Very true. Also, is the plan to improve AI over the original X-wing? I apologize if this was already covered earlier in the thread, but from what I scanned I didn't see mentions of specific AI types.

Actually NM, I found the Google doc with the AI info in it. Sorry for the inconvenience!

No, it's a good point.
In fact, if you have a peek at past 5 or 6 pages, I was commenting there that the AI I first wrote was significantly harder than the one in the original game. To the point that there were basically no dogfights happening. Both rivals would start aiming and hitting each other from weapon max distance and the combat would last less than 2 seconds.

Of course, that "broke" the test mission. The ships were not acting, or at least lasting, as they did in the original. And if that isn't working properly, then the mission balance changes dramatically, making it too hard, too easy, or directly impossible to complete.

So I had to limit the AI to the easier levels of the original game. But that doesn't mean that we could add some kind of Nightmare mode where the AI is harder. Or have full new missions and campaigns that make use of the harder AI ranks.

Interesting. It is difficult to get AI to act like human opponents, that's for sure. In my opinion good AI with human- like traits make the game far more enjoyable and immersive. AI in many games is either too easy or simply on murder bot mode, and it doesn't respond the way a player would, which kind of sucks the fun out of tactical dog fighting.

Just throwing this out there, but an interesting goal might be to give the AI dynamic confidence. So for example, if the AI wing-men are close at hand it is more aggressive, but if the wing-men are being destroyed it has has a probability that it wants to flee or at least go full evasive, rather then aggressive. Similarly, and you might already be doing this, it would be interesting to have the AI lose coordination and confidence if the leader of the squad was killed.

A stretch might be to make AI fight dogfights like actual players, ie using maneuvers involving more then one fighter. For example, it would be very unique if the AI could use a pincer maneuver, or use different styles of attack plans. As an example, one AI might fly evasively to draw a line of fire while the other falls in behind to pick off targets. Even more difficult would be to create a variable list that AI uses to monitor the battle situation then choose a pre-determined attack plan based off of the variable values. For example, the AI reads the variables that there are 10 enemy fighters to it's 4 fighters so it chooses a team strategy based on retreating. These kind of behaviors are very difficult to code and may be simply over the line of what we want to do with the project. From experience, I have a hard time enough getting the AI to do similarly clever things with dynamic way-points in my team's Free-space Open Star Wars mod. Oh dear I think I'm ranting now...lol

All valid suggestions, but again, that seems to be more appropiate for a pure dogfight game, where:
- Dogfighting is all what there is to do. No other goals to track, transports to protect, cargo to inspect, ships to disable, etc.
- The player has some kind of self healing and respawning ability.

In X-Wing versus TIE fighter (or in Battlefront 2), most of the time it was like that in all those melee engagements where players would respawn indefinitely and didn't matter at all what other ships were doing, the player could focus on keeping himself alive.

In X-Wing and TIE Fighter, and the rest of the campaigns in the series, the player doesn't respawn. The whole mission ends if the player is downed once. Also, there is no selfhealing of any kind.
The AI in X-Wing and TIE Fighter is purposedly limited so that the player has a chance to complete the missions.
The player learns enough about keeping himself alive during the Historical Missions training. Then the difficulty goes up again during the Tours of Duty, since they are most of the time about keeping others alive.
That is much harder if the AI isn't under certain limits.

For example, in some mission, a group of TIE Bombers are dispatched to destroy a medical frigate with torpedoes. The player must rush in his X-Wing to intercept them before they are able to release them.
If the AI were actually clever, the TIE Bombers would release a barrage of missiles against the incoming X-Wing, then bomb the frigate at pleasure.
Actually, if the AI was clever, it wouldn't release the squadron of TIE Bombers piecemeal, that is, in waves of 3 ships. Instead, it would release the 12 of them at the same time.

And so on and on.
For good or for ill, X-Wing isn't a pure dogfight arcade game, but more like a puzzle-mission-solving spacesim. Each mission is like a riddle to solve, with all flight groups acting in a particular choreography, and the player needs to figure out which missing piece he has to impersonate so that the mission ends in success.
Look at it like a Rube-Goldberg machine where some gadgets need to be interacted with at the right moments for it to work. That is X-Wing and TIE Fighter.

Of course, we could have a Skirmish mode for those just wishing for instant action. That should be super easy to implement, and then there we could release the beast and give the AI all kind of tricks, feints, maneuvers, and abusing practices.

I see your point. In my team's FSO Star Wars mod when the player dies the mission is over and the player also has protect and defend/escort duties, almost always without self healing. With that being said it is more of a military-movie sim rather then a puzzle sim like the original x-wing. I agree that being a remake that your description fits better :)

On a side note your 'release the beast' analogy made my day!

I'm not dead, again. But almost.
If there has been little progress reported here, the reason is that at the beginning of august I took a 3 weeks vacation trip, then, as soon as I came back, I have been sick with a nasty cold that doesn't want to go.

Anyway, I have been working on the game in the spare time between my daily job and my evening fever. I integrated some dashboard graphics for the missile lock and the shield instrument, kindly provided by FekLeyrgTarg.
But mainly I have been dealing with a facet of the AI that was not yet developed: reactions.
I think I have explained this in the past, but I cannot remember for sure, so there I go again.

Every AI ship follows a flight plan that carries out their assigned orders.
Sometimes something happens that requires immediate attention by the AI ship and a momentarily change of plan while the exceptional situation is occurring.

Instead of adding all possible handling for all possible exceptional situations at every step of every flight plan for every order in the game (that would involve a lot of redundant logic, and when you are adding redundant logic you know you are doing something wrong); I have opted for creating the concept of Event->Reaction. When a particular Event happens, the ship affected jumps from whatever they are doing to the Reaction indicated.
A Reaction basically is a simple task to perform.

As an example, I will describe the first reaction I have added as a testbed for this subsystem: Evasive Maneuvers.
As I described in the past, one of the shortcoming that I was observing with the current AI was that they were so bloodthirsty that they would basically limit themselves to joust other ships and exchange laser fire until one of them would blow up.
Believe it or not, that made them extremely lethal, even if suicidal. In particular shielded ships became really hard to face by unshielded ships.
In the test mission, there are 3 Y-Wings against several waves of TIE Fighters. In a normal game of X-Wing, that would be kind of balanced.
In the state of XWVM at that time, however, that was a total defeat for the TIE Fighters, because the Y-Wings could just joust them and survive the incoming fire, while the TIE fighters blew up practically one-shot (When the AI has a good shot at their targeted ship, they switch to dual fire, that is enough damage to oneshot a TIE Fighter).

It didn't matter that 3 TIE Fighters were shooting at the Y-Wing. The shields would protect it while it repeatedly slaughtered the whole wave of TIEs during approach.

However that is not what happens at all in the original X-Wing game. If you observe the game, as soon as a ship has received enough damage, it breaks off and starts performing evasive maneuvers.
Not only this increases the survivability of the damaged ship. It also interrupts the combat, giving a chance to those poor unshielded TIE Fighters to last longer.
And in general, it makes all dogfights last longer, especially those where the player isn't involved, so it serves the game too for maintaining those battles in the background that are there to give some missions some thematic feel, while the player is doing something else.

So for creating this Evasive Maneuvers reaction, I have been exploring several ways for the AI to decide when to break off and start doing barrel rolls and follow a twisty path to avoid being hit.
Perhaps you people can provide a better way, but what I have done so far is adding the concept of "pain buffer" and "pain threshold" to AI ships. Whenever a ship is hit, it adds the damage it just got to the pain buffer. When the pain buffer goes above the pain threshold, then the In Pain event triggers, and that leads to the Evasive Maneuvers reaction.
The pain buffer is slowly emptied, so that with time, the ship no longer is "in pain".

This is done and working. Now, some orders specifically instruct the AI not to perform evasive maneuvers, so this reaction will be disabled for them.
At this moment, the Y-Wings and TIEs exchange fire for a couple of seconds, then the Y-Wings break off after sustaining enough damage, and the TIEs chase them, and it becomes a more interesting dogfight to look at, and definitely closer to what it used to look like in the original.

The evasive maneuvers themselves, I have implemented them by making the ship pick a random vector and aiming at it, once every 2.5 seconds, as long as they are "in pain" or for 5 seconds, whatever is longer.

Other applications of these reactions could be, for example, the "wingmen escort" behavior we can observe in some ships: A flightgroup of 2 ship or more has assigned an order to attack a particular target. The entire wave flies towards the target. If the wave is attacked, the flight leader stays on target, while the wingmen attack the attacker.
This is really effective to make the player lose focus on the mission objectives. The wingmen distract the player in a dogfight, while the leader sneaks past and destroys the taget, that the player usually needs to protect.

Many of the advanced AI tactics proposed by Wookiejedi could also be programmed as reactions.
A pincer maneuver could be implemented like this:
Event: Target flightgroup is targetting us, and is flying in close formation.
Reaction: Divide wingmen in two groups, assign group A a waypoint to the left of the target group, and group B a waypoint to the right of the target group, then focus fire.

The Bait tactic would be basically a variant of the Wingmen Escort. The attacked ship starts performing evasive maneuvers nonstop, while the other ships in the wave target the bait's attacker.

A nasty Merry Goose Chase tactic could be like:
Event: A ship is targetted by many enemies.
Reaction: The ship focuses on fleeing the battlefield at max speed, while all other ships in the wave attack the ships attacker.

That one would be really annoying and unfun for the player, I suspect. Imagine that you are facing a wave of 3 TIE Advanced. You target one, and that one starts fleeing, soon getting out of your weapons range, meanwhile the other TIE Advanced target and attack you while you are an easy target for them because you are desperately trying to aim at a fleeing ship. As soon as you change your target, it all repeats.
It would be a royal pain to play against a squadron that behaves like that.

I like those ideas for AI. Another thought - there's a number of different variables that could be added to Pilot AI skills to make them more diverse. For instance:

Bravery: bonus to pain buffer
Gunnery: bonus to weapon accuracy
Maneuverability: bonus to evasion time

Then instead of just making the pilot get better at everything from Rookie to Top Ace, they're given a pool of points that could be distributed among the three main skills.

How about also adding a "Teamwork" skill that unlocks additional squad maneuvers (some of those advanced AI ideas above) with each rank?

I would worry that the Merry Goose tactic could very easily lead to uncompletable missions, just due to the simple fact that the enemy craft are often faster - particularly bad for Y-Wing missions, where just trying to catch up to gunboats that are 3 clicks away and going in the opposite direction is a difficult experience.

Naturally, I would have no objection to this type of behaviour as part of an 'enhanced AI' option, and agree that it would make pure dogfights more interesting!

The "event->reaction" system, to my mind, seems to mirror the vanilla behaviour - the classic example being ordering a friendly craft to perform an evasive maneuver, where it puts the current order on hold, evades, and goes back to what it was doing before.

Speaking of evading, it's worth mentioning the observed behaviour in the original game where extremely badly damaged craft will attempt to fly home - whether this is entirely realistic is up for debate (you could argue that some pilots would keep going; or kamikaze; or flee and regenerate shields in that situation), although as a long-time player of the original games, it's a behaviour I now automatically anticipate when playing. Would the proposed "pain meter" take this scenario into account?

I don't think the Merry Goose Chase tactic should be enabled in "classic" X-Wing mode.
All advanced AI tactics and exploits should be there only for Skirmish mode (just procedurally generated mission with ships and objectives thrown together, like the majority of missions in the Melee collection in XvT), or perhaps for other hypothetical modes that I will address in another post. Countbuggula's comments about particular "Traits" that improve or worsen some aspect of a pilot would also belong more to some kind of new RPG Campaign mode than to the Classic X-Wing TOD mode, where you get to improve your squadron of pilots to face harder and harder scenarios.
Remember the Rube-Goldberg nature of the classic missions. Any substantial change in the AI would break them.

Now that you mention it, the "Retreat to repair" behavior could indeed be an Event-Reaction. The triggering event would be having a damaged hull (under 50%?), plus a certain "die roll". If the die rolls affirmatively, then the ship decides to change his orders to Fly Home.
I have seen this behavior only on shieldless TIEs. Have any of you seen it on rebel fighters or other ships?

Because I plan on streamlining modding and adding a mission builder inside the game, so that we don't need to resort to 16-bit programs to make custom missions, I thought that once the Classic X-Wing mode is completed, we could put some time on creating a Squadron Campaign mode, where the player leads a squadron of rebel fighters thru a open-ended campaign mode. In that mode, you could improve or alter the traits of your pilots, and the equipment of your ships, as you rebel cell gains prestige among the Alliance or you make deals with sympathizing civilians. But this belongs to another post, at another time.

Not necessarily - remember that (at least in the DOS version) you could already assign locally saved pilot AI profiles to all members of your flight group, giving them potentially anywhere from Rookie to Top Ace ability. I don't see this as being all that different, just giving a bit more variety or flexibility within that range. It just means that you could potentially have a veteran AI with Top Ace accuracy but has no guts and wimps out at the first sight of danger. Or someone with all guts but no real skill. Those ranges would still be within or close to the normal scaling from Rookie to Top Ace, at least as far as individual attributes go.

That actually brings up a good point though - do we know exactly how the skill ranking system worked for the different AI levels in the original game? I remember from my own experience that assigning Top Ace pilots definitely made a noticeable difference, just not sure exactly how. That might help point us in the right direction here.

I think the hull damage threshold is <25% remaining - but I think this is based on some triggers available in ALLIED. It does seem to fit with the pattern of behaviour I've seen while playing, however.

I think I might have seen this behaviour on friendly craft also, however I would need to experiment specifically to be sure. While not 100% relevant, I can definitely confirm that it happens to Rebel craft in TIE Fighter (you can see this behaviour all the time in small strike craft groups such as Y-Wings and A-Wings).

Yes please! :D (for both the integrated mission builder and the new Campaign mode - I will happily pledge any assistance I can, including building a full-length campaign [and even better if we are able to include extra stuff like platforms])


I can't say 100%, but from a mission design perspective (using X-Ed), flight groups can be assigned an AI level from Rookie to Top Ace, but it doesn't seem to be any more fine-grained than that, at least from the known mission specifications. I would imagine that in the original game, selecting a pilot of a given skill level just creates the selected craft with the equivalent AI level when the mission is initialised.

As to what changes exactly with each skill level from an AI behaviours perspective, I'm afraid I don't know, other than more accurate targetting (leading the target) for the top level(s?) as identifed earlier in this thread.

It's worth a try, the thing about assigning (or developing themselves) some traits, even in the Classic mode.

About the different AI ranks, I am currently just making it determine the "thinking frequency", that is higher the higher rank the AI has.
If that doesn't make sense for you, just let me say that AI ships think at particular moments. A Top Ace thinks about 5 times per second, while a Rookie thinks once per second.
Considering the speeds these ships move at, you can imagine the "reflexes" displayed by the Rookies will be clumsier than those of the Top Aces. Aiming depends on thinking, also, so higher ranks correct their aiming more frequently than lower ranks, increasing their chances to actually hit.

This is an implementation totally based on my intuition on how the original game works.
It would be wonderful if someone could actually find out more differences and report them. :)

How about the ability to assign points to your own traits every time you advance, and trait points are assigned semi-randomly (to keep them from being too unbalanced) for NPC pilots - unless you assign another pilot record that you've already customized. Keeps it from being too complicated (like having to manage leveling up of an entire squad - you're correct in thinking that sort of thing is best left to a future custom mod or campaign) while still retaining just slightly more than the functionality from the DOS X-Wing. That makes sense. In fact, "Reaction" or "Reflexes" could potentially be another attribute/trait.

Another question: do we know how many points/experience is required to level up each rank? And how many points each ship destroyed/hit/disabled counts for? That's all important information to properly implement the ranking system.

Somewhere I found a list of points for ships, primary objectives, and such for TIE Fighter, but not for X-Wing. I guess only experimentation will allow us to find the right numbers. It's not such a high priority task, but at some point we will need to find out.

I know that every laser shot substracts some points, and every missile/torpedo shot substracts a lot more. Mines, satelites and other minor objects, I think, don't give points unless you destroy them all. Or something like that.

I am currently developing both the boarding mechanics, as well as devising a way to automatize importing new ship models in a less work-intensive way.
You just need to see MajorParts' new models for the TIE Fighter and the Lambda Shuttle!