When do you abandon such an objective if you have insufficient forces available.

Or do you do it as a mod to the formation code. If the latter, how far do you stretch a frontage to maintain a link with its neighbours? In many cases, doing this will make the force vulnerable. When do you allow hanging flanks? Do you adjust the formation for changes in visibility ( eg at night )?

I can tell you now that any methodology cannot rely on extensive route calcs. Well not until machines get a lot faster. One option may be to utilise a control map. Determine the gaps between "real" objectives and then spread out a number of "front line" objectives in between, based on some arbitrary distance. Adjusting the actual objective locations to take advantage of good defensive terrain and choke points.

OK that's enough musing for now. Any suggestions welcomed. [:)]

[Jane Doe]

How about using a drag and drop feature that could be incorporated into a 2 phased defense order.
Like once you've clicked on the defense button, you need to choose an area where to defend then (that's what's already incorporated in AA), it would then automaticly "ask" you to draw an "abstract" frontline using click-drag-drop technique. It would be easier to join the fronts from one bn to another. I think it is good for attacks too.

I don't know if you understand what i'm trying to say?

[raydude]

A point of clarification - are we still talking about implenting better defensive lines for the AI? That is - for the units the single player is fighting against?

Ray

[Arjuna]

Jane Doe,

While that could work for the human player, how do we get the AI to make the same decisions? Providing means for the human player to make better decisions and have more control is relatively easy. The hard part is developing the AI to do the same.

[Tom Stearns]

Do we not already have a front line designation system for the human player when we set the depth, length and facing for a formation given a defend order or attack order for that matter?

[jcj]

The issue is making the strategic AI better able to form lines when appropriate. Generically, I believe that the AI will be better overall if it forms lines too often rather than not often enough. But, perhaps much can be gained by not looking at the issue so broadly. In a divisional sized engagement, especially where infantry is the prime component, there should be some amount of force tasked with flank protection at all times. In the example I gave above the force sizes are actually a little more than a division; 3 paratroop brigades and the airlanding brigade. For a force of this size, attacking a fixed objective a short distance away there must be some provision for flank protection. Otherwise, defeating this force becomes fairly easy with a coordinated attack on their flanks. On the other hand if they had just one company sized unit let alone a battalion on each flank, it would be impossible to roll-up their entire line so easily.

Maybe the best way to do this is to first have the AI define its main effort. That is, what forces constitute the main effort, then have other forces assigned to protect the main efforts flank. Three states could then be enforced, aggressive, moderate and conservative. Relating to the previous example an aggressive state would mean only light screening forces were present on the flanks and the flanks did not extend out too far. Another important piece of the puzzle is how these screening or flank forces are deployed. There must be a bit of intelligence to their deployment. In an agressive state where the flank forces are light, those forces must be placed at critical junctures. The key aspect of the system I have outlined is that flanks are created as an extention of 'main effort' forces. The terrain and other objectives are secondary considerations to the placement of flank protection.

While all the above may be very difficult to implement I think it will improve the AI vastly. Furthermore, I think it will enhance some of the other features already present in the AI, specifically a fair degree of creativity. The attached picture is a well formed line from when I played the Allies in the above mentioned scenario. Unfortunetly, I did not take a picture of the Allies when played by the AI. Perhaps I can run through the scenario again.

[Arjuna]

Thanks for your thoughts Rob. I agree that any implementation of flank protection must be subordinate to the "main effort". Assigning a portion of your force to this would be fine. One question that springs to mind. How would you see the concept of "tieing in" or linking up with neighbouring forces working. This has always been a difficulty. For instance, we have two forces, say Bde strength each. One is in the North and the other South say 4km apart. Now the Bde in the North might determine that its eastern flank should extend 1.5km and be angled from North to South East. However the southern Bde might think it best to extend a 1.5km flank angled from South to North albeit a few km east of its present position. The result you get a significant gap. Now who'se responsible for that and when is the tie in done. The issue is compounded becuase each Bde may develop its plan at different times. Now in reality, the Div HQ would direct where the boundary went and perhaps this is where the decision should be made and the fact made part of the orders. However, I have been loathe to address this issue in this manner because of the route calculations required to determin where that boundary should be. And you do need to do some. If you just specify a distance then you will end up with silly cases where the boundary forces the Bde to split its forces across major rivers etc.

Very interesting isn't it... [:)]

[jcj]

Well, I think you more or less answered your own question. The divisional commander would have a large say in how brigades are placed in order to coordinate their actions. Although I suspect brigade commanders often worked this out on their own as well. For example, they may say to each other that they would meet in the middle or if there was a landmark such as a town one commander would say he would be responsible for the town and everything west of that would be the responsibility of the other commander. Maybe instead of trying to create a general system you could address individual situations with specific rules. For instance if there is a river, or other impassible terrain then the brigade commanders would use that as the demarcation. If there is nothing unusual about the situation then they meet in the middle. I bet if you sat down and thought about it you could cover 95 percent of situations with 5 or 6 simple rules.

Then again I really don't know how the game is programmed or how to translate intuition into code. If such a system is feasible I think it would be effective. However, I think addressing the problem of open flanks and the process of forming any line, let alone one that is not perfect, is most important. But, the more I think about this "problem" the more confused I get, not just about a solution but about the problem itself. A recent poster commented on the lack of reconaissance around and in front of units. Perhaps the solution partly lies in this problem. I think most humans understand that while they can't see the units in front of them, that they are there. Maybe the AI can't make this leap of faith. Prior to seeing the poster's comment I hadn't thought about the lack of intelligence too much. Having played a few games since seeing the comment, I now see it as a bit of flaw ... one that can be overcome by a capable human player, but how novices (or the AI) must feel about this I don't know. What I am saying is that maybe in real life commanders had more information about where enemy units were then in the game, and if the game more accurately modelled reality in this respect that both the AI would be better, and novices would find the game more approachable. I would like to also make it clear that I am not certain that the game is not accurate with respect to intel and reconaissance.

[Arjuna]

Rob,

Thanks for your thoughts. Yes I realise that in part I answer my own questions. But then I find this discussion method very good for crystalising ones thoughts.

I agree that a few rules would go a long way to covering most situations about where the boundaries should be. The rules are not really the problem. The issue is how to minimise the route finding and still come up with enough "info" for the right decision. For example, how does the AI determine that there is a river between two objectives. Answer - by determining the route between the two and interrogating the route data, loc by loc. Interrogating the data is trivial and quick. The real speed problem lies with calculating the route in the first place. On large maps routes take a lot of CPU processing time - less so over short "traversable" distances, more so over longer distances and where passage may be denied because of rivers etc ( the very things we're trying to find, hey! ). I think that until we introduce a better method for tracing routes on the larger Control map ( 1km x 1km ) this may have to wait.

As to Recon, yes we would like to enhance this area and it is on the wish list and slotted in for Normandy ( Game 4 ). If we get time we may make some minor inroads for Game 3. But we'll have to see how we go for time.