AI of Ground Units Discussion

It is a joy to see that AI is a very frequent discussed subject lately. Most of this discussion was about the AI of the planes, therefore I would like to take to opportunity and open a discussion about the AI of ground units.

Currently the AI of ground units in most sims is fairly simple. They can follow waypoints, sometimes follow roads and shoot things. With the big focus of LOMAC on CAS, that will get even bigger with 1.1 and 1.2 ( and hopefully includes FAC at a later point ), I think it would be good to think about the AI of ground units. Some things I would like to see:

-halt of advances

-surpression of enemy fire

-retreating ground units

Currently you have to kill every ground unit in CAS missions to support your troops. It would be cool if enemy ground units could be forced to halt their advance or even forced to retreat. I envision two ways, a "hard kill" and a "soft kill" one. First the hard.

A ground unit would halt/retreat after X amount of allied ground units in radius Y get destroyed. Of course the exact number would have to be found out first ( or perhaps beeing changeable in the editor ). This simply means that ground units abort the attack ( or give up their defense lines ) if own losses are too big.

Now the "soft" way.

The keyword is firepower. A unit would halt/being surpressed or even retrat if X amount of firepower falls down within radius Y and time Z. In simple words this means that if you put down enough explosives near ground units you can force them to take cover or "convince" them that it is better to pull back :) This might apply more to infantry units if we ever get them, but perhaps it would also work for vehicles. Testing and lots of adjusting would be needed to find out the right balance.

Factors that would influence it:

-skill: A more skillfull unit is more difficult to surpress/force to retreat

-moral: If a possible dynamic campaign would feature moral of units, it might influence the will to stand heavy losses

-weapon type: Napalm is alot more scarier than a normal bomb

-target type: A tank is less impressed by cannon fire while infantry jumps for cover

The way it would be set up in the editor; I don't know exactly. Perhaps a retreating unit would just turn around a follow its previous waypoints back. or perhaps the user would have to set up retreat waypoints for every unit. Another idea is to let the user set up retreat areas where all retreating units pull back to.

Another thing to consider. Should that retreat system work for every unit individualy ( every tank descides on his own when to pull back ), or should it be on a bigger scale. Per example the user defines groups of ground units, that will retreat together if the "commander" decides to. This would mean that even tanks that didn't face a lot of action would retreat of their whole group does so. I would prefer the last one, but such a 'high level of command' AI would be more work to implement.

I think the 'halt advance'/take cover/'be surpressed' actions would be decided by every unit itselfe.

This is all very theoretical, so I would like to make a little example to illustrate what all this would do:

Imagine the following mission. You are in your A-10 and fly a C-SAR mission, covering the rescue of a downed pilot by a chopper. You see an Igla being fired from a little group of trees, but you don't know where exactly the shooter is. You roll in and drop a pair of Mk-82 on the trees. Even if you don't kill the manpad, the explosion will keep the soldier on the ground for some time and even then he will think twice to uncover his position again ( =surpression of enemy fire by firepower ). Then there is a group of soldiers advancing to the downed pilot to capture him. It is almost impossible to take out the individual soldiers. You roll in for several passes with the Avanger cannon and lay down surpressive fire. Altough you don't kill them, you force them to take cover and stop their advance. This will buy the downed pilot enough time to get rescued by the helicopter (=halt of advance by firepower ). Lastly there are 3 tanks coming down the road to stop the rescue. You kill the first two by Mavericks, the last tank turns around and runs ( =retreat by "hard kill" fire ).

You see that a little more intelligent ground units can spice up the usual 'kill every ground unit' quite a bit. Throw in some radio comms and a FAC for the sandys and you have heaven on earth :)

If you have comments or other AI ideas for ground units, please post them.

Well, it would require addition of offenseive and defensive 'scores' to each unit, then an enemy unit could compare the 'visible' enemies offensive scores with its own offensive capability (both in terms of ranging and score) as well as its defensive score. Should it fnid itself outmatched, it could attempt to either hide or flee, or try to get a better position, like hull-down.

Personally, out of all the AI ground units, I think the SAMs need the most work.

You pop-up out of terrain in range of a Hawk, and 0.5 of a second later it has a missile inbound for you. IRL several things have to happen first. The search radar has to detect you, then the tracking radar has to actually pin-point locate you, then a missile battery has to be assigned to target you, then fire. The ex-instructor from the USAF I spoke to about this suggests it could take 40 seconds or more for all this to happen before a missile is inbound for you, even if you are well within range at the initial detection point. In LOMAC it is instant (IRL you can't engage a STT instantly either, from the ground or in an aircraft, but that's another issue entirely).

You get the same sort of thing with AI aircraft on occasion. On more than one occasion I've pop-up out of the terrain in front of an AI Su-33 to observe that he had somehow known I was about to clear the terrain and had ALREADY fired an R-73 at me (maybe they were using F5 :lol:).

There is a lot more depth to SAM/AAA operations than is modeled in LOMAC. Just have a read through the Falcon 4.0 manual to see some of the tactics that are used in real life. Flashing radars on and off to confuse RWRs, waiting until aircraft are within the no-escape zone before tracking or attacking aircraft, turning radars off when a SEAD strike it detected, etc. The SAMs in LOAMC are lacking any of this. Currently it's like they follow a very simple rule such that: radar on = true && if (jet <= SAM range), then fire.

Well, having such a detailed ground AI system may be slightly beyond the scope of ED's capabilities given the current time and resource constraints. It may be more realistic to have smarter SAMs (turn their radar on/off, wait to within range to ambush, etc.) and simulation of main battle tanks like the M1 or T-80. For example, they can increase range and accuracy of their main gun rounds to 3000-3500 km (typical max range), use anti-air machine guns/coax instead of their main guns to engage helos/fixed wing, trash the 'health bar' system, do better armour simulation (i.e. mobility kill, etc.) and so on. I think that's more in line with an air combat simulation, IMO.

In playtesting one of my A-10 missions(A-10_AleksandrovkaGap), I discovered that in a 6 or 7 unit group had the leaders destroyerd, the others are behind it stopped dead. Don't know why

When testing another mission(A-10_CentrePush) the T-80s and others were shooting each other in tha ass.

The lead tank crested the top of the bridge and had targets ahead of him. He fired. The tank behind him also fired, and hit the 1st tank blowing up the bridge and stopping traffic. Then the 3rd tank in line fired and hit the 2nd in the rear. Then all stopped including the BTRs who needed to go over the bridge to complete the mission.

I've diverted the T-80's off the bridge to provide covering fire for the BTRs and avoided the whole tank/bridge issue. I also moved the enemy units 45 degrees to avoids the bridge being hit.

In Steel Panthers, they have a Retreat Hex or a retreat Edge flag that's placed on the map by the human scenario maker.

One flag for each side.

All units head back to that - it might be simpler to do that than a retreat hex for each unit.

Not perfect but few systems are.

Levels of tactical awareness

There should be an operational level - one side deciding

and a tactical - one group deciding

Operational AI

Every 4 minutes the Operational AI counts loss and retreats and then decides if it's worth it to gain the objective.

The type of objectives - Destruction of a Enemy Target, Destruction of a Defended object, Arrival at a waypoint, Arrival with enough troups to hold an area, Defend a Target. We wouldn't have enough troops in the Ground war to take a look at the idea of surrounding or being surrounded.

Right now we have only 2 Objective states - SURVIVE and DESTROY. Maybe ARRIVAL should be added.

Leave the Objective calculation to the Operational AI as the loss of one group will not affect the gaining of the objective. Traditionally, there's a 3-1 numerical advantage that's built up before an attack(there are modern multipliers like air power).

Offensive objectives require more survivors while Defensive objectives can be held with fewer troops

Tactical/Group AI

Every time a Group loses a unit or has a 1 unit suppressed, it then decides if it can move forwrd to the next waypoint. Limit the Group decision to Retreat, Stop to Rally, or Go On.

Suppression

To be a Tactical/Group AI thingy

Every time a hit is made within a unit radius(by bomb or ground Attack) figure hits and proximity.

Figure out a group Suppression Value and then three value levels.

• Not Suppressed - keep moving
  Partly Suppressed - moving/firing at half the rate
  Fully Suppressed - not moving/firing

After 1 minute(or reasonable amount of time for an officer to rally the troops), reduce the figure by 30 percent(Rally Rate) and check the level again.

Varying troop quality

As this is a flight sim and Everything is an armoured vehicle, it be a waste of time to try and differentiate between troop types.

The best that could be done to differentiate is a Global variable of the Rally Rate(troop quality) for each side.

Elite - 40%

Veteran - 30%

Average - 20%

Recruits - 10%

Conclusion

One truism of game programming is - if you cannot SEE an effect, do not waste any CPU cycles on it. From 2,000 m in a flightsim, you cannot see anything but moving dots that occasionally fire.

The above thing that I stated should be the maximum for Ground AI. It's SEEable and CPU doable.

ZoomBoy27 - ...The tank behind him also fired, and hit the 1st tank...

I've flown that mission. There is no real need for that tank unit to be in column on the road. Indeed if they are expecting opposition, they might well be in another formation. Try having them deploy off the road as they near the bridge (say at 3km) and changing to Rank or Cone...

PS. The briefing for that mission says Su25_Snipe_Hunt. You can always use a text editor to directly edit the .mis file and provided you keep the exact number of characters in the file, it will still work... :wink:

...The tank behind him also fired, and hit the 1st tank...

I've flown your mission Centre Punch. There is no real need for that tank unit to be in column on the road. Indeed if they are expecting opposition, they might well be in another formation. Try having them deploy off the road as they near the bridge (say at 3km) and changing to Rank or Cone...

The final version had them deploying off the road. I did learn about Cone formation in my next mission.

What does Rank formation do?

What does Rank formation do?

Try it and see :wink: You need to experiment with number of units in the group as sometimes changing to Rank causes the whole group to stop dead in their tracks!

As for Alexandrovka Gap, I found something odd. I was just about to take off and I had to deal with a phone call, didn't pause, came back about 40 min later, QUIT and got a SUCCESS :?: This made me curious, so I ran it again, watching the ground unit battle which was good fun. It seems the friendly Shilka in "Reserve Group" kicks *ss!! killing all of UKR Group3. One or two other bad guy units survive elsewhere but none get near enough to kill the friendly command group that SURVIVE's.

I think the best addition to gameplay is changing how the ground AI responds to airplanes and not other ground units.

A column of tanks driving merrily down the road, a bomb takes out one of them, the rest carry on as if nothing happened.

The most simple way you can change this: If the group loses a unit from an airborne attacker, they halt and go to a defensive spread. Implementing cover-seeking routines might be asking much, the next best thing is defensive manouvering. The tanks disperse and drive around in circles/curves, making them hard to hit with dumb bombs. The collision avoidance routine for buildings and stuff is already there.

Another goodie: smokescreen. (fps hit though)

If no unit is lost for five minutes or some other set time, carry on to next waypoint.

Simple (perhaps) and adds very much to CAS/interdiction, which as of 1.0.2 is shooting ducks in a barrel.

I'm cross-posting an old post of mine from the Ubi forum here. It's from the debate about tanks engaging airborne targets with their main gun (something that was discussed over here aswell). While this is not strictly AI-related, it does touch on the subject and adds to what olaleier says about behaviour when under aerial attack.

"There are several solutions that would contribute to eleminating this problem. One would certainly be to improve the helo AI and make them use pop-up/shoot-and-scoot tactics, at least when engaging with missiles. Another would be to drastically reduce the accuracy of main gun rounds fired at airborne targets and bring it more in line with the effectiveness against ground targets.

However, the most important one IMHO is to enable tanks (and other AFVs beside the BTR-70) to use their machine guns, which they don't seem to do currently (with the exception of said BTR-70). The machine gun should also be made their primary choice of weapon when targetting aircraft and helos. The question whether tanks should engage helos and if they should do it with the main gun is a bit moot IMHO: they should most certainly shoot back, but they should do so preferentially with the machine gun (that's what it's there for, afterall)!

I don't know what exactly ED's reasoning for omitting the machine guns is, so I have no idea what potential problems may arise. I heard they wanted to include them only for remotely operated, unmanned turrets (such as on the BTR-70) to avoid having to animate a gunner, but I personnally couldn't care less about that. I wouldn't notice the lack of a machine gunner from my cockpit in the heat of combat."

Thanks for your comments.

One truism of game programming is - if you cannot SEE an effect, do not waste any CPU cycles on it.

You made some very good suggestions and especialy this sentence is very true. With this in mind I selected my proposals. I tried to select actions that the player can see from his point of view, actions where the player can see his influence on the ground war. I don't need a perfect ground war for a flightsim, rather just to such extend as the player interracts with it.

One thing to consider tought. When we will get a realistic FAC envoirment somewhen ( and I realy hope we get one at some time ), it will broaden the players insight into the ground war considerably. Trough the FAC the player can "see" to ground action very closely and has a much deeper view on his own influence on it. Therefore I would like the ground AI to be fit for a FAC envoirment once we get it. Or rather: A more complex ground AI is a first step to realistic FAC-CAS/C-SAR.

Another thing to consider. With Lock On 1.2 we will get the Ka-50 as it seems. From the hellicopter point of view we will have a much closer look at the ground war and interaction with it will be much more deep. I would even go so far that a combat helo needs more of a "ground war envoirment" than an air combat envoirment. The coming Ka-50 moves priority of ground unit AI up quite some bit.

As this is a flight sim and Everything is an armoured vehicle, it be a waste of time to try and differentiate between troop types.

Currently this might be true, but considering future developements I disagree with that statement. You are right that in the current Lock On, everything smaller than a vehicle is out of the players scope and that introduction of infantry units wouldn't cause a noticable change to gameplay.

But once FAC is introduced, infantry becomes "visible" for the player. Not necessary trough his own eyes, but the FAC will be the palyers eyes. This would make CAS work more interesting, as infantry is affected by different weapons than vehicles ( "firepower" weapons to surpress infantry ). I would like to have at least a difference between MBTs, light vehicles and infatry types of units.

And again the coming Ka-50 will have an effect in that regard. With the close look you have at the battlefield from a helicopter cockpit, targets aren't just armoured vehicles anymore. With targets such close in the players view, different behavior of different unit classes is much more justified.

Right now we have only 2 Objective states - SURVIVE and DESTROY. Maybe ARRIVAL should be added.

Excellent idea !

The most simple way you can change this: If the group loses a unit from an airborne attacker, they halt and go to a defensive spread. Implementing cover-seeking routines might be asking much, the next best thing is defensive manouvering. The tanks disperse and drive around in circles/curves, making them hard to hit with dumb bombs. The collision avoidance routine for buildings and stuff is already there.

 

Another goodie: smokescreen. (fps hit though)

Very good suggestions. I especially like the smokescreen idea.

To use SwingKids words: Thanks for the interest :) And keep it coming.

A relatively simple, but major enhancement would be non critical situation AI randomizations.

With the exception of critical situations (eg. ground avoidance, the last couple of seconds before firing) the AI would be slightly randomized for each craft. Randomization could effect things like situational awareness, attention span, agressiveness, turret turn rates and acceleration of angular acceleration.

A relatively simple, but major enhancement would be non critical situation AI randomizations.

 

With the exception of critical situations (eg. ground avoidance, the last couple of seconds before firing) the AI would be slightly randomized for each craft. Randomization could effect things like situational awareness, attention span, agressiveness, turret turn rates and acceleration of angular acceleration.

This is already implemented in FC to some degree AFAIK.

Also a very important thing is the ability to make a decision to continue with the mission, to retreat or to alter route - the latter in particular is important in terms of bombed bridges etc.

Just keep in mind to be carefull and to know how to count on emergent behaviours. It's interesting to create, on units, really basic behaviours making each of them partially rely on its neighbours ones, this creates (without materially coding it) an emergent group behaviour. The trouble is this behaviour can be eminently realistic, or eminently dumb. :D Only a shi*load of tests can ensure to keep the good results and get rid of the rest. :wink:

Edit:Making both unit IAs and group AI usually only gives dumb groups... reason of my suggestion.

Another great thing would be a "level of awareness" for air defence units.

There is one golden rule for a-g work, that is never to return for a second attack. Because then everyone on the ground will be ready to welcome you. Unfortunatly this isn't well represented in most sims. The first surprise attack is as dangerous as any follow up attack.

What would help against this is that AD units get a level of awarness. That level would determine how well a unit expects an air attack and how fast it will react on it. The the level would raise and fall with input from own sensors or trough 3rd party information and fall with time. Such an awareness level would be requirement to simulate an Integrated Air Defence System, but would also apply for non integrated units like Shilkas or Stingers.

Some levels I can imagine:

-Default: The soldier at the controlls would do the usual guard dutys without expecting an imediate attack. A quick attack would need considerable reaction time. The attack might be already over when a system with short range sensor like the Strela is ready to fire.

-Level1: The unit is aware that enemy units are in the region, perhaps trough warnings from other units ( if some sort of link exists ) or because it sighted enemy aircraft itselfe some time ago. Reaction time is shorter, but as the direction of a possible attack is unknown it still needs time to check 360° for enemys.

-Level2: The unit is aware of the direction of the comming attack trough own sensors or 3rd party infos and has its weapons and the controllers attention are trained in that direction. Reaction time is short as the controller expects an enemy any time. But an enemy plane from a different direction might come as a surprise ( diversion tactics ).

This is just a first try, this system would of course need lots of refinement.

The surprise factor would change the way we operate the A-10 and Su-25 quite a bit ( and make it more realistic ). And it would of course also lay a very good basis for future strike and SEAD mission types.

There is one golden rule for a-g work, that is never to return for a second attack. Because then everyone on the ground will be ready to welcome you. Unfortunatly this isn't well represented in most sims. The first surprise attack is as dangerous as any follow up attack.

Yeah, but it means another point has to be implemented first: the solution I proposed to make ground attack waypoints stand for ONE-PASS attack, otherwise your wingmen will keep flying over the target until their rests give nutriments to grass. :?

Yeah, but it means another point has to be implemented first: the solution I proposed to make ground attack waypoints stand for ONE-PASS attack, otherwise your wingmen will keep flying over the target until their rests give nutriments to grass. :?

True. There could be an drop down list on the attack waypoint in the editor to select attack options ( single pass/multiple passes, level/dive/pop up/loft delivery profile etc. ). The editor and flightplanning need a major boost anyway.

I think we should open a topic dedicated to airplane AI. I know it is a often discussed subject, but specific suggestions are always a good thing. I am currently thinking out and collecting some things I would like suggest.