AI Discussion (dev branch)

[mad_cheese 593]

On 31.10.2017 at 10:53 AM, Theo1143 said:

 

 I know right...

Exactly me point even basic things don't work anymore.

As stupid example take the hostpital at altis it is a building who has been there since they released altis.

You should think since there is pathing on top of the roof they should be able to walk there, well you WRONG.

If you put a heli on top of the roof then order them to go in while he is standing beside the heli.

He will walk all the way down walk to the garage then teleports into the heli and sits in the pilot seat like nothing happened.

Now say this is a mission where u see a guy running down stairs 10 sec later u hear the engine going on from the heli on the roof.

Oh yea that really helps by the realism (sarcasm is real on this one)

 

This could be less frustrating if you consider that units just can not board a vehicle (chopper) via another vehicle (house). Even if the house has a path, this does not mean that the unit can go anywhere and do whatever.

 

I would also love more functionality of AI in buildings, the system is maybe a bit rudimental. But the setup you described is a bit unfair,  AI are not programmed to do that stuff. It's not a basic thing and never worked. 

 

 

[fn_Quiksilver 1636]

[snipped]

[Vasily.B 529]

7 hours ago, fn_Quiksilver said:

 

your right, 16 years in the same engine and they've never been programmed to walk naturally in/on buildings. 

its sad that this is the most recent changelog

 

 

 

I fell this too, but as BIS have no AI programmers, we cannot expect to have it fixed in next months. 
But there is some move in right direction in BIS, that can handle those issues at least in some way - they opened doors to third party studios for premium DLC, keeping old rule - content (vehicles, weapons for $, features and mechanics for free with platform update). Thats some oportunity to get some Ai programmers that can release DLC (for example) "AI command tactics" which will improve basic functions (searching houses, walking off the streets, driving improvements) for free, and add some command features for $. I know i'm dreaming, but as Aerosmith was singing "dream till the dream come true".

Yes, they wasted much time by spending manpower into other projects like for example Visual Update, which is serious mistake and messed up lighting in day/night. I dont know when AI programmers have left BIS precise, but as they were employed, chiefs could make some push for fixing those issues - for example driving "improvement" project could start much earlier. Now as theyre gone chance is lost and lot of things is messed up/left unfinished.

This is pure example of wrong planing roadmap and milestones. If i would be director of BIS i would dismiss people responsible for letting that situation to happen. 

[kklownboy 43]

My theory for the lack of vocal disgust about broken AI/driving walking etc... is its so OBVIOUS that we all just expect it to be fixed...It is in everything that this happens... since the very beginning of A3...(A2 was way better...) We have all learned to just not use 50% of the game due to this fail. The 50% we do use is awesome,

 

BUT this is just one big DUH!!   i mean fuck... or what the f, even REALLY?... Come on mannnnnn.....

 

 

[Vasily.B 529]

18 minutes ago, kklownboy said:

My theory for the lack of vocal disgust about broken AI/driving walking etc... is its so OBVIOUS that we all just expect it to be fixed...It is in everything that this happens... since the very beginning of A3...(A2 was way better...) We have all learned to just not use 50% of the game due to this fail. The 50% we do use is awesome,

 

BUT this is just one big DUH!!   i mean fuck... or what the f, even REALLY?... Come on mannnnnn.....

 

 

Saint words, but from my experience i can say we can moan here forever and this will give us nothing. Nothing beacuse lack of manpower in BIS.

[R0adki11 3949]

22 minutes ago, Vasily.B said:

Saint words, but from my experience i can say we can moan here forever and this will give us nothing. Nothing beacuse lack of manpower in BIS.

 

Please try to stay on topic and away from such inappropriate comments. You of all people should know how the forum rules work.

[ProfTournesol 956]

Moaning about things that obviously the engine cannot handle better (AI pathfinding into buildings and bridge crossing) is useless and boring. Better wait for next game installment.

[fn_Quiksilver 1636]

[snipped]

[Strike_NOR 898]

There are a few things that would tremendously improve the AI, which I would hope to see done in ArmA 3 before the devs completely abandon improvements.

 

• Driving skills. There are so many times where the AI simply shows it is not capable of following roads, adapting speed, avoiding collisions, avoiding teammates etc. Many times it'll snag onto a fence or something and just freeze there. There seems to be a lack of the AI recognizing that it is stuck and trying to reverse and find a different path around the obstacle. There are also so many poor decisions on hill terrain, where they chose incredible inclines to go up or down. Causing their vehicle to freeze halfway up a hill.
• Walking in/on objects. The AI seem to be able to move right through collision objects in many instances. Walking through walls, doors etc. Also they really struggle on piers, bridges etc, running off buildings or obstacles and dying.
• Taking proper cover. AI that get caught in the open seem to favor staying in the open rather than deploying smoke and moving for hard cover.
• Recognizing threats at different levels and prioritizing them. A grenade nearby should trigger running away from it, a tank approaching should trigger a priority 1 urge to get to concealment, vehicles impending explosion should trigger them running away. Airstrikes, strafing runs etc should trigger infantry to move away from vehicles and spread out etc.

 

The game has so extremely much going for it, and with proper AI, it has the potential to rival anything out there. Yes, I realize that having advanced AI will tax the system further, but that's kind of where dynamic simulation comes into play. By only rendering and calculating nearby AI, a lot of calculations can be done without.

 

There are very many games that do AI very well, even in sandbox games with huge open worlds, but the game has been designed with AI performance from the ground up. Every object is tailored for AI to take cover behind, every corner is known to the AI, every bridge, every roof, every door, ladder etc is known to the AI. It also knows where enemies are and will position itself behind cover. However, these other open world games have a lot less going on at once than ArmA. A typical FarCry 4 scenario is really just an isolated bubble of the environment, restricted to a few 100 m radius. Nothing outside is simulated, only distant terrain is drawn. Enemies within the bubble are normally only restricted to a handful at once and bodies, weapons and vehicles despawn quickly after dying/taking damage. ArmA does not do this, and that's a double-bladed sword. It is realistic, to have everything "exist" at once and never disappear or spawn during the game, but it is also very CPU intensive.

 

I hope that ArmA3 AI gets a thorough cleanup before everything shifts to enfusion, and hopefully, what the devs learn from coding AI on enfusion, can be used to improve arma 3 AI  just a little bit more before it is abandoned.

[lex__1 422]

I've already posted a video here, where different cars, most of which perfectly ride across the bridge.
Some have problems crossing the bridges.
Each type of machine has the features of the movement in front of the bridge:
- some do not slow down
- some reduce speed
- some change the rectilinear motion
- some do not change the linear motion
- Some begin to make a turn before the bridge or on the bridge
There are AIs that do not have big problems in driving, getting the right track, or no mistakes in the way, or the skill of using the vehicle and its features in running.
I think that each of you has seen how some machines with AI can perfectly and quickly move in different conditions.
The only thing that all AI works equally is the absence of a fast detour of objects, in which they often rest, and try to pass through them.
Is it really impossible to restore order and eliminate this problem?

[fn_Quiksilver 1636]

have snipped a bunch of my recent posts to this thread, i feel ive done my part in raising awareness here and elsewhere regarding some AI issues. That said my posts were hurtful in some ways and that doesnt help anything in the long run.

[nkenny 1057]

I feel that the main problem is the  incongruities between selected AI ROE/orders and expected behaviour. While an annoyance as a mission maker, this fatally undermines any attempt to squad lead or command, by high command or ZEUS, multiple teams of AI soldiers.  It seems that small tweaks could be made for great effect. 

1.0 More aggressive AI: 
1.1 I wish the AI engaged more aggressively, but perhaps less accurately with certain weapon systems.  Compare the way players will deploy weapons (sometimes indiscriminately) to how the AI will.  Especially low grade AT weapons like RPG-7 should see the AI cheerfully fire at any target of opportunity. 

1.1.1 Low flying helicopters and infantry specifically.  Think Black Hawk Down like scenarios. 
1.2 Vehicle crews targeting garrisoned buildings. 

2.0 Stealthier AI: 
2.1 The current Stealth ROE is anything but.  The AI may whisper, but the mode of movement is by rushing leapfrog.  A sensible mode of movement would be limited to walking pace and going prone on spotting a relevant threat. 
2.2 Stealth ROE has few or no ways of limiting fire to AI equipped with suppressors. A simple way to set this up would be great. 

2.3  Sympathetic ambush AI:  If one member of a team is spotted and opens fire, it would be a sensible change to have the entire squad respond.

3.0 Cover/Suppressing AI: 
3.1 The new suppressive fire mode is good. The lack of ability to easily define covering arcs not so much. A way to easily define covering arcs from both editor and mid-mission would be a wonderful change. 
3.2 On similar note: Halting or going firm,  setting up an allround 360 defense would be helpful. 

4.0 Lack of visual and audible feedback of suppression states
4.1 Perhaps the most grievous  is the lack of feedback to help determine a player that the enemy is suppressed (or in a cowering mode).  This undermines any attempt at fire and manoeuvring. 
4.2  Crew abandoning burning vehicles also tend towards being laser accurate. 

5.0 Lack of interface for certain common infantry actions
5.1 Deploying Smoke or Frag grenades
5.2  Deploying mines or demolition charges

5.3  Assaulting or Garrisoning buildings. 

5.4 Force fire of ready AT or HE weapons at point targets

5.5 Rushing/storming/blitzing to or away from a location 

6.0 Infantry formations
6.1 Current infantry formations are poorly suited for fighting in Urban areas.
6.1.1 The AI will sprint at inappropriate times (to keep in formation) 
6.1.2 The AI will flank unnecessarily (instead of squeezing in behind formation leader)  
6.1.3 The AI is generally unable to keep up with formation leader in both speed and agility.
6.2 Current infantry formations are poorly suited to conduct rapid rushes both towards and away from enemy. 

Compare and contrast Tom Clancy's Rainbow Six: Rogue Spear 
The 1999 tactical shooter dealt with a much more limited environment. Even so the user interface and tactical options available to the commander was in many ways greater. 

By combining two  ROE setting, Mode and Speed, the player could adapt tactical approach to a large number of scenarios.  See attached picture from game manual.
 


Not shown is the mid-mission  Quick command where the player could set up (or adjust)  ROE and specific actions on the fly:  Including cover arcs, allround defense, ROEs and as I recall certain grenade usage. 
 

 

Compare and contrast Tom Clancy's Ghost Recon 
The much simplified 2001 tactical  shooter is another interesting point of contrast. Again simpler in scope and expansion to any Arma game, but easily superior in user interface. 

ROE was again a section of two modes:  Movement and Combat. Covering arcs were also easily defined mid mission.

Movement had three options:  Hold,  Advance, and Advance at all costs. 
Hold,  the team would hold and seek cover. 
Advance, default option, the team would advance until fired upon. When fired upon they would adopt the hold mode. 
Advance at all costs.  Just what the name suggests. 

Combat had three options: Assault, Suppress, and Recon.
Assault, the default option. 
Suppress, the team will lay down a great deal of fire, sacrificing accuracy for effect. 
Recon, the team will hold fire unless fired upon. 


 Conclusion 
My point is not that Arma should adopt any Tom Clancy's control interface wholesale. Arma is much more advanced than either of the suggested titles.  What I suggest is that Arma3s AI, in addition to the navigation and movement problems mentioned by other posters, fails along three lines: 

1. Correspondence of  expected behaviour to user interface 

2. Ability to actualise common infantry tasks in user interface

3. Consistent and conventional visual and audio feedback of current state of both friend and enemy forces. 

-k 

[Strike_NOR 898]

20 hours ago, nkenny said:

It seems that small tweaks could be made for great effect. 

 

Whoa, this is a really good/extensive list of improvement suggestions.

 

I share most of your views and thoughts on this. 

 

There is a more complex, but sandbox-friendly idea that has popped into my head on many occasions. 

 

I would hope to see an AI system that works closely to how a human soldier, squad or platoon would operate when dealing with battlefield awareness and tactics.

 

I'd call it something like Dynamic Frontline System (I realize there is a PVP conversion mod called Dynamic Frontlines, but this is not entirely related).

 

Introduction

 

How it would work is as follows: Each unit (U) has an individual "sphere of influence" which follows it, carrying the "side frontline". This dictates where "the average frontline" appears. If no known about threats are nearby, this average frontline encompasses a large area around friendly unit(s).

Example 1. Single blufor unit surrounded by its sphere of influence, with its side frontline and finally average frontline (not to scale, this could cover the whole map).

 

Each unit then has its own "weight factor". The higher the weight, the greater the sphere of influence becomes. These weights also combine with other members of the same group (G). Two soldier units of equal weight, will double the group sphere of influence weight. Two individual units that are not in group, will widen the side frontline, but not increase its weight. If any two units get too far away, their sphere of influence breaks into two separate spheres. If a group loses a unit, its weight decreases by the lost units weight value.

 

Example 2. Pair of blufor units not grouped, overlapping sphere of influence leads to wide, but weak area of influence (top).

Pair of blufor units grouped leads to higher collective weight and greater sphere of influence (bottom).

 

Weight factor

Like I mentioned above, weight is carried by each unit in arma. This can be best understood as the combat strength of the unit. When you combine units in a group, the group leader carries the collective weight of the entire group. As a quick example, if each soldier is "1" weight, then a group of 12 soldiers has a weight of 12. The sphere of influence in this group increases in strength, pushing the frontline further from the group center.

 

Differentiating weight per unit would, in its simplest form, be decided based on unit type. A regular soldier has the lowest weight, machine gunner has more weight, anti-vehicle soldiers have more weight, snipers, special ops etc. The list goes on. This essentially means that a heavy weapons squad generally has higher weight than a regular squad.

This same system carries on to crewed vehicles. Transport - unarmed/armored = lowest weight. Transport armed/armored = higher weight. MRAP<APC<IFV<MBT. This means that for example, a single MBT (Main battle tank) could have more weight than 2 infantry squads combined.

 

Aircraft, as in real life, do NOT influence the front-line of a battlefield directly because they do not seize land. They are to be considered as a support role for attacking/defending at the frontlines and behind them. They play a key role in reconnaissance and battlefield awareness.

 

Merging spheres of influence

As mentioned before, a group shares the collective weight of all group members, and hence the sphere grows as weight increases, or shrinks when it decreases. However, they do not merge with other groups, unless the two groups use a "join" function in the game. If two individual groups "side frontline" touch (group A radius + group B radius = physical distance between A and B) then their frontlines merge, just as shown in the figure "example 2" with individual units. This creates a wider sphere of influence, but it may have an uneven distribution of weight, taking on a more "pear shape" appearance. 

 

This aims to help the system understand that the area between two nearby groups has more strength than the area between two distant groups.

 

Example 3. A single tank unit has heavy weight merges with an infantry group nearby that has lower weight. Since they are separate groups, the spheres do not add together, but stretch their influence between each other.

 

If you are paying attention to what's happening at the larger scale here, the system is becoming aware of the combat strength of allied units and groups in an area. It is also able to recognize where it is strong, and where it is not.

 

The enemy and dynamic frontlines.

Now we have seen some examples of how allied units influence the "side frontline" by means of spheres of influence that alter with weight and group size. Now it is time to apply that same system to an enemy faction and give the system something to work against! First we have to talk about two different terms:

 

The subjective frontline and the objective frontline.

 

The subjective frontline is a side-specific system that only uses the information that is available to its own side. In other words, it always knows the position of friendly forces, but only knows about revealed/spotted enemy forces. This system taps into the already working spotting mechanics of arma that reveals targets as unkown, eventually type, and later specific unit. At a distance, revealing an enemy squad will only give the system information about 12 men. If we use the weight parameters from before, this means the spotted enemies will have a collective weight of 12. This is then used to calculate the frontline. If the enemy either moves closer to reveal soldier type, or friendly forces have powerful optics that can reveal soldier type at long range, then the different soldier weights are also added to the system. This means that instead of having a weight of 12 (generic men), it now has the weight of 20 (heavy weapons, marksmen, grenadiers etc).

 

Summarized, the subjective frontline only knows about spotted units and the quality of that information (how much details are revealed). When time passes, spotted contact information fades out, and the system adjusts accordingly.

 

The objective frontline is not side-specific and knows about all units on both sides at all times with full information. This is particularly good for arcade game modes or multiplayer where "fog of war" is less relevant. This mode will immediately calculate the frontline across the map so both AI and players know where everything is happening in real time. This way, you would essentially be "safe" in friendly territory and crossing the frontline means that you are moving towards enemy forces that are generating that frontline in realtime.

 

Mission makers decide when to use subjective frontline (realism, fog of war), or objective frontline (fast-paced action, good situational awareness) based upon their preferences.

 

 

The dynamic frontline is where ArmA is taken to a whole new level of combat.

 

Once a frontline system (subjective or objective) knows about both friendly and enemy forces. Their weights and location in relation to each other are averaged and form a frontline. This frontline could be visualized on the game map, or it could be hidden so that only the AI utilize it to make decisions. The whole objective being to improve AI combat awareness and tactics, but an added bonus of showing players the frontline dynamically (the visualization is where it is similar to the PVP mod "Dynamic Frontlines").

 

An example of how the system may look with all units known:

Example 4. The clusters of units have a high weight that pushes the frontline towards the enemy units of lower weight. The frontline moves towards the side that has the lowest weight and is constantly shaped, dynamically.

 

 

Another example showing how fog of war may look:

Example 5. The enemy cluster is now unknown/hidden (not spotted). The subjective frontline system only knows about the single unit and draws the frontline based on this. Frontline does not match the actual situation.

 

By simply looking at the two above examples you can immediately see why Recon roles are important. Recon can reveal enemy units and help develop the frontline. Observation helicopters, scouts, aircraft, UAV's now directly contribute to battlefield situational awareness. Cool huh?

 

 

What can the AI use this system for?

The AI would now know where the frontline is. In other words, a perpendicular line between friendly and hostile forces. It can use this simply to issue "scan direction" orders to it's troops, or form line and have the line formation aligned with the frontline. This way, AI would prepare for the threat by looking and forming up efficiently to combat the enemy. It also help the AI adapt to changes in the combat. If a new enemy force is revealed, the frontline may shift and cause the AI to shift attention to a new area.

 

Even better would be if the AI could compare the frontline angle to the angle of suitable cover objects. It would prefer hiding behind objects in relation to the frontline system. The AI could chose to garrison buildings and prefer occupying positions that face the known frontline. They could prefer to run for cover by placing something with fire geometry between them and the frontline.

 

Even more, it could be implemented into an AI commander system that coordinates groups of AI dynamically. Such a commander system would be able to detect areas of weakness and reinforce them, or spot enemy weaknesses and attack there, essentially flanking. In order for the AI to successfully flank, it needs to know where the frontline is, which comes from the dynamic frontline system.

 

CAS missions, artillery strikes etc could be dynamically generated to cope with sudden influxes of enemy troops.

 

Mission designers can set the "frontline resolution" to tailor the system to the scale of their mission. A huge war can have frontlines with segments that are 100's of meters long, or a small scale "capture the town" can have a 10m long frontline segment. The point being that having a very high resolution in a large scale operation may tax the system.

 

 

 

 

Anyways. I can't code, but someone who can may comment on this and see if it is doable. If there already is a mod with dynamic frontlines, then there should be a possibility to make a system like "ALiVE" to take advantage of that frontline system to command AI troops around the map in an... "intelligent" matter :) .

 

Feel free to give share your thoughts on this? I am rather serious about the idea and genuinely think it can work, even in arma 3, with the current engine. 

[escforreality 35]

Strike_NOR that is an excellent suggestion! Something like this natively handled by the engine could revolutionise the the AI. 

[Strike_NOR 898]

1 hour ago, escforreality said:

could revolutionise the the AI.

 

If anything at all, it has to be implemented as an optional module based system, much like the way one would use the alive mod. Hopefully, such a system could tap into the BIS "Dynamic Simulation" module to simulate map-wide warfare. 

 

Instead of placing all unit waypoints manually there could be an AI commander module. The mission creator would pop down a "Dynamic Frontline" module, set it up to be either objective or subjective (objective manages frontlines for both sides, subjective manages per side). Then pop down an AI commander module for each side (which sets aggressiveness, style (tactics) etc.). This way it would be an all out sandbox warfare simulator, which is completely random and adaptive each time.

 

By just using the frontline module alone, the AI would better react to threats, take cover properly and know where and how to flank. Even in squad on squad engagements.

[escforreality 35]

Alive mod is exactly what I was thinking of (I basically only play ARMA with it now), they clearly do something for their AI commander to pick where it fights but this is a great option to build on that theme. I'm not sure how they are doing it but what you are suggestion of the front line changing based on the AI knowledge of enemy composition is great.

 

39 minutes ago, Strike_NOR said:

By just using the frontline module alone, the AI would better react to threats, take cover properly and know where and how to flank. Even in squad on squad engagements.

But ignoring the whole Alive mod thing, this alone is a good enough reason to do it.  

[froggyluv 2136]

Have you tried HETMAN?

[Strike_NOR 898]

11 hours ago, froggyluv said:

Have you tried HETMAN?

 

No, but reading the description it seems to be a simpler version of ALiVE without unit virtualization. It essentially takes units from one side and sends them to player-determined markers in the order of choice of the player.

 

I am looking for a system that improves low-level AI also. I'd like the frontline to directly influence the status, arrangement and positioning of troops on squad level.

 

Picture the following: An AI squad is placed in kamino firing range destroyed village with a hold/guard (etc) waypoint. They run to position and halt there. Since the "frontline" is now surrounding them (no known threats) they spread out evenly behind cover, and in buildings and occupy areas that are facing the frontline. Basically this means they will establish a 360 perimeter defense.

 

Now introduce an enemy squad advancing from one side. As soon as the AI spots the enemy squad, the frontline dynamically shifts from omnidirectional to directional, between the known threat and allied forces. This means that whoever is facing away from the frontline, now automatically runs from their position towards cover/buildings on the side of Kamino village that are facing the frontline.

 

This is the kind of system I want. That the squad assumes 360 defense without intel. As soon as enemy is known, the frontline is established towards them. The entire squad moves to counter the threat, effectively bringing more firepower to face the enemy. This kind of reaction really opens up the possibility of flanking maneuvers. The pinned squad, approaching kamino village, can now send a few guys to the sides to effectively flank them.

 

In vanilla ArmA, some of the defending AI would just be wandering around the other side of the village, or be given orders to "attack" the enemy, meaning to move in their general direction until contact is made, then fire (dumb move). The smart thing would be to find some cover that is facing the frontline and be ready for the enemy to advance.

 

In the same scenario, if a new enemy group approaches from a different angle, as soon as they are spotted the dynamic frontline shifts more towards the average area of the two attacking groups. The defenders now adjust properly and some are sent to watch/take cover by the new frontline area.

[mad_cheese 593]

Probably been noted many times before band it may not be the time to expect changes in general, but:

 

Having multipe AT/AA units in a group is currently ineffective. Even with multiple targets around. Units will loften ock on to the same targets and shoot at the same time, which is a waste of extremely valuable firepower.

 

 

 

 

[nkenny 1057]

 

@Strike_nor
I will first give an outline of what it seems you are suggesting then some critique.  It seems that what you are describing is actually a learning heuristic combined with a dynamic response mechanism on two levels.
 

I. Recon
A SIDE will develop information regarding combat strength (in your nomenclature: weight) by interacting with enemy forces. The side maps out an enemy frontline in relation to its own. Frontline is by this a measured in three categories: (1) location and (2) size/expanse of danger area and (3) abstract combat strength contained within danger area. 

II. Tactics
In turn the AI will tailor low level response to information about dangerous, or potentially dangerous (as measured by abstract combat strength) along lines of (1) behaviour and (2) facing.
(1) Combat behaviour. Stance and use of cover; more dangerous zones demanding greater attention to use of cover.
(2) Combat facing. Group facing determined not by current waypoint, but by distance and combat strength of nearby combat foci.
 

III. Strategy
High level response would be handled by units linked to an administrative HQ (be it a module or actual unit or abstract SIDE) with the ability to consider a number of variables:
- Enemy combat strength
- Own combat strength
- Ability to project combat strength
- Sides rules of engagement

High level response could  include considerations of reconnaissance, resupply, reinforcement, and conservation of assets-- however I believe that this will vary greatly from mission to mission as to require a human level intelligence  or else strict human defined limits to be effective.

IV. Criticism
First I will consider practical implementation a to b, then consider the internal tactical and strategic benefits of such a system c to e. 

Practical implementation
a. If this system is applied on a general level it will break with many existing missions. High level command AI, such as enabled by the mod/script VCOM, does exactly that. Indeed the configuration, made by mission maker, necessary to establish sensible areas of operation may preclude it from being a time saving, drop-in mechanism.

In your own example this is evident. A unit set to a GUARD waypoint is a very different in tasking from one set to HOLD. The reason why the AI is set to this waypoint may will also differ! A single solution will not give answer to the myriad of scenarios possible in Arma3.

b. High level AI responses is not a quick fix. While the AI of Arma3 could do with a rough shaking, decrepitating the entire existing waypoint structure seems a project better suited for Arma4.  At this level of development, Arma3 is in in need of fixes and enhancements, rather than rewrites.

Internal tactical and strategic value
c. It seems like you are describing a new game mode-- focusing on information gathering and weighing options rather than immediate AI response and actualisation of offensive and actions. This is interesting, but harder to make into a generic one-size-fits-all package. If unfamiliar with mission making and mods I recommend checking: HETMAN and VCOM.  

d. Introducing an expensive  super structure to handle low level tactical choices is interesting. But what does it actually change? Already the AI will respond (in a directional) sense to known enemy forces. Units set to GUARD waypoints will move to support (in accordance to a combat strength evaluation) to enemy presence.  The biggest benefit seems to alter behaviour of AI patrols moving into enemy frontlines or to units patrolling within a zone which suddenly is marked as enemy dominated frontline. Certainly limmersive, but how expensive is this compared to the minor benefits gained? 

 

e. The theory of modern military strategy has moved away from envisioning frontlines.  Modern military forces are unable to muster the manpower necessary to populate them, and in any case revealing ones presence is to invite destruction by artillery or bomb.  Even so handling the vast range of variables that must be considered by a high level commander are beyond what a AI could be expected to manage.  If map specified Areas of Operations,  or limited support weapons and QRF forces are linked to an AI module then this ties neatly to c.  It sounds like a new game mode, rather than a unitary one-size-fits-all solution to our AI woes. 

V. Concluding
While the idea is nice it seems that you are describing something different to basic enhancements to player-AI interaction in Arma3.  A high level AI commander capable of smart development of information and responding to said high-level information is suitable for some, but not all scenarios played in Arma3.  Therefore it makes more sense to focus the lower tactical level of execution of common infantry tasks.

TL;DR:  Sounds like a fun game mode, but beyond the scope of improvements to AI decision making and basic player to AI interaction. 

 

-k 

[h - 169]

AI getting stuck after healing seems to still be a feature, since Alpha/beta or whatever it was..

Just observed it happening thusly:
- we all were prone after a short firefight
- nr 3 shouted 'injured!'
- ordered medic to heal, medic actually went to a nearby enemy body (the enemy ran pretty much over us = the injury) an healed it instead...
- 3 was obviously still injured so told him to FAK himself because medic started some Lazarus project apparently, 3 did as ordered, and then stood to crouch and FAK'd himself again at which point I already knew he got PTSD now and won't move. Which was true. He would follow no orders at all anymore, just stay in that one spot crouched.

- Left him there as enemy honeypot....

[froggyluv 2136]

5 hours ago, Strike_NOR said:

 

No, but reading the description it seems to be a simpler version of ALiVE without unit virtualization. It essentially takes units from one side and sends them to player-determined markers in the order of choice of the player.

 

 Mmmm, perhaps thats an option but you can load the mission the AI commander will choose everything without any input markers from me.

 

Quote

 

Picture the following: An AI squad is placed in kamino firing range destroyed village with a hold/guard (etc) waypoint. They run to position and halt there. Since the "frontline" is now surrounding them (no known threats) they spread out evenly behind cover, and in buildings and occupy areas that are facing the frontline. Basically this means they will establish a 360 perimeter defense.

 

Now introduce an enemy squad advancing from one side. As soon as the AI spots the enemy squad, the frontline dynamically shifts from omnidirectional to directional, between the known threat and allied forces. This means that whoever is facing away from the frontline, now automatically runs from their position towards cover/buildings on the side of Kamino village that are facing the frontline.

 

 

First off I really like your ideas and the direction of your thinking. But we got some problems here - would you want a 360 defense if 1 full direction was empty ocean or LOS block by row of buildings or sheer cliff? The terrain makes this sort of AI planning very difficult -hey its tough to get right in a 2d strategy game so no surprise its very tough for open world 3d shooter.

 

@Facing directions - sorry but the AI squads cant get that right now as it is youll have a few soldiers facing directly away from only known threat -I guess this is them trying to keep a 360 defense but it looks and plays bad.

@Garrison Buildings to Enemy Threat: Yes but many buildings dont have windows facing all sides, so youll have AI using useless building positions which essentially take them outta the fight. Works well if they are just fleeing as in self preservation tho.

[Strike_NOR 898]

3 hours ago, froggyluv said:

But we got some problems here

 

Thanks for the input and i challenge you with a quote.

 

Don't stop at pointing out problems, try coming up with the solutions

 

Loads of fps sandbox games already incorporate enemies using objects for cover. Even GTA has dynamic cover behind vehicles, objects etc. I dare to dream of BIS figuring out this for infantry gameplay too. (Dont confuse with wall-hugging, or clinging to objects mechanisms, but the idea that they recognize cover in the first place.)

[road runner 4344]

Given than Tanks DLC will be the last one created by BIS, and that they're now advertising for 3rd party DLC creators, it's quite obvious the current AI issues are never going to be fixed. 

 

it's now a case of " you can only piss with the cock you've got"

 

BIS are not going to fix the AI chaps, so better get used to these little motards!!

[Strike_NOR 898]

On 11/15/2017 at 3:21 PM, nkenny said:

*Snip*

TL;DR:  Sounds like a fun game mode, but beyond the scope of improvements to AI decision making and basic player to AI interaction. 

 

-k 

 

Tusen takk (Thanks for the rest of you) :)

 

Many very good observations and some sincere feedback.

 

Some of my inspiration came from the way frontlines worked in IL-2 Sturmovik (A WWII sim, where frontlines are more relevant to the time).

 

They way it worked was by placing map markers (flags) in red or blue colors, these would all have the same "strength/weight" and push the frontline to the "average" point between friendly and enemy flags.

 

Like you say, some big improvements are for Multiplayer TVT game-modes where the players decide where to go or where to send AI troops (player commander). This may allow players to act as special forces (recon, sabotage, deep-strikes etc) while AI populate the "generic infantry combat" role in the background.

 

I do notice that AI try to use cover by moving behind them and peeking (strafing sideways wile leaning) to look around corners, but it feels more random to me than consistent. One sorely missed aspect is AI recognizing specific objects as "good cover". AI would prefer hiding in a sandbag circle, rather than behind a stonewall if both are equally available to him - such things.

 

I do like your comments about using it for setting awareness, stance etc. 

 

Finally, it would have to be a module system because it would allow future missions to be developed WITH it, and older missions to work WITHOUT it :) It would also be a drop and setup module (setup side, faction, tactic style). (Asymmetric would seek to infiltrate enemy lines to cause havoc. Defensive would simply do all it could to hold the area it is given upon mission start and offensive would seek to expand.

 

It's easy to use your imagination and see a much better dynamic war going on when there is a frontline. Everything is "safe and calm" on friendly side, aware towards the line, and finally combat/aware/stealth if you are in enemy territory.