[WIP] bCombat infantry AI Mod [SP]

[bravo409 13]

I have do the video again the notes didn't save sorry

[ebarstad 18]

Thanks.

It's true for player controlled group, but should not happen for AI groups, at least i did not see this happen.

Can anybody confirm AI groups are not switching back from combat mode ?

please let me know.

I'm playing with the default bCombat settings and I just did a test with two AI groups facing off against each other. They start in SAFE mode, walking towards each other to a MOVE waypoint half way between each other. Upon contact, they go into COMBAT/DANGER mode but after all enemies from one side have been killed, the remaining soldiers switch to AWARE mode.

I think you asked about accuracy, too. Compared to vanilla, I'd say the AI isn't accurate enough, especially in CQB. In the same test described above, soldiers within 50-100m of each other were very inaccurate. Mind you, they were also using cover quite well, but the fire fight probably shouldn't have lasted as long as it did. In general, though, I prefer inaccurate AI to aimbots.

[bravo409 13]

So check my video when I just got home and they have notes in the video that pop up to tell you what I see or experiencing..https://www.youtube.com/watch?v=a-CxYtFsv-E

...........https://www.youtube.com/watch?v=5sJt8XXdmcg.....

Let me know if you dont see them thanks

[fabrizio_t 58]

I'm playing with the default bCombat settings and I just did a test with two AI groups facing off against each other. They start in SAFE mode, walking towards each other to a MOVE waypoint half way between each other. Upon contact, they go into COMBAT/DANGER mode but after all enemies from one side have been killed, the remaining soldiers switch to AWARE mode.

Thx ebarstad, this what i see as well.

Mr_centipede, is it possible that some enemy units were still around in your test, maybe at distance? I imagine group would be still locked into combat mode in that case.

I think you asked about accuracy, too. Compared to vanilla, I'd say the AI isn't accurate enough, especially in CQB. In the same test described above, soldiers within 50-100m of each other were very inaccurate.

Mind you, they were also using cover quite well, but the fire fight probably shouldn't have lasted as long as it did. In general, though, I prefer inaccurate AI to aimbots.

Yes, accuracy is indeed lower compared to vanilla.

It's capped to 70% and depending on suppression is reduced.

It is true than in CQB accuracy looks lower, but it happens with vanilla AI as well, it's not something introduced by bCombat.

With bCombat it is just more noticeable.

I think the reason for this is units fire in full-auto on close distance and accuracy is being affected by this.

---------- Post added at 10:19 ---------- Previous post was at 09:43 ----------

So check my video when I just got home and they have notes in the video that pop up to tell you what I see or experiencing..https://www.youtube.com/watch?v=a-CxYtFsv-E

...........https://www.youtube.com/watch?v=5sJt8XXdmcg.....

Let me know if you dont see them thanks

I've seen the second video now.

About the final 1-on-1 clash, you posted this note:

i didn't see any of the colors turn red or blue so just wondering when it should

Consider that this clash happened from very close distance.

As per default bDetect is set up to ignore any bullets fired from < 20m., which can well be the case.

This is done for mainly 2 reasons:

* the time for a bullet to travel such distance can be well under a single game frame, so it's highly likely to be missed anyway.

* In my opinion at this distance there's no need to trigger suppression effects, nor to alter accuracy. Units should fire and kill, or be killed.

Any evasive maneuver would be impossible.

For higher distance you would have seen some suppression, but probably not as high as you expect.

The firefight lasted 5 seconds, 6:06 to 6:11, both units knew eachother and were frontal.

In such a situation, assuming linearity for sake of simplicity, the maximum amount of suppression a healthy unit may suffer from bullets would be:

(( <time> / 0.2 ) * <bcombat_penalty_bullet> * <1 - skill> )  %

So effect would highly depend on skill:

For skill=0.7:

((5 / 0.2) * 4 * (1 - 0.7)) = 30 %

Blue balloon.

For skill=0.5:

((5 / 0.2) * 4 * (1 - 0.5)) = 50 %

Blue / yellow balloon.

For skill=0.3:

((5 / 0.2) * 4 * (1 - 0.3))  = 70 %

Yellow balloon.

For skill=0.1:

((5 / 0.2) * 4 * (1 - 0.1)) = 90 %

Red balloon.

This was a simplified example.

Due to penalty being applied non-linearly in v0.12, these percentuals are further damped.

For skill=0.5 morale hit would be roughly 40% instead of 50%, for instance.

Obviously other environmental events may add to this (explosions, wounds, casualties, ...).

Edited November 14, 2013 by fabrizio_T
Fixed wrong calculations. Recovery rate not supposed to kick in.

[mr_centipede 31]

I'll try to retest the combat behaviour, maybe it was just on my end... will report later.

[fabrizio_t 58]

I'll try to retest the combat behaviour, maybe it was just on my end... will report later.

Thank you.

[mr_centipede 31]

Yep, it was on my end. It's Hetman's fault. I'm sorry for using with other mod, but in my defense, I assumed (wrongly) that HAC wont activate if they were not called, as in have the leader variables. But it seems it still active even without the HAC leader. So sorry for the false alarm.

Still regarding not exiting the combat mode, if the enemy group is fleeing and no longer a threat ( no shots were fired), AI group seems taking forever to revert back their combat mode. At least for me seems never, and the enemy is very far away already.

But original problem of AI not exiting combat mode after a firefight (no enemy left) is false alarm on my end. A conflict with Rydgier's HAC mod. Sorry about that

[fabrizio_t 58]

Yep, it was on my end. It's Hetman's fault. I'm sorry for using with other mod, but in my defense, I assumed (wrongly) that HAC wont activate if they were not called, as in have the leader variables. But it seems it still active even without the HAC leader. So sorry for the false alarm.

Still regarding not exiting the combat mode, if the enemy group is fleeing and no longer a threat ( no shots were fired), AI group seems taking forever to revert back their combat mode. At least for me seems never, and the enemy is very far away already.

But original problem of AI not exiting combat mode after a firefight (no enemy left) is false alarm on my end. A conflict with Rydgier's HAC mod. Sorry about that

No problem, nice to hear that it was sorted.

About AI being slow to fall back to aware ehen enemies are fleeing: it is very likely that fleeing units are still raising AI events on enemies.

If so the latter would keep in danger mode as long as the former are in sight.

Since bCombat raises max. spotDistance a bit (from 0.4 to 0.6), fleeing units may have to move farther to become undetected.

Edited November 14, 2013 by fabrizio_T

[b0s 18]

Good luck with this! A all around AI mod is much needed. Any plans for buildings... like house search and garrison?

[ebarstad 18]

fabrizio_T: Are AI leader-created fire teams out of scope for bCombat? For example, using bCombat in CQB, I'll often see one AI sent to flank an enemy, but it would probably be safer if two were sent instead, as a team. I hesitate to bring up the comparison, but I know GL4 AI leaders will actually assign units to coloured teams, but I don't recall if the AI actually does anything with those assignments (i.e., flanking, overwatch, etc.).

[fabrizio_t 58]

fabrizio_T: Are AI leader-created fire teams out of scope for bCombat? For example, using bCombat in CQB, I'll often see one AI sent to flank an enemy, but it would probably be safer if two were sent instead, as a team. I hesitate to bring up the comparison, but I know GL4 AI leaders will actually assign units to coloured teams, but I don't recall if the AI actually does anything with those assignments (i.e., flanking, overwatch, etc.).

I imagine it would be possible to implicitly order some unit to follow the one which has been already commanded to attack/flank, but i'm not sure how well it would work.

About CQB, i'm even doubtful that flanking is any useful with ArmA 3 AI, but that's a personal opinion.

---------- Post added at 20:20 ---------- Previous post was at 20:19 ----------

Any plans for buildings... like house search and garrison?

Not sure at the moment, plenty of stuff to be finalized already.

Not to speak about overall balancing ...

[ebarstad 18]

I imagine it would be possible to implicitly order some unit to follow the one which has been already commanded to attack/flank, but i'm not sure how well it would work.

About CQB, i'm even doubtful that flanking is any useful with ArmA 3 AI, but that's a personal opinion.

Ok, no worries. Thanks for the answer.

Are you interested in pathfinding at all (bCombat vs Vanilla)? If so, I can upload some comparative screens of the scenario I run which shows (in map view) the paths each unit took during a fight. Here's an example: http://steamcommunity.com/sharedfiles/filedetails/?id=193520118

[fabrizio_t 58]

Ok, no worries. Thanks for the answer.

Are you interested in pathfinding at all (bCombat vs Vanilla)? If so, I can upload some comparative screens of the scenario I run which shows (in map view) the paths each unit took during a fight. Here's an example: http://steamcommunity.com/sharedfiles/filedetails/?id=193520118

I'm very interested into this kind of comparison.

Any deterministic way to do analysis is welcome.

Please post your findings.

[bravo409 13]

Fab did my videos help give you a good idea what we are seeing? Did I test this right is there anything I should be testing thanks?...

[ebarstad 18]

I'm very interested into this kind of comparison.

Any deterministic way to do analysis is welcome.

Please post your findings.

Cool. I just ran 6 quick tests of the same scenario. Here are the screens:

Vanilla

Test 1 (Run at 4x speed)

Test 2 (Run at 4x speed)

Test 3 (Run at 4x speed)

bCombat

Test 1 (Run at 4x speed)

Test 2 (Run at 4x speed)

Test 3 (Run at 1x speed)

I think the biggest thing I discovered is that I should have run the tests at normal speed (d'oh) -- see how different bCombat Test 3 is from the rest.

In general, in vanilla, the AI noticed each other later, killed each other on first contact, then started manoeuvres in and out of cover to get to their waypoint in the centre of town. With bCombat, there were some first contact kills, but the AI began moving to cover typically before the first shots were fired. In fact, in the last test, BLUFOR noticed the enemy much earlier and went into cover before OPFOR even noticed them. They then moved closer to the OPFOR (out of sight) and came out at a different position, from which BLUFOR managed to kill most of the OPFOR squad. BLUFOR then moved much more cautiously toward their waypoint, spreading out a little more.

If you look at the large images, you can also see where a unit was killed as well as from where they were shot. It's all a bit cluttered.

[mr_centipede 31]

Cool. I just ran 6 quick tests of the same scenario. Here are the screens:

Vanilla

Test 1 (Run at 4x speed)

Test 2 (Run at 4x speed)

Test 3 (Run at 4x speed)

bCombat

Test 1 (Run at 4x speed)

Test 2 (Run at 4x speed)

Test 3 (Run at 1x speed)

I think the biggest thing I discovered is that I should have run the tests at normal speed (d'oh) -- see how different bCombat Test 3 is from the rest.

In general, in vanilla, the AI noticed each other later, killed each other on first contact, then started manoeuvres in and out of cover to get to their waypoint in the centre of town. With bCombat, there were some first contact kills, but the AI began moving to cover typically before the first shots were fired. In fact, in the last test, BLUFOR noticed the enemy much earlier and went into cover before OPFOR even noticed them. They then moved closer to the OPFOR (out of sight) and came out at a different position, from which BLUFOR managed to kill most of the OPFOR squad. BLUFOR then moved much more cautiously toward their waypoint, spreading out a little more.

If you look at the large images, you can also see where a unit was killed as well as from where they were shot. It's all a bit cluttered.

What script did you use to make that? Could be useful for us the rest of the testers as well

[gagi 10]

+1 this path script is great !!!

[ebarstad 18]

What script did you use to make that? Could be useful for us the rest of the testers as well

I used Kronzky's Unit Tracker script.

[lordprimate 159]

hmm, I just sat down to do some testing, you had to have modified it to get it to work. as i get errors using that script.

[fabrizio_t 58]

Fab did my videos help give you a good idea what we are seeing? Did I test this right is there anything I should be testing thanks?...

I still have to see first video.

Care to share the mission featured into that?

EDIT: i've seen the video and your notices.

I noted 2 problems:

* your sniper not opening fire

* your sniper not firing on the target you ordered to.

So i did some experiments, trying to mimic your mission setup (with bCombat v0.13, which is almost the same as v0.12 targeting wise):

* I placed a group composed by player + BLUFOR sharpshooter on the hill near Agia Marina, as well as a OPFOR group into the village, near the bridge.

* Distance to enemy was a little over 500m., so a bit off bDetect maximum range preset, but this does not matter since we will deal only with sniper actions.

* I've ordered group into STEALTH (\-7-1), HOLD FIRE (\-3-2)

Then:

* Took binoculars and revealed some enemy units

* Assigned some target to the sniper (2-2-<some target>)

* Ordered FIRE (\-3-1)

after doing that my sniper started taking fire on target as expected and switching target as commanded.

So basically i can't reproduce the issue.

I'll post a video about this later.

By the way, i was able to command AI snipers open fire on target up to 650m. with no problems apparently (except they don't hit basically anything).

---------- Post added at 09:55 ---------- Previous post was at 09:34 ----------

Cool. I just ran 6 quick tests of the same scenario. Here are the screens:

Vanilla

Test 1 (Run at 4x speed)

Test 2 (Run at 4x speed)

Test 3 (Run at 4x speed)

bCombat

Test 1 (Run at 4x speed)

Test 2 (Run at 4x speed)

Test 3 (Run at 1x speed)

I think the biggest thing I discovered is that I should have run the tests at normal speed (d'oh) -- see how different bCombat Test 3 is from the rest.

In general, in vanilla, the AI noticed each other later, killed each other on first contact, then started manoeuvres in and out of cover to get to their waypoint in the centre of town. With bCombat, there were some first contact kills, but the AI began moving to cover typically before the first shots were fired. In fact, in the last test, BLUFOR noticed the enemy much earlier and went into cover before OPFOR even noticed them. They then moved closer to the OPFOR (out of sight) and came out at a different position, from which BLUFOR managed to kill most of the OPFOR squad. BLUFOR then moved much more cautiously toward their waypoint, spreading out a little more.

If you look at the large images, you can also see where a unit was killed as well as from where they were shot. It's all a bit cluttered.

I find curious that by running at 1x you saw that kind of difference, your FPS are fairly high, with a min. over 30FPS, so no (vanilla) AI degradation should occur, nor bDetect adaptive degradation.

Judging by screenshots bCombat does make units move / flank / divert / spread more.

It would be interesting to see what vanilla AI is doing at 1x.

Edited November 15, 2013 by fabrizio_T

[ebarstad 18]

hmm, I just sat down to do some testing, you had to have modified it to get it to work. as i get errors using that script.

No, I didn't change anything. It throws an undefined variable for me at the start of my mission, but it works fine.

---------- Post added at 09:30 AM ---------- Previous post was at 09:26 AM ----------

I find curious that by running at 1x you saw that kind of difference, your FPS are fairly high, with a min. over 30FPS, so no (vanilla) AI degradation should occur, nor bDetect adaptive degradation.

Judging by screenshots bCombat does make units move / flank / divert / spread more.

It would be interesting to see what vanilla AI is doing at 1x.

The fps are high because I watched the whole mission from the Map screen, but (as you say) they probably wouldn't have dipped below 30fps anyway, with only 16 units. I'm going to run the tests on normal speed to see if there's much difference.

[fabrizio_t 58]

Testing v0.13

Edited November 15, 2013 by fabrizio_T

[ebarstad 18]

Ok, I did another set of tests, all at 1x speed.

----

Vanilla

Test 4

- 6 minutes

- 1 BLUFOR survivor

Test 5

- 7 minutes

- 1 BLUFOR survivor

Test 6

- 4 minutes

- OPFOR wiped out at first contact

- BLUFOR friendly fire event

- 7 BLUFOR survivors

bCombat

Test 4

- 9 minutes

- 5 BLUFOR survivors

Test 5

- 9 minutes

- 6 BLUFOR survivors

----

I think one of my favourite parts of bCombat is that the AI units still end up at their waypoint -- even though they come under fire, spread out, and hunt down the enemy. I think Fabrizio_T has created a pretty awesome balance: on the one hand, the player gets better AI and, on the other, the missions are less likely to get broken because the AI still respect their waypoints. Win-win.

[fabrizio_t 58]

Ok, I did another set of tests, all at 1x speed.

----

Vanilla

Test 4

- 6 minutes

- 1 BLUFOR survivor

Test 5

- 7 minutes

- 1 BLUFOR survivor

Test 6

- 4 minutes

- OPFOR wiped out at first contact

- BLUFOR friendly fire event

- 7 BLUFOR survivors

bCombat

Test 4

- 9 minutes

- 5 BLUFOR survivors

Test 5

- 9 minutes

- 6 BLUFOR survivors

----

I think one of my favourite parts of bCombat is that the AI units still end up at their waypoint -- even though they come under fire, spread out, and hunt down the enemy. I think Fabrizio_T has created a pretty awesome balance: on the one hand, the player gets better AI and, on the other, the missions are less likely to get broken because the AI still respect their waypoints. Win-win.

My favourite part is the 9 minutes vs 6 minutes duration. 50% more :)

With ArmA WWII would have lasted about 30 minutes ...

[fabrizio_t 58]

Testing v0.13

Some pitched fighting in urban area.

Suppressive fire is on, as you may notice by the sheer number of gunshots.

units do blind firing and / or shoot in movement when target is not in sight or in order to advance on threat.

You may notice a (kamikaze) BLUFOR unit advancing firing full-auto at 1:27 to 1:30.

Aggressive grenade throwing is used to clear resistance.

You may notice a quick close distance grenade duel at 1:00 to 1:05.

---------- Post added at 22:54 ---------- Previous post was at 22:15 ----------

Some WE combat photography :)

I still think ArmA 3 is a looker.

Surrendered

Killed

Fragging.

Edited November 15, 2013 by fabrizio_T