Harder AI

[Cuddles the Warmonger]

I know the AI is generally just supposed to be a punching bag for the player to slaughter over and over again, but is there any chance they will have a higher difficulty level than in shock force and battle for Normandy. I don't mean like a whole revamp or anything, just an AI competent enough to know that it probably shouldn't meander up the strategically important hill or that it shouldn't continuously pour tanks into a choke point once the first I dunno eight T-90s are smoking hulks. 

 

Or even just the little things, like not walking out of the spawn, sending out probing parties rather than recces "in force", or maybe even semi-coordinated assaults rather than bum-rushes 

Edited January 11, 2015 by jamo552

[Ryujin]

As far as I know there's no AI, it's just scripted by the mission maker. It just carries out the pre-scripted orders regardless of what happens. 

[MikeyD]

Most of the people who brag about "always beating the AI" either haven't played the game in awhile or have a 'convenient memory lapse' about quitting the battle halfway through after they've got slaughtered and restarting fresh knowing where the enemy is.   CMSF NATO module campaigns were notorious for humbling the most die-hard players.

 

The game AI now has multiple types of triggers for choreographing troop actions, you can currently see them in action in all the CMRT scenarios. This is going to be the second title using triggers so there was no 'learning curve' for the scenario designers. They're old hands at it now.

[womble]

Where the AI script approach is weakest is the QB. Since MarkEzra, who does the lion's share of QB map production, cannot know what force composition he's writing scripts for (even to the point of "will it be infantry-only or armour-only?"), the plans are necessarily generic. One of the classes of situations you're describing (persisting with trying to use the same chokepoint, when to do so involves nagivating a clot of burning wrecks would be addressable at the level of the AI. Choosing which hill to meander up and which to advance on with purpose, and how to approach the evolution, is down to the AI plan, with the planner needing to be assiduous about using multiple unit groups to provide overwatch and scouting. Such detail is very difficult to assign without knowledge of the force composition, so the QB AI is never going to be a maneuver challenge of the level that a scenario plan can be. Triggers are a big improvement. BFC have said they will continue to improve the AI. Branching triggers would be another step change in the ability to plan, but would, of course, have less impact in the field of QBs.

For now, though, it is a massive challenge to make the AI attack effectively. A challenge that is too much for the generic plans that can reasonably be created for QBs. If you're playing QBs, you would be well advised, IMO, to restrict yourself to situations where the AI doesn't have to maneuver too much, so no MEs and the AI defending against your Probe/Attack/Assault. If you want to defend, have the AI Assault and give it 150% troops. At least then the broken remnants might be able to take the VLs because you've run out of ammo.

[SlowMotion]

And quickest fix is to start playing against human opponents.

[domfluff]

To be honest, I think the AI in Combat Mission is pretty great, all things considered.

The tactical AI (i.e., getting behind cover, using cover to advance, choosing the right weapon to use, panicking and running away etc.) is generally pretty sensible or at least mostly believable - the soldiers rarely do anything completely stupid, unless you force them to. Given how hard a problem that is to solve, I think that's pretty impressive really.

The scripting tools for scenario design are (of course) up to the map maker, and will depend heavily on their abilities. There are definite limitations here (not being able to remount vehicles, not being able to engage suspected targets, etc.) - and even with triggers the AI has little idea of the overall state of the battle. However, for the most part it's a tool to get the troops moving from A to B, and it does that reasonably well.

Sure, it would be nice if the strategic layer was more powerful, and if the operational layer had the ability to implement battle drills and the like... and equally if the whole lot could be dynamically generated on the fly (!), but for the most part it gets the job done fine, and difficulty is mostly a function of scenario design.

The old CM way of doing things was to base everything on objective flags, yes? This meant that if the scenario design put all the flags at the AI's end they would defend, and the opposite would be an attack. Something in the middle would create a meeting engagement. This clearly is a more dynamic option than the system in CMx2, but what we've had since CMSF is an awful lot more subtle.

[Stagler]

There is the scripted battleplan made by the map designer. Other than that the only ai is what controls soldiers when they panic - which moves them into cover and such. So ai in cm is very limited. In future titles a full ai should be something to aspire to that can make decisions based on your troop groupings on the map and place his major attack forces against them would be an acceptable base for a start.

[Chudacabra]

And quickest fix is to start playing against human opponents.

 

 

Agreed, although I think battles attacking against a well set-up computer opponent can be just as challenging, but playing a human is more fun.

[Jock Tamson]

And quickest fix is to start playing against human opponents.

With respect, that is no fix at all for those of us who want to play single player, which I believe to be the vast majority of customers.

 

For someone like me, who has a pair of newborn twin babies in the house, I can get 10 mins warning that I have a bit of time to play CM, and it might not re-occur again for 3 weeks.

 

It would be nice to be able to fire up a QB and expect a decent game as a single player.  I understand the limitations of the current system - I have made plenty basic scenarios for my own amusement - but there are other games out there that do have a non scripted strategic brain, Graviteam Tactics and Command Ops being the obvious examples.  The former would be my go-to for a dynamic single player experience were it not for the fact that I am fond of the CM style of play.

 

AI plans in CM are now into, what, their 8th year?  I own every CMx2 game including Afghanistan, and all the modules.  I guess that is several hundred pounds worth.  Typically, when a new title comes out, after a few weeks of play I am left feeling that I am funding the multiplayer experience.

 

Of course, as we learned in "the other thread", increased exposure to an exponentially larger market of gamers will not lead to more sales, as 100% of people who might want these games have apparently already found them.  Even though a search on Google for "ww2 PC wargame" does not reveal a Battlefront URL until the 11th page of results.  So I don't suppose we will ever see the increase in sales required for the resources to implement a real sea change in some of the CM architecture.

 

Modding could be one answer - several mods were made for Arma2 which introduced an artificial commander for Opfor - however I doubt the desire is there to open up that avenue.

Edited January 11, 2015 by Jock Tamson

[MikeyD]

The solution to a relatively weak AI in QB is to build proper scenarios and share them on Repository. People certainly weren't shy about building scenarios for CMSF! Playing in the editor is half the fun of the game. But, it appears to be a consequence of the steadily improving QB in the game that players have come to rely more and more on the QB generator instead of build-your-own. Plus, as scenario quality has improved prospective scenario designers seem to have become intimidated. For their first map they try to recreate all of downtown Antwerp and soon become frustrated. Really, your scenario doesn't need to be a doctoral dissertation on modern European history. Two houses on the map and AI infantry told to move from one house to the next is a 'scenario'. If the result is fun then the scenario was a success.

[antaress73]

My experience is that The AI can kick your a$$ in defense. Much less so on the attack. But thats normal since attacking is more difficult even for human players.

Edited January 11, 2015 by antaress73

[womble]

...there are other games out there that do have a non scripted strategic brain, Graviteam Tactics...

And do any of those have the complex multilayerd 3D environment from which to derive their inputs that the AI in CM has to winnow through to decide what's important?

...I am fond of the CM style of play.

Might it be the detail and complexity of the environemt having anything to do with that?

Modding could be one answer - several mods were made for Arma2 which introduced an artificial commander for Opfor - however I doubt the desire is there to open up that avenue.

Pretty sure you're right, there.

You can still get a pretty good performance out of the QB AI if you don't ask it to attack. The first thing I'd poke at for improvement there would be in the force-picker for the AI, so that a) you get a sensible force without having to go in and tweak it and find out what you're up against, and it's optimised for the number of groups the AI plan has, and the assignment to those groups is sensible (rather than assigning the AT platoon to the moving element and the tanks to the static one, for example...)

[Jock Tamson]

And do any of those have the complex multilayerd 3D environment from which to derive their inputs that the AI in CM has to winnow through to decide what's important?

 

It would be an entirely separate thread to those working in the 3d space, "pulsing" at whatever number of frames the devs felt was appropriate.  Remember we are talking about an operational layer that reacts to the bigger picture info, could be as infrequently as every five minutes. 

[MarkEzra]

Red Thunder brought triggers and a necessary learning experience for me. I now have a better understanding of the depth of this process. CMBS will introduce some refinements to the QB Game to improve the player's understanding of the maps and scoring. I have also been working on the AI groups, their orders, and how they respond to triggers. It's been an exciting time.

[womble]

It would be an entirely separate thread to those working in the 3d space, "pulsing" at whatever number of frames the devs felt was appropriate.  Remember we are talking about an operational layer that reacts to the bigger picture info, could be as infrequently as every five minutes.

It's not the decision making that's harder. It's the digging out what's relevant from the environment. Figuring out what terrain is significant, assessing the position on the very broad spectrum of operational capabilities and conditions that CM can represent. The number of dimensions to be analysed and their interactions makes developing the algorithms to do that very challenging indeed; it's not the CPU that's the problem.

The first part of making the AI's decision-making better is improving its assessment of the situation. It has already been said that currently the AI has very little awareness of its environment. "Friendly troops close" and "enemy to the flank" and "Is that slope exposed to fire?" aren't part of its calculations, only "Am I hurting or scared?" and "Is it hard to go through those next few AS?" Assigning values to those nebulous judgements and then organising their interaction is the tricky system development, not writing the code to have the CPU crunch the "rules".

[c3k]

With respect, that is no fix at all for those of us who want to play single player, which I believe to be the vast majority of customers.

 

For someone like me, who has a pair of newborn twin babies in the house, I can get 10 mins warning that I have a bit of time to play CM, and it might not re-occur again for 3 weeks.

 

It would be nice to be able to fire up a QB and expect a decent game as a single player.  I understand the limitations of the current system - I have made plenty basic scenarios for my own amusement - but there are other games out there that do have a non scripted strategic brain, Graviteam Tactics and Command Ops being the obvious examples.  The former would be my go-to for a dynamic single player experience were it not for the fact that I am fond of the CM style of play.

 

AI plans in CM are now into, what, their 8th year?  I own every CMx2 game including Afghanistan, and all the modules.  I guess that is several hundred pounds worth.  Typically, when a new title comes out, after a few weeks of play I am left feeling that I am funding the multiplayer experience.

 

Of course, as we learned in "the other thread", increased exposure to an exponentially larger market of gamers will not lead to more sales, as 100% of people who might want these games have apparently already found them.  Even though a search on Google for "ww2 PC wargame" does not reveal a Battlefront URL until the 11th page of results.  So I don't suppose we will ever see the increase in sales required for the resources to implement a real sea change in some of the CM architecture.

 

Modding could be one answer - several mods were made for Arma2 which introduced an artificial commander for Opfor - however I doubt the desire is there to open up that avenue.

 

 

In no particular order...

 

- Congrats on the twins!

 

- Only get 10 minutes warning that you have time to fire up the game? That's what pbem is made for. Find a good mate (or several) and take the games at your own pace. (In one pbem I'm engaged in, we sometimes exchange only 1 turn in a week. That's fine by both of us. Real Life has a way of getting in the way. This is a GAME and should be FUN. Easygoing pbem is part of that.) Without minimizing the validity of your criticisms, head to head via pbem is a great experience. (If you want to get your feet wet, PM me.)

 

- Modding the AI will not happen. That is such a deep part of the code that it cannot be opened up to outside manipulation.

 

- MarkEzra, answering you upstream, did not mention that he has produced virtually every single QB map...and the AI plans to go with them. I'm in awe of what he's done. He mentioned the AI triggers. The trigger function seems simple, but has amazingly complex inter-relationships. As designers, such as MarkEzra, get more familiar with the nuances and power of the triggers, the AI will act/react in a much more dynamic/realistic manner. It will continue to improve.

 

- AI plans are NOT into their "8th year". If you look at the introduction of triggers as a whole new approach, then AI plans have only been out since CMRT. (A year?) Compare CMRT AI with CMBN's initial release AI. That's a whole new ball of wax. (Note that CMBN v3 has AI triggers, just like CMRT. Added to preclude misinterpretation by folks not familiar with the upgrade system.)

 

- I agree that dynamic QB attacks are nowhere near what a human opponent would conduct. AI defense is quite good, however. (In most cases.)

 

Ken