Playing the Ogre: Teaching the Computer

[wolf90]

Hi, everybody! I'm back with more questions. . .

In developing the Ogre computer game, we want the computer Ogre to be a worthwhile opponent. If the Ogre always plows up the middle, very quickly this will be a boring game to play. We need to try to define the various strategies an Ogre might take to beat your pesky little tin cans. So what I'd like to hear from all of you is a summary of how you play the Ogre, with the following questions core to your answer:

- Do you prefer fast forward strikes, or more probing the edges of a defense? Why would you choose one over the other? What might make you switch from one to the other?
- What type of defenses might dictate your style of attack?
- What circumstances might make you back up and try another route or strategy? . . . or, how target-rich does an area have to be for you to move away from the CP or other primary target? (Obviously a function of how many treads you have left.)
- How does your tread status effect your overall strategy?
- Does the "edge of the world" strategy appeal to you? Why would you use it, or abandon it?
- How do you use your missiles? Fire them early, or save them? Or both?
- What target priority do you give your missiles?
- What target priority do you assign the units if more than one are within range? This is something that might best be presented in percentiles, allowing some variability from game to game
- What else can be "tweaked" from game to game?
- When do you ram? When do you shy away from ramming?
- When would you overrun a position?
- When might you prefer more definitive kills (3-1 odds or greater) and when might you prefer to maximize your attacks (multiple 1-1, etc)?
- When would you immediately follow up a D result with another attack on the disabled unit, assuming there are un-D’d targets also available?
- How do you like to use terrain as an Ogre? Do you use towns for defense, or dive into water? Do you try to scrape off units by skirting woods or swamp? Do you destroy bridges?
- Do you avoid ending the move near terrain that might profitably be occupied by the enemy on his next turn?

I'm sure there are many more subtleties to playing the Ogre that I haven't touched upon. Please feel free to elaborate on any other idea you think of. Think of this as a class where you, the master player, are giving a dissertation to a newbie on how best to handle the Ogre.

This thread can wander far and wide, but please keep the primary question foremost in mind when you answer.

So . . . how do YOU play the Ogre?

D.

(. . .and yes, there will be a defense thread soon. Just focus on the Ogre for this discussion!)

D.

[Buzzardo]

I have all the impulse control of a rabid mongoose. I do try to zig and zag a bit and I'll often follow a side rather than go down the middle, but I never back up my Ogre. As for treads, full speed ahead! Damn the torpedoes!

I lose a lot. :)

-----------------------------------

I'd really like an AI that I can program, either by way of a bunch of checkboxes in a dialog box or by way of a file (XML, JSON, flat file, whatever) that I can customize. I prefer the file route, because I can create a bunch of them and swap them in and out and because I can then share them with other players. I can envision long conversations here about a file that works well (or not) and why it works (or not).

I'm biased, though. I've done a fair bit of rules-based programming in tools such as Jess, so that's how I tend to think of these kinds of problems.

[Misplaced Buckeye]

It seems like the best AI would be simliar to the one in Star realms which allows for Easy, Medium and Hard. It also seems that the programmers would benefit from attending an ogre event and watch how different gamers of different experience levels play the Ogre to program those traits into the AI. Similarly to play the defense if that option is going to be in the game. Playtesting at SJG Hq should put them ahead of the curve. Maybe videotaping the game if the programmers can't attend and that allows them to back up the game an analyse it.

[offsides]

Wow, where to begin. Granted, it's been a while since I've played the Ogre too much, but let's see.

I rarely go straight up the middle - I'll usually offset at least a few hexes one way or the other, based on how the defenders are set up and the crater/ridge layout. Howitzers are the #1 factor - I like to stay out of range as best I can until I've thinned the defenders, but occasionally I'll just charge. CP placement is #2 - if it's in the middle, then it doesn't really factor in, but otherwise I'll almost always at least lean to that side, if not going up the edge. Again, defender placement and force makeup will affect things - I'm more inclined to stay near the edge against a lot of GEVs (less room for them to escape), whereas against a force heavy on non-GEVs I like a little more room to maneuver and try and force them to either commit to being in retaliation range or letting me escape their attack zone. I almost never retreat since the longer you take, the easier it is to run out of treads, but occasionally I'll pull back for run sideways for a turn while I'm still at M3 to force a little more separation between some of the units.

I try to keep 1 (Mk III) or 2 (Mk V) missiles in reserve for those HWZ, but when in doubt I'll fire them before they get hammered. HWZ are clearly the #1 priority for missiles, with HVY being #2. I will occasionally burn one on something else if I'm playing a Mk V, but a Mk III just doesn't have the missiles to spare on anything else. Of course, if you get down to use-it-or-lose-it, anything goes, so just pick whatever target most in your way :)

Ramming is something that I'm conservative about to preserve treads, but at the same time if I'm surrounded or blocked from moving where I need to be, someone's getting crushed :) OTOH, overrunning INF is a given, as long as it's in the right direction (or at least not the wrong one) - the automatic reduction is a no-brainer. The other time ramming is critical is when you're down to minimal firepower and still have armor in the way - then the 1 or 2 treads you lose is still less than what they'd kill over 2-3 turns before you can hit them with a gun (assuming you still have any).

Early on, I favor hitting new targets over disabled ones, at least newly disabled ones. Depending on what the situation looks like after the Defender's turn, I may take them out or I may just push on. Early on, when you can move faster, sometimes it's better to just press on and try to leave the disabled units behind. Later, when you've got less movement, taking them out may save you from getting immobilized. It all depends on what the available targets are and what the firepower/odds options yield. A 2-1 or better is usually worth going after a clean unit, since there's a 50% or better chance of a kill, whereas a 1-1 may be worth using on a disabled unit since it's a 67% kill - but again, it all depends on the units. Plus, at M2 or less anything you let get behind you can generally stay out of range while keeping up, so there's that as well.

All of this is based on the original Ogre Map. I've rarely played with an Ogre on GEV maps, so I don't have anywhere near as good a feel for it...

[HeatDeath]

The Ogre, every turn, should, for every hex, calculate how many defender fire factors can be shot into that hex, and how many defenders it can expect to kill from that hex.

Then it decides which hex to move to. It can use a user-adjustable aggressiveness parameter to determine whether it wants to maximize enemy kills vs minimizing potential incoming fire factors.

You can also weight this choice by how far north each hex is - call that the "mission orientation" parameter.

You'll also want a heuristic that strongly weights rushing artillery if it's under a howitzer umbrella.

The defending AI is trickier, but one approach would be to reverse engineer this [rough sketch of an] algorithim, with an eye to minimizing the Ogre's northward progress, and try to arrange themselves so the Ogre will stay as far south as possible.

Yes, I've thought about this a little, why do you ask? :)

[offsides]

Originally Posted by HeatDeath

The Ogre, every turn, should, for every hex, calculate how many defender fire factors can be shot into that hex, and how many defenders it can expect to kill from that hex.

Part of me wonders if even modern computers are fast enough to do this without too much lag time. But I have to say that's an excellent suggestion for an algorithm to figure out where to move. The rest of it is, as you said, based on the aggressiveness and other mission-based factors, but this? You've just nailed the core Ogre AI in one sentence. *applauds*

[HeatDeath]

The processing cost on this goes up as the product of the number of hexes times the number of units, but even in the worst case scenario, that's a few thousand calculations per turn. A reasonably modern PC can do simple calculations like this ten thousands- to hundreds of thousands of times in less than a tenth of a second. For those of us who developed our intuitions for programming on Commodore 64s, or even 486s, modern PCs can be absolutely mindblowing.

The algorithim is probably actually how many defender fire-factors can be removed from a position in a given hex. The Ogre is sensitive to how much fire it can prevent itself from taking in the future, not so much how many units it is facing, though of course the one is at least roughly proportional to the other.

On green map scenarios, you would generalize the "move north" bias with a "move toward your objective" bias. [An objective is either a hex or an enemy unit.] In the case of multiple objectives, you would weigh them based on the density of enemies between you and them. [The "rush artillery" heuristic might be accomplished by designating artillery as ad hoc high priority objectives if you happen to be under their umbrella.]

On reflection, a version of this algorithm would be a good starting point for each individual defending unit. On the orange map, you would hardcode the objective to be the incoming Ogre, and set the aggressiveness level very high. They will also need a fairly heavy bias to stay within 1-2 hexes of their platoon-mates [units the user has specifically grouped them with that typically have similar speed, though platooning armor with infantry can be very useful, and should be an option for the player. All units in a platoon should typically have the same objective, or things might get weird.]

[EDIT: Defense platooning might be a good distinction to make between an "easy" defending AI and a "medium" or "difficult" defending AI.]

Defending infantry in a Breakthrough-type scenario may usually have their objective set to their current hex, with mission orientation dialed up high so they don't move.

Come to think of it, you could have a really interesting game mode if, instead of micromanaging the units, the player just reviews their platoons, their objectives, their aggressiveness and mission-orientation levels, and let the individual units figure out what to do and how to do it.

[HeatDeath]

The other main problems that require AI coding is Ogre fire-allocation, and where the Ogre should enter on Smash-the-CP scenarios.

Ogre fire allocation is basically mechanical - have the longest range guns fire at targets that only the long range guns can. The have the AP guns fire at the infantry, Then distribute the other guns so as to probabilistically remove the most fire factors from the defenders. [The probability math here will take into account that two 1:1 attacks are a significantly better probability of a kill than 1 2:1 attack, etc [the AI programmer will REALLY need to be provided with that probability article from the Ogre Book!], though actually that might be a good insight to reserve for the "difficult" level of the AI.] There might be scope to have user-controlled parameters on whether to save missiles for "high priority" targets [howitzers, mobile howitzers, CPs, etc] or to include them in the pool of 'use fire factors first on targets only those fire factors can hit.]

Whether, in general, to enter the south edge of the map at the left, center, or right end is an interesting question. I'd have to think about it a little more.

[ianargent]

Originally Posted by offsides

Part of me wonders if even modern computers are fast enough to do this without too much lag time. But I have to say that's an excellent suggestion for an algorithm to figure out where to move. The rest of it is, as you said, based on the aggressiveness and other mission-based factors, but this? You've just nailed the core Ogre AI in one sentence. *applauds*

You can do a lot of the calculations while the human defender is thinking; just as a human Ogre player is going to be doing a rough calculation of this "in the background" as the defender moves their pieces. The human will only be able to keep track of this in a rough way, but makes up for it by imagination. The AI can track every potential move hex, but only do what its been programmed to do.

[Dave Crowell]

The old Ogre computer game ran not too slowly on my old Amiga 500, so I think that running teh calculations for an Ogre is well within the processing capabilities of modern computers. ;)

As for Ogre strategies, avoid Howitzers, don't save missiles-better to fire them than lose them unfired, use missiles to kill Heavy Tanks, MSL Tanks are priority targets for Mains. Kill GEVs when ever you get the chance. If you let them get away they are like being nibbled to death by ducks. Ignore infantry as a target for every thing except AP guns and reducing them by moving over them.

Don't ever overun infantry on a green map, especially if they are in a town!