Search the Community
Showing results for tags 'ai'.
-
I've been looking into how AI units detect aircraft at night, and the results are actually really interesting. After checking through the detection.lua file and running a bunch of tests, I found that sky_bodies_illumination_factor is set to true, which means the moon phase directly affects how visible you are to AI. The game uses four moon stages quarter, half, three quarter, and full, with each one adding more light that helps AI spot you more easily against the night sky. sky_bodies_illumination_factor = true, other_light_sources_illumination_factor = true, --no effect if sun_illumination factor disabled darkness_threshold = 0.0, --no effect if sun_illumination factor disabled What stood out the most in testing was how external lights change your visibility. The parameter light_source_detection_distance is set to 10,000 meters when light power is at full, and my tests showed that position lights make you a lot easier to see and far more likely to be engaged by AI. Strobe lights also increased detection, but not as quickly as the position lights, while formation lights barely made a difference. as_light_source = { light_source_detection_distance = 10000.0 --for light power = 1.0f }, The file also shows that other_light_sources_illumination_factor is set to true, which means your own aircraft lights actually light up your airframe and make you glow against the dark background. In other words, you are basically making yourself a visible target when your lights are on. Another important part is the darkness_threshold value, which is set to 0.0. That suggests AI detection does not completely disappear at night. It is just greatly reduced unless you are illuminated by your own lights or the moon. max_detection_distance = 50000.0, --m, absolute limit But detection times increase significantly at longer ranges average_det_time_max_dist_0 = 1.0, --s, average detection time of target ahead at the maximal distance average_det_time_max_dist_180 = 10.0, --s, average detection time of target behind at the maximal distance Overall, both the code and testing show that flying with lights on at night, especially under a bright moon, makes you far easier to detect and engage. What should be the cover of darkness actually turns into a disadvantage if you are not careful with your lighting. Contributer: @mach_point_yuh, @lightningw0lf, @zenki1419
-
Can we get an update on one of the biggest issues with the current weather system, please? AI can still see through clouds, IR missiles can still track through clouds. None of the visually interesting cloud presets really work for singleplayer and cooperative play. WWII and modern ground attack are notoriously frustrating without clouds blocking AI's vision and weapons/sensors. Heatseekers (especially IR SAM) have an unfair advantage that flat out prevents certain scenarios to be playable in certain weather conditions. I would appreciate an update on the matter. I think it's one of the bigger long-standing issues DCS is plagued by.
-
Hello, Since ED is currently prototyping new ATC and presumably other AI interactions, I hope I can provide some useful suggestions. I'm very critical of LLM "AI," but for video games - especially dynamic and immersive simulation environments like DCS - the technology is, in my opinion, extremely useful. To my surprise, the technology has now reached a point where it is also technically viable for a product like DCS, with very lightweight and fast models that can run locally on client machines, thereby solving a lot of economic obstacles. Also, most of these models are open source and relatively easy to implement. The cornerstone would be TTS/STT on the local machine, meaning speech generation and natural speech processing from the player (through a microphone). Models are now fast enough to run alongside DCS without significantly impacting performance. However, minimum system requirements would increase, as at least 2 CPU cores and a few GB of RAM would need to be dedicated to this task. This approach would not allow free-flowing conversations with the AI. The logic would still be scripted. A full LLM capable of handling open-ended conversation with the player would require a high-end workstation PC if run alongside DCS. However, models are becoming more efficient and average player PCs are becoming more powerful. ED should therefore anticipate the feasibility of locally running LLMs to further enhance the experience. In the meantime, ED could create an API for external LLMs to hook into their system, not to completely take over the logic but to augment the scripted logic with flair, personality and more dynamic conversations. Such an external LLM could be cloud-hosted (e.g., using the user's paid account with providers like OpenAI) or locally hosted, either on the player's machine or on a dedicated PC on the network. The proposed system would be fully multiplayer-compatible, since all communication with the AI is exchanged as text, while TTS and STT are handled locally on each client. If a larger LLM is desired for more free-flowing conversation, it could be run either on the server or in the cloud, with the server API managing the interaction between the game and the LLM. Thanks for reading. PS: One major advantage of this hybrid approach is that the LLM component could be easily maintained and expanded by the community. With a clean API separation, there would be no security risks for multiplayer or core gameplay, but plenty of room for creative extensions. For example, different LLM “personalities” could be developed - from a strict flight instructor helping beginners step by step, to a more casual RIO adding flavour and banter to the experience. Neither development nor maintenance of these LLMs would be on ED, as long as they provide a robust API for the external LLM to hook into.
-
@Flappie@BIGNEWY hello ,bug report: (those bug make the mi24 problematic ) issue when player play as pilot and the CPG is an IA . ->in a mission(iff/ia realistic ) because of the partial vision obstruction petrovich can't see a bmp1 or tank yet if i take place as gunner i can see them well, -> also the iff ia is not as good as a cpg player , i can identify a unit farther than petrovitch at 100% , the identification range must increase , also petrovitch should tell if the ia is radar sam (sometime i can see shorad with radar , ican't tell if it is roland or sa8 but i know it one of them as cpg , because i see a radar rotating , the ia must report at least sam with radar rotating at least you know you are in danger . -> petrovitch UI need to be redesign with the ability to filter the type of unit : you want report of only enemy, only friend ,only neutral , or all you want only air air unit , only sam , only aaa , only armor , only ifv , only infantery etc... (it is very usefull in area full of unit) -> petrovitch ui we need to tell the ia like a radar how wide he scan etc.. 10 ° 30 ° 45 ° 65 ° wide (also the elevation almost like a fighter radar below horizon or only top of it or both), when we ask to scan he only report sitac but doesnt not try to track , we should have the ability to scan at specific area or an area around our own helicopter as referential (2 type of scan) , when he scan you can see the aiming mark moving around -> petrovitch need to report the angle and distance of each target(or group with vehicule number)(static or moving how fast and direction also he tell if it is partially obstructed before you shoot not at the last second!!) it see to build a picture -> when you track a target the array of target should stay as an option so we can switch target fast -> the list of unit in UI maybe should become a 2D array -> petrovitch need to report gun fire close to us and from where it come ,same for missile , and filter better if it is not running at us and if it is close fire or middle or far --------- issue : when the player play as CPG and the pilot is an IA . ->now if you play as cpg the ia pilot has some bug when you see through the optic , if you ask to change the altitude the ia is unable to change it , you need to to stop using("enter the optic") the optic so he can change the altitude , the expected behavior "he changed the altitude even when i look in the optic" ->also we need an ui alternative where you can push all command directly with key ,so you no longer need to change the UI mode" like hover/ hover transition/combat) we need access directly to all key directly -> the pilot ai need to report gun fire/ missile close to us and from where it come , and filter better if it is not running at us and if it is close fire or middle or far , and start doing evasive action -> the pilot ia for some reason try to move the altitude higher than 80 meter when attacking in dynamic if player play cpg , this is huge issue when you need to fly very low (NOE) we need the ability to use cpg as player at NOE -> when we switch from pilot to cpg , petrovitch should keep the missile selector to last missile used , currently it go back to zero it is a problem when you have to switch missile selector everytime in middle of fight because of switching pilot/cpg... i will make a video soon , provide track , and image
-
Hi everyone, I'm encountering an issue with the AI - they appear unable to accurately deliver either the CBU-99 or Mk 20 Rockeye II. In the track below there is no wind or turbulence and apart the absence of clouds uses default weather conditions. With the Mk 339 fuse (set to default settings), the AI always seems to drop the bomb long. A similar thing can be seen with the FMU-140 (again, using default settings). With the exception of the BLG 66, all other cluster bombs (including unguided ones) do not seem to suffer from the same issue and the AI is able to employ them with far better accuracy. AI_Mk20_Mk339_350kts_min.trk AI_CBU-99_Mk339_350kts_1kft.trk AI_CBU-99_FMU-140_350kts_2kft.trk AI_CBU-99_FMU-140_450kts_2kft.trk AI_Mk20_Mk339_450kts_min.trk
-
First and foremost, I love DCS World and I play it very active with my Friends. This is also my first bug report since 2011 and for me to do this, something big has to have happened... While things like the TD Box in the Jeff were annoying, currently, the AI does not seem to be able to detect an Active Radar Homing Missile. Tests were done with the AIM-120B, AIM-120C, AIM-54C, R-77, SD-10 and R-77-1. Track will be provided as attachment. I would like to notify people that Tests with a clean DCS World install have been conducted. The Track File provided has Mods installed, but it is not the cause of the Bug. Friends found the same issue and during a public questioning on Spuds Server, Killshot also tested it and found the issue. Perhaps I'm really complaining on a high note here, I really don't want to upset anybody but given the lack of attention to this bug on the various Discords and Forums... Am I the only one who thinks that such a Bug is... borderline unacceptable to be around for one Month? I fly DCS a lot, and sure it doesnt stop you from playing the Game itself. But Fox 3 BVR is one of the major parts of Modern Air Combat. And for Singleplayer Mission/Campaign creators like me and my Friends, this is almost game-breaking since it completely destroys a properly created dynamic Combat environment and ruins realism and immersion as a Fox 3 Missile is guaranteed to hit an enemy Aircraft currently... Lastly, I forgot to add that, as an experienced Mission maker, I of course tripple checked the correct "Reaction to Threat", "RoE" and "Usage of Weapons" Commands for the AI. I have also tested it with every AI Difficulty level. I hope this brings some attention to the Issue so that it is at least patched in the next big Update... As of right now, we unfortunately cannot play any modern BVR Sorties. Fox3.trk
-
Or even just ground unit sectors in general. Can we either assign sectors of fire per waypoint or request the devs to rework the logic? Attached are some examples of what I'm talking about. I know, ground combat is lame and a low priority compared to adding some feature to a hornet. But this does need to be addressed as ground units die often because they're looking in the wrong spots. The orange triangle should be the units sectors of fire. The blue line are where the barrel is pointed. These AI skill is set to excellent by the way. The image with the Brads. The far left one is looking far right even though there is an enemy to it's left about to engage them. The image with the Abrams. The lead vehicle should be looking forward of the element, the second would be looking left or right of forward movement. Both forward moving vehicles are looking behind them while there is an enemy to the front. The rear vehicle should be pulling rear security but is actually looking at their forward movement. Thanks
-
This was pointed out by Cowboy 307 on Discord. Not sure if aircraft specific, but when doing the F16 Air Refueling Instant Actions on Caucasus and PG the AI F16s continue to get in line for refueling even if you call Intent to Refuel far out like in the PG instant action. Weird part continues when you abort refueling, the AI F16s will also abort but stay with the tanker. When calling for Intent to refuel again the AI F16s will go right back to refueling. The only way to stop this loop so you can refuel is to get in line right behind the AI and as soon as they get done call for Ready Precontact. Attached are the two tracks and should be around 5 minutes each give or take. Caucasus AAR F16.trk PG AAR F16.trk
-
aar..log In the instant action AAR training missions on the Caucasus as well as the PG map, AI cycles between lead and wingman refuelling for a couple of seconds before rotating to the other aircraft seemingly in a infinite loop. Tried to butt in but that doesn't seem to work. Honestly miss the return to pre-contact message since I am not able to get there.
-
I have an AI aircraft waiting on the runway, ready to take off. However, it should wait until a unit is in its zone before it's allowed to actually take off. Is there something in the Advanced Waypoint Options to do this? I could delay its take off time, but there's no real way of knowing when the unit will arrive. Failing any other options, I guess I could do the ol' switcharoo and get rid of the first plane that's waiting and spawn a new one when the unit is in the zone.
-
Hi everyone, Despite the following being in the changelog: I am still experiencing issues with 5-inch/38 calibre guns, as fitted to the USS Samuel Chase and Essex-class. In the below tracks I have a single aircraft approaching the USS Samuel Chase from the starboard rear-quarter. Wind is calm, the ship is not in motion and I've set the weather to clear skies. In both cases the aircraft approaches from the same direction and remains in level flight, the only difference between the 2 tracks is the altitude the aircraft flies at. While the 3-inch/50 guns engage more-or-less as expected (when the aircraft briefly exits their engagement zone, they cease tracking and reset to their default positions instead of attempting to continue tracking the target - so far only the KS-19 appears to do this) in the 6kft track the 5-inch gun never engages, despite doing so in the 2kft track. 6000 ft should be more than within the capability of 5-inch gun (in-fact it outranges and has a higher ceiling than the 3-inch guns - compare this table (5"/38) with this table (3"/50)). With the Essex (which I've separately reported here) the guns track the target in both cases (the directors though do not), but only in the 2kft track do they actually engage. However, even so, they take a long time to commence firing (the Bofors 40 mm actually opening fire first) and they only get a single salvo off before resetting(-ish) and no 5-inch gun further attempts to engage the aircraft. In the 6kft track, they track the target, but never engage and simply reset, pretty much exactly mirroring this bug report for ground-based AAA - where so far only the KS-19 behaves as expected. SamuelChase_5-inch38_Engage_2kft.trk SamuelChase_5-inch38_NoEngage_6kft.trk Essex_5-inch38_Engage-ish_2kft.trk Essex_5-inch38_NoEngage_6kft.trk
-
I noticed this bug in the Hunters Over the Yalu Campaign Mission 11. The Mig-15 AI is set to ace level and randomly crashes itself into the ground at the beginning of the mission when the player flight is taking off. The chance of the AI crashing itself is random as reflying the mission does not always result in the AI crashing. It seems to happen roughly 20 - 30% of the time. Track file and YouTube video of the bug attached. Recorded on DCS 2.9.17.12034. caseyjonesdead.rar
-
Pretty simple request, currently there is a trigger action called "UNIT AI SET LIFE" if this function could be exposed to the LUA api that would be incredibly useful for a variety of scripts/scenarios
-
У нас есть тяжелый двухвинтовой вертолет для армии США, давайте предоставим такой же для ВМС.
-
Hi everyone, I've noticed that in some testing scenarios, large-calibre WWII AAA will only track targets without ever actually firing - instead at some point they stop tracking and reset (seen in the. Sometimes, guns will start engaging but will only fire a few rounds before ceasing. Adding the Kommandogerät 40 seems to resolve the issue for German Flak guns and the allied director resolves the issue for QF 3.7". However, the Kommandogerät 40 does not resolve the issue with either the 75 mm Type 88 AA gun, nor the 8 cm/40 3rd Year Type gun. However, the director shouldn't be required for guns to actually start firing (especially if they've detected and are tracking the target) - instead directors should make guns more accurate. The same issue also applies to ships (namely the Essex), which I've reported here, but have yet to make extensive tracks (I've seen it sometimes engage higher altitude targets, but my prevailing experience is that 5"-guns don't engage reliably, nor do the directors track targets reliably). The same issue does not occur with the KS-19, which is perfectly capable of engaging targets without its SON-9 director. 8.8cmFlak37_Engage-ish.trk 75mmType88_Engage-ish.trk 80mm3rdYearType_Engage-ish.trk KS-19_Engage.trk 8cm3rdYearType_NoEngage_3kft.trk 8cmFlak18_Engage_3kft_Kdo.trk 8cmFlak18_NoEngage_3kft.trk 75mmType88_NoEngage_3kft.trk KS-19_Engage_3kft.trk KS-19_Engage.trk
-
Hello, pls comment this. So no any opponent for the Corsair? Overall, I wouldn't really like shooting the second Corsair. We can as well talk about Corsair vs. Mirage dogfights, e.g. Generally, it doesn't make sense, but maybe something wasn't said or in this interview they only talk about the work of the ED team and not other developers. So it will be or not?
-
I am trying to understand what is the logic happening behind AI Task Push and Triggered Actions. I created a mission with several triggers. Each trigger executes "AI Task Push" "Attack Ground Unit N". As far as I can see when AI Task Push happens, the unit executes exactly the task and if it a task was already in execution it just forgets it completely. Example: Trigger 1 happens, AI starts attacking group 1 Trigger 2 happens, AI forgets about group 1 and starts attacking group 2 When group 2 is dead, AI returns to its route and forgets about group 1 Can I make it not forget the task it was doing and return back to it?
- 1 reply
-
- mission editor
- ai
-
(and 2 more)
Tagged with:
-
So I am looking at making a mission where I need the Puma helicopter. I remember there being a French helicopter mod pack. However, I cant find it anywhere. Does anyone have a link? Or is a dead mod? Thanks a bunch
-
Hi everyone, I'm experiencing an issue with AWACS that go for AAR. The problem is that shortly after they connect they "jitter" causing them to disconnect. I've attached a short video of the issue and the mission used to reproduce it. I've experimented with different altitudes and speeds, but it had no effect. ps: in the video I'm running the sim at 3x but the issue occurs even at normal speed Thanks for your help. awacs.mp4 awacs_refuel.miz
-
Hi everyone, I've discovered an inconsistency with AI-fired AGM-84Ds. When "Attack Group" is used, the missiles always only strike the group leader, ignoring any other ships (even if those other ships have larger radar cross sections and are directly in the path of the missile - in the attack group track, you can see missiles overfly larger escorting ships on their way to the group lead). Even with a far tighter formation, the missiles only track the group lead, ignoring any other ships. This they do, even if other ships are obscuring the group lead. As if the missile magically knows where the group lead is and its seeker isn't being modeled. However, when "Search then Engage Group" is used, the missiles more randomly target ships in the formation, leading to every vessel being hit at least once (with the escort ships, which are both larger and closer to the missile's flight path) having more missiles target them. This is more expected behavior of the Harpoon and seems to better replicate the Harpoon's seeker. The bug here is that missiles fired by the AI using "Attack Group" should behave like they do when fired in "Search then Engage". For anyone wondering why I wouldn't just use Search then Engage every time. Well, search then engage requires the firing unit to detect the target themselves, which the low-flying aircraft will only do at fairly close distances. By using Attack Group, I can replicate pre-planned targets (where the Harpoon is fired at a pre-planned waypoint) - allowing the launching platform to remain in EMCON and/or replicate targets transmitted to shooters by other AI aircraft. AGM-84D_AttackGroup.trk AGM-84D_AttackGroup_TightForm.trk AGM-84D_SearchThenEngage.trk
- 1 reply
-
- agm-84d
- search then engage
-
(and 2 more)
Tagged with:
-
The big hill at the NE end of Spangdahlem means that AI aircraft always crash when trying to land on runway 23. This seems to be 100% reproducible, at least with the Hornet. (It's actually a pretty tricky approach for human pilots, too, but it's suicide for the AI). Broken AI Landing Spangdahlem.miz
-
Hi everyone, For whatever reason, I cannot seem to get the AI F-4E-45-MC to use Paveway II and Paveway III series laser guided bombs against points on the ground. Upon reaching the waypoint with the bombing task, the aircraft simply turn around, follow the rest of their waypoints and land. I've tried numerous options on the weapon to use (AUTO just ends up using the internal cannon, I've also tried guided bombs, guided and bombs) to no avail, override AI attack avoidance decisions also doesn't appear to change the behaviour (and in any case, this mission is empty apart for the AI aircraft - there's no threats). It also doesn't seem to matter if the task is set to CAS and whether or not you use the CAS, Attack Group/Unit or Search then Engage tasks, none seem to work. The issue seems to be fairly exclusive to Heatblur's F-4E-45-MC (though the F-5E-3 also failed to drop the bombs, but then, there wasn't anything to provide designation). Attached is a track showing what happens with the GBU-12, the issue is also present on the GBU-10 and GBU-24B/B (the latter should actually be an A/B, but that's for HB to fix). EDIT: The problem is also happening with a few other weapons as well, it seems mostly isolated to guided air-to-surface weapons, with exceptions (so far the AI is fairly reliable when engaging with the AGM-12). In AI_F-4E_AGM-65D_nofire.trk, you can see that the AI won't fire Mavericks at a tank platoon, the same is true for every other Maverick the F-4E-45-MC has available. It also happens regardless of what the Maverick is mounted on (be it the LAU-117, the LAU-117 on the special weapons adapter or the LAU-88). In AI_F-4E_AGM-45A_nofire.trk, you can see that the AI won't fire an AGM-45A at the SON-9 (the guidance section is set appropriately), I have got it to fire at the Big Bird and the P-19 however - it seems a bit inconsistent. I also can't get it to fire at the Fan Song or Low Blow. EDIT 2: The AI will only fire AGM-45A if the SEAD advanced waypoint task is active, it doesn't work if you only have an Attack Group/Unit or Search then Engage task set. The GBU-8 is as with the Paveway II and III described above, see F-4E_AI_GBU-8_nodrop.trk. Both Walleyes however are unaffected. F-4E_AI_LGB_bug.trk AI_F-4E_AGM-45A_nofire.trk AI_F-4E_AGM-65D_nofire.trk AI_F-4E_GBU-8_nodrop.trk
-
Anyone else experiencing very weird pathfinding decision by ground units when set to "On Road" and moving on the Autobahn? Specifically this is happening in my mission along the A4 and A64 near Bad Hersfeld, Alsfeld and Fulda.
-
We desperately need an option or variation of the CAP task, to limit units to engage only when enemy units are within a set range relative to the CAP flights current position. Contrary to popular belief this does not exist in DCS currently. The option available only limits engagement to a distance around the flight route, which is a feature that is utterly pointless, since "engage in zone" does more or less the same.
-
Hi everyone, The recent update has introduced some unexpected behaviour for the AI when performing a racetrack orbit. Unlike previously, where upon activation of the task the AI would fly directly to the next waypoint before making its 180° left turn, now, upon activation of the task, the AI immediately performs a 180° turn to the right, immediately followed by a 180° turn to the left. They will then fly to the next waypoint, make a correcting turn to the right and then commence the 180° left turn to begin the racetrack. Fortunately, subsequent "laps" of the racetrack are perfectly excuted - completely as expected and the AI no longer needs to make corrections upon finishing a turn, which is fantastic to see. However, the initial behaviour upon activation is undesirable, especially for practice missions with tankers and receivers that start from the air. There’s no reason for the initial right, followed by left turn, all it does is displace the aircraft laterally from the intended course line and it's only something the AI later corrects when reaching the first turn. Instead, what should happen is that the AI should proceed to the next waypoint as it did before, perform its 180° turn, fly the reciprocal course until it reaches the start point, do another 180° turn and so on, as it does for subsequent laps. Orbit_2.9.12.5336.trk