Engine rewrite .. single player mode

[aka_tom_w]

I think the sugestion here was a request to give any scenario designer some additional tools (scripts or

SOP's or hints like flags that may or may not be seen by the opponent) to guide the AI in solo games.

This to my mind sounds like alot of fun because then two scenario designers could challege each other

via the AI in their own scenarios. Picture this, one scenario designer challenges another scenario

designer using identical (or roughly similiar) terrain and units to "program" (or just "hint" whatever..)

the AI to defeat the other human player. Both designers swap files and play Solo against the AI

programed by the other guy

See who does better against the AI !

To me THAT sounds like fun!

-tom w

_tom_w

Member

Member # 1515

posted November 06, 2002 11:20 AM

quote:

Originally posted by TSword:

Agree 100 %

The scripting should not augment but rather complement the current AI.

The bigger moves, coordination, assess terrain and the like are the problem of the

AI. Here scripting could aid tremendously. Of course only games vs. AI would gain

something. New type of battles would be possible, more historic accurate battles

would be possible and most important a challenging Computer driven attack would

be possible. A rather small investment for huge gains when compared to other

requirements.

Greets

Daniel

Thanks Dan!

That's what I (we ) meant:

"The scripting should not augment but rather complement the current AI."

If some of these "complementary" tools were available in the editor the scenario designer could "hint"

the AI in one direction or another perhaps, with scripts and SOP's and things like flags that are not in

fact flags but short term battlefield goals or objectives, to help the AI along during the battle.

Thanks

-tom w

[ December 08, 2002, 07:23 PM: Message edited by: aka_tom_w ]

[Eden Smallwood]

Originally posted by Pingu:

You guys are so unhip it's a wonder your bums don't fall off.

- Douglas Adams

Adding scripts would be banging the drum of defeat.

It would be puckering on the piccolo of progress. If a designer provided no script, the AI would do what it does now, or hopefully even better in the rewrite.

I don't think there's a valid objection other than that the benefits would not perhaps be worth the effort of coding. To that end the most interesting thing would be to hear from a qualified person on what concrete examples of improved behaviour might be possible, and how much of a bear this would be to code.

Eden

[ZealotBurner]

As far as multiple humans running large battles or one human controlling a small part of a larger AI force. Play Sid Meier’s Gettysburg or Antietam for example. The multiplayer in Gettysburg is awesome with a couple of humans on the same side.

[James]

There have been people saying that an AI good enough to challenge an experienced CM player is years away. But it may be possible now, take a look at the AI in Battlefronts Airbourne Assault. It works on the lines of real life command & control. Instead of trying to simulate a player it simulates real miltary proceedure. The higher level AI c/o's actually plan their manovers while the lower level AI c/o's execute them with the initative to adapt their orders to the actual situation. I could go on for ages about it but my dinners on the table soon - take a look at it on the Battlefront site and try the demo and you'll see A very realistic and challenging AI in action. Whats more its not script based and can adapt to different maps.

In short if the AI of AA was scaled down to the company-platoon-squad level and put into CM the single player experience would be great. Such an AI is not years away, its out there now. Got to go dinners on the table, take a look at AA and you'll be pleasently suprised, then imagine it transplanted into the CM engine.

[Battlefront.com]

Hi all,

I think people are getting a couple of different AI concepts confused here. For the sake of a better discussion, I would like people to use the following terms:

Scripted - the scenario designer designates that unit x does behavior y under z condtions. The lower level AI (TacAI) is coded to respond in certain ways based on the situation. The amount of variability depends on the choices programmed into the system to choose from. Generally there is very little higher AI (StratAI) in such systems, other than basic "sit and do nothing" or "go for broke, attack everything" type behaivors.

Unscripted - a higher level AI (StratAI) chooses a unit to perform a mission based on an overall plan. The StratAI is programmed to understand basic concepts and has basic abilities to assess various situations and determine which is the best course of action. Once it figures out what its overall plan is, it issues orders ("missions") to units under its command, such as "relocate from here to there", "defend until further notice", "attack using Area Fire", etc. The TacAI is coded to behave in a certain way depending on the mission and the situation as it unfolds.

The pros and cons are pretty easy to grasp:

Scripted - Pros

1. Easy to program compared to Unscripted

2. Easier to achieve desired behavior/strategy for a given scenario

3. Easier to challenge players, at least initially (see Cons)

Scripted - Cons

1. Brittle. If player does something designer did not expect, or code was not written to handle, the entire scenario can become a joke for the player. Or worse, the entire GAME can become easily defeated. Memories of SSG's original Reach for the Stars and Carrier Command come quickly to mind

2. Scenario design often has to be a secondary concern because the primary concern is to not expose the shortcomings of the AI's limited abilities.

3. Can not handle dynamic battles, such as those made from a random battle generator.

4. Much easier to "trick" either in a specific tactical circumstance, sceario wide, or game wide. This is because the TacAI uses much more predicatable "triggers" which the player can learn to avoid or exploit.

Unscripted - Pros

1. Can handle any map and force given to it, even if not equally well in all situations (much like a Human).

2. AI behavior is inherently not predictable, but of course this depends on how dymaically the AI is programmed.

3. AI can adapt to changing situations within a battle to the extent it is programmed to do so.

4. AI has a more or less "Human" feel to it, to the extent it is programmed to be.

Unscripted - Cons

1. All the pros are dependent upon skillful and extensive programming. This makes it infinitely harder and more time consuming to code than Scripted AIs.

2. Lacks predictability when it is desired, for example in historical battles.

3. Like Scripted AIs, the scenario designer often has to compromise the design in order to not expose the AI to its shortcomings.

Note that I did not mention "cheating". also made no mention of "hints", "markers", or anything else like that. Either system can be programmed to do these things in any combo the programmer desires.

OK, now that we have some terms and descriptions on the table, hopefully we can all speak the same language now I'll make some comments in my next post...

Steve

[aka_tom_w]

Thanks Steve

Interesting reading

I am looking forward to your next post in this thread !1

-tom w

[Battlefront.com]

Obviously CM's AI is Unscripted. It was the harder road for us to go down, but it is the far better one for a game like CM because of the dynamic nature of the game. Not only in terms of random QBs, but just in terms of the number of units. Take a look at games like Command and Conquer, Warcraft, Quake, etc. How many different units are there in those games? How many different types of offensive and defensive abilities? How dynamic are the maps from one scenario to another? If CM were that simplistic, a Scripted AI system might have worked for it. But of course CM is a tad bit more complex

CM's AI also doesn't cheat. Cheating AIs can be a bit better than they otherwise would be, but the player almost always figures it out and feels "ripped off". The best games I have seen that have cheating save such behaviors for an option. Unfortunately, the AI tends to stink even with cheating so it makes no difference in terms of the level of challenge.

CM does use "hints" to some degree. Exit Zones, Flags, parameter settings, and a few other things are used by the AI to figure out what it should be doing.

Where do we NOT see CM's AI headed towards? Scripting. We think it would be a big mistake and therefore aren't going down that road, even though it is much easier in terms of development energies.

What we do see happening is helping out the AI with more "hints". And not just the enemy StratAI, but both friendly and enemy TacAIs. We see a basic four way approach to AI improvement:

SOPs (Standard Operating Practices) - these allow more dynamic unit behaviors, such as "if you come under fire, favor movement over safety" or "if you come under fire, favor safety over movement". Obviously the unit might ignore its specified SOP, just like a regular order in CM, but it provides better guidence for the unit when unforeseen situations come upon it.

Hints - in the editor the player will have the ability to "suggest" various things for the StratAI to consider. For example, advance routes, defensive lines, strategic terrain features, etc. These hints are designed to help the AI make dynamic and unscsripted decisions, not to force it to behave a certain way all the time every time. However, it is likely that the player can do this if he really wants to.

For example, if the AI is instructed to attack on this map, here are three suggested routes of advance. Each one is rated in terms of how "good", "safe", and "fast" it is. Depending on what the StratAI decides to do for that battle, it might take the slower safe route on the left vs. the riskier fast route on the right. This means that the defending player can't know exactly which path the StratAI will favor. Of course, the player could make only one route and make secondary ones so horrible that the AI will likely never choose them.

Task Forces - units can be grouped together, before the battle, and assigned a specific goal. For example, 1st Platoon with 2 tanks will take Hill 123 while 2nd Platoon will take Village X. These Task Forces can be made loose or ridged, depending on the desired behavior.

Strategic Considerations - these are things like "take as few casualties as possible" or "take your objectives even if it means suicide attacks". These types of parameters can help set the "tone" for the overall battle and all the things that flow within it. For example, if it is a "do or die" type situation, SOPs will be more risky, Hints will be chosen with safety as secondary consideration, and Task Forces will be broken up only if it is in the strategic interest of the entire force (i.e. if one platoon gets nailed, thee other might not come to help it but instead take its objective).

Now... the interesting thing here is that these things can be done in random QBs to some degree. The StratAI can "guess" what it should be doing for a new battle by assigning its own SOPs, making its own Hints, designating its own Task Forces, and setting its own Strategic Considerations. It might not be as good as what a Human would come up with, but it is likely to be better than how it behaves now.

Final thought... the above is what we are shooting for, and therefore this is not some sort of promise being made. We think the above should take CM's AI to the next higher level, even if we can't implement everything we want to as well as we want to.

Steve

P.S. Airborne Assault is a totally different game and therefore can not be compared to CM directly. What is possible in that game system is not necessarily possible to do in a tactical wargame. Also, Airborne Assault is not as diverse or as "on the fly" as CM and therefore that must be considered too. Having a tight focus makes things much easier to do. CM's focus is far greater in scope and therefore inherently more difficult.

[Zarquon]

Amen. That's all we ever asked for.

Will it be finished in, say, late January? Can we preorder?

P.S.[OT]: The only try I gave AA resulted in the AI (Brits) landing on that heath, taking up defensive positions, sending a company or two towards Arnhem that circled through the woods for a few hours before returning, all the while being surrounded by more and more german units. I dubbed it the Maginot AI before uninstalling.

[aka_tom_w]

Originally posted by Battlefront.com:

Obviously CM's AI is Unscripted. It was the harder road for us to go down, but it is the far better one for a game like CM because of the dynamic nature of the game. Not only in terms of random QBs, but just in terms of the number of units. Take a look at games like Command and Conquer, Warcraft, Quake, etc. How many different units are there in those games? How many different types of offensive and defensive abilities? How dynamic are the maps from one scenario to another? If CM were that simplistic, a Scripted AI system might have worked for it. But of course CM is a tad bit more complex

CM's AI also doesn't cheat. Cheating AIs can be a bit better than they otherwise would be, but the player almost always figures it out and feels "ripped off". The best games I have seen that have cheating save such behaviors for an option. Unfortunately, the AI tends to stink even with cheating so it makes no difference in terms of the level of challenge.

CM does use "hints" to some degree. Exit Zones, Flags, parameter settings, and a few other things are used by the AI to figure out what it should be doing.

Where do we NOT see CM's AI headed towards? Scripting. We think it would be a big mistake and therefore aren't going down that road, even though it is much easier in terms of development energies.

What we do see happening is helping out the AI with more "hints". And not just the enemy StratAI, but both friendly and enemy TacAIs. We see a basic four way approach to AI improvement:

SOPs (Standard Operating Practices) - these allow more dynamic unit behaviors, such as "if you come under fire, favor movement over safety" or "if you come under fire, favor safety over movement". Obviously the unit might ignore its specified SOP, just like a regular order in CM, but it provides better guidence for the unit when unforeseen situations come upon it.

Hints - in the editor the player will have the ability to "suggest" various things for the StratAI to consider. For example, advance routes, defensive lines, strategic terrain features, etc. These hints are designed to help the AI make dynamic and unscsripted decisions, not to force it to behave a certain way all the time every time. However, it is likely that the player can do this if he really wants to.

For example, if the AI is instructed to attack on this map, here are three suggested routes of advance. Each one is rated in terms of how "good", "safe", and "fast" it is. Depending on what the StratAI decides to do for that battle, it might take the slower safe route on the left vs. the riskier fast route on the right. This means that the defending player can't know exactly which path the StratAI will favor. Of course, the player could make only one route and make secondary ones so horrible that the AI will likely never choose them.

Task Forces - units can be grouped together, before the battle, and assigned a specific goal. For example, 1st Platoon with 2 tanks will take Hill 123 while 2nd Platoon will take Village X. These Task Forces can be made loose or ridged, depending on the desired behavior.

Strategic Considerations - these are things like "take as few casualties as possible" or "take your objectives even if it means suicide attacks". These types of parameters can help set the "tone" for the overall battle and all the things that flow within it. For example, if it is a "do or die" type situation, SOPs will be more risky, Hints will be chosen with safety as secondary consideration, and Task Forces will be broken up only if it is in the strategic interest of the entire force (i.e. if one platoon gets nailed, thee other might not come to help it but instead take its objective).

Now... the interesting thing here is that these things can be done in random QBs to some degree. The StratAI can "guess" what it should be doing for a new battle by assigning its own SOPs, making its own Hints, designating its own Task Forces, and setting its own Strategic Considerations. It might not be as good as what a Human would come up with, but it is likely to be better than how it behaves now.

Final thought... the above is what we are shooting for, and therefore this is not some sort of promise being made. We think the above should take CM's AI to the next higher level, even if we can't implement everything we want to as well as we want to.

Steve

P.S. Airborne Assault is a totally different game and therefore can not be compared to CM directly. What is possible in that game system is not necessarily possible to do in a tactical wargame. Also, Airborne Assault is not as diverse or as "on the fly" as CM and therefore that must be considered too. Having a tight focus makes things much easier to do. CM's focus is far greater in scope and therefore inherently more difficult.

That sounds GREAT!

Thanks for the update

it sounds like a bright future for sure!!!

-tom w

[Eden Smallwood]

Originally posted by Battlefront.com:

SOPs (Standard Operating Practices)

Hints

Task Forces

Strategic Considerations

Ooo, Ooo, Ooo! Those sound like Really Good Directions™ to move in.

Just a note on nomenclature- what you've described up there is, I think, close or very close to what 'we' have meant by "scripting"... some of us- most of us? Just me? Some of us, anyhow, didn't mean the 'lame/static/brittle' scripting, part of the reason it's been a messy debate on this one.

At any rate, what you've described up there sounds really good. It sounds like you guys are hip to what's up. I for one am way content with what you at least are shooting for.

Final thought... the above is what we are shooting for, and therefore this is not some sort of promise being made.

Hey everyone, did you hear that? He PROMISED!!!

Eden

[Lou2000]

Battlefront,

Thanks for the info / update ..... yet again it shows how BFC actually take note of the activity on this forum. There are lots of companies who dont !!

Steve - any thoughts on the suggestion of the player only controlling a percentage of forces with tha AI controlling the rest. As I posted it earlier im obviously keen on hearing what the 'man at the top' thinks

Would this be workable with the future AI ...or even the current one ( thinking patches here ). Or is it something that has been considered before and rejected due to gameplay / AI restrictions

As in your last post ....not asking for a 'promise on content' but some official thoughts & views

Thanks

Lou2000

[Battlefront.com]

Yes, an interesting TWO YEARS lies ahead of us, so let's just keep in mind that stuff like this isn't cranked out (unfortunately).

Lou2000,

Well, the new engine is being designed to be multi-multi player. This will allow two sides to go against each other. It is too soon to tell how many per side or if the AI can assume a role on the same side as a Human. Having a Human and AI work together is far from easy.

Steve

[Michael Dorosh]

Originally posted by Battlefront.com:

Final thought... the above is what we are shooting for, and therefore this is not some sort of promise being made. We think the above should take CM's AI to the next higher level, even if we can't implement everything we want to as well as we want to.

Steve

How does anyone NOT respect that level of honesty?

Sounds terrific. But we will still whine for a campaign layer.

Incidentally, all the CADPAT trousers on ebay now purporting to be from the manufacturer? They are, but they are not the IR treated ones - there are several differences between these pants and the issue trousers - lack of ass reinforcement, leg tapes, etc. I think the manufacturer has been permitted to use the camo pattern, but not the IR material itself, to make up pants for sale to the public. They have the CADPAT brand name, but will not have a military tag.

If that makes a difference to you, word to the wise. My civvie friend has two full sets of CADPAT - issue IR stuff, pants and blouses - and I still have not been issued mine.

But the Canadian Broadcasting Corporation does, apparently, have priority over the reserves in Alberta regarding issue of CADPAT - the Chicken Cannon crew on Royal Canadian Air Farce is fully kitted out in the new uniforms. dulce et decorum est pro patria mori....

[ December 10, 2002, 02:14 AM: Message edited by: Michael Dorosh ]

[Lou2000]

Steve, thanks for the (quick) response.

and thanks to the others who posted here ....

This thread turned into one of the more interesting 'Future-of-CM' threads, with some good discussion and views, not forgetting BFC feedback, rather than just another CM 'wish-list'.

Lou2000

[massimorocca]

I wonder if the program could have a learning ability like the chess programs. May be in two years..........

[xerxes]

Wow Steve, that's exactly what I've been trying to describe, only your description makes sense.

I really like the task force concept, and the ability to set unit level goals (minimize casualties/hold at all costs/etc). The really slick thing is that this all corresponds to the orders a commander would actually give. Unlike scripting, a good CM player would be able to give the right SOPs to make the ai play a scenario well. And, and, at least some of the SOPs would be available to players during regular play. Very very cool.

Interestingly, you could have an ai vs. ai mode. With the addition of AI SOPs it would be an odd, but interesting game to have two players "Setup" the ai and let the engine resolve the entire battle. It would also be great for scenario testing/design.

Fascinating.

[Pak40]

Originally posted by Barleyman:

</font>[*] Allow scenario designer some simple scripting for AI. Think "attack vectors", "target area" and maybe "defense zone". This would tell the poor AI where to set up for a fight. Fallback/regroup-area would be nice, too.

The VL flags already create a "Target Area" as well as a "defense zone". The VLs tell the attacker where he needs to attack and tells the defender where he needs to defend.

The problem with attack vectors is that it's too much like scripting. The attacker AI would use the attack vectors every time you re-played the scenario, making the scenario predictable. The first time that the scenario is played would be great but the second time would be almost the same. Maybe there's a way to make it use random attack vectors.

[aka_tom_w]

Originally posted by xerxes:

Interestingly, you could have an ai vs. ai mode. With the addition of AI SOPs it would be an odd, but interesting game to have two players "Setup" the ai and let the engine resolve the entire battle. It would also be great for scenario testing/design.

OK!!

that is COOL

That is a great Idea

I really like the idea that two players might program the their own AI, then hit go and watch the battle unfold as the Strat AI dukes it out with the Strat AI

This is "sort of" possible now in a TCP/IP game set to one minute if both players program out ALL the moves and then sit back and watch both sides attempt to auto excute the moves. I have done once in CMBO (But it would be even more fun in CMBB with new commands and cover arcs !) and it is sort of fun to watch, every minute (timer set to one minute) a new turn starts and without any new orders both sides try to do what they were prgramed to do on the opening move. (both players, (playing head to head over a network), agree to set out ALL their moves on the first turn and do nothing thereafter, there is NO Strat AI (not really) and the TAC AI is left to fend for it self after first contact, it was a curious thing to watch, we only ever did it once)

BUT if the Strat AI could play the Strat AI and both players could program SOP's and sit back and watch the battle movie.

And and, get this....WITH a FULL movie uninterupted playback feature (its on "the list" last time I checked) after the computers spend all night crunching the game (no player intervention), the players could watch the WHOLE thing like a Movie in the morning!!!!

Now that would be facsinating!

SO what are you waiting for BFC boys?????

GET TO WORK!!!!

-tom w

[ December 10, 2002, 11:32 AM: Message edited by: aka_tom_w ]

[xerxes]

Heh, now that would be a totally different game design. Oddly, it would be more similar to a real commander. Another possibility would be to have an option for a VERY long turn. Say 10 minutes. So a 30 minute game would have only 3 "turns" where you change the SOP/objectives, etc.

[aka_tom_w]

Originally posted by xerxes:

Heh, now that would be a totally different game design. Oddly, it would be more similar to a real commander. Another possibility would be to have an option for a VERY long turn. Say 10 minutes. So a 30 minute game would have only 3 "turns" where you change the SOP/objectives, etc.

I hope they stick with the one minute turns, lets just trust the Strat AI to work out the details and make battle plan and stick with it.

I know we are just dreaming here and this concept of AI vs AI seems a LONG way away from other cool idea's like Multi-multi player and task forces and other strategic AI considerations.(Not to mention important "real" features like Relative Spotting and full movie playback).

But is is fun to theorize since they have not really started to build any new code yet (at least not that I know of....)

-tom w

[ December 10, 2002, 11:27 AM: Message edited by: aka_tom_w ]

[MikeyD]

An idea I had floated some time ago involved 'invisible' flags. This would keep the game unscripted and would help the scenario designer steer the action without giving away too much to the opponent. Imagine using invisible flags to direct a unit over a bridge, or to execute a wide left flank or an encirclling maneuver.

Invisible flags would act something like regular flags except they would be invisible to everyone but the AI, they would be specific to each setup color (red invisible flag markers, blue invisible flag markers, etc.), and they would have no point value. Of course they'd be most useful in scenario design, their usefulness might be limited in creating QBs, unless someone could think how to get around the problem. Maybe opening a game you could be givien the option to toggle invisible flags on and off. Comments? Questions?

[ December 10, 2002, 12:34 PM: Message edited by: MikeyD ]

[aka_tom_w]

Originally posted by MikeyD:

An idea I had floated some time ago involved 'invisible' flags. This would keep the game unscripted and would help the scenario designer steer the action without giving away too much to the opponent. Imagine using invisible flags to direct a unit over a bridge, or to execute a wide left flank or an encirclling maneuver.

Invisible flags would act something like regular flags except they would be invisible to everyone but the AI, they would be specific to each setup color (red invisible flag markers, blue invisible flag markers, etc.), and they would have no point value. Of course they'd be most useful in scenario design, their usefulness might be limited in creating QBs, unless someone could think how to get around the problem. Maybe opening a game you could be givien the option to toggle invisible flags on and off. Comments? Questions?

I think that is still a good idea...

Flags invisible to the player(s)

-tom w

[James]

Steve thanks for clarifying my AA speculation. What your planning is execellent.

The problem with attack vectors is that it's too much like scripting. The attacker AI would use the attack vectors every time you re-played the scenario, making the scenario predictable. The first time that the scenario is played would be great but the second time would be almost the same. Maybe there's a way to make it use random attack vectors.

id:Pak40

Pak 40, you can do this to some extent by useing dynamic flags in the editor.

[Battlefront.com]

Michael Dorosh,

Yeah... I know all about that CADPAT stuff on eBay. I've got my sources for the real thing if I want to pay an arm and a leg for it. I don't, so I'll just wait until they start surplussing the stuff. The price falls through the floor! Of course, if that place in Montreal has 'em at a decent price... I'll own a set sometime in January

MikeyD,

What I described above will be sorta like invisible flags. However, they will be flags with user defined attributes. The player could also assign paths with attributes, Task Forces with attributes, etc. The concept is to go much more towards a real world command model than the game currently does in order to help out the AI. However, unlike a command level game, the player would still be just as involved in the game as right now. No sitting back and watching 10 turns without interaction. That is not something we want to make. We'd be out of business soon after if we did

Steve

[Marlow]

Very cool Steve. Something to look forward to.