Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Alex]

I think the AI could do with more feedback to the player.

...

A fair point! Not the easiest thing in the world to do well, though.

We were under the impression that one of the stronger factors regarding the player's current flagship is that allied AI will strongly prioritise targeting the enemy ship the player is actively shooting. 

Is this actually a factor?

I thought it was both one of the stronger factors in the allied AI and also one of the aspects more extensively tested by the community.

The AI will for some purposes look at what nearby allied ships are targeting, but the player ship is not "special" in this regard. If anything, it's harder for the AI to figure out what the player is targeting, so it's less likely to have this info and act on it in the case of the player ship.

Also: where this does kick in generally does *not* influence general target selection by the AI. Unless ordered to "Eliminate", it will pick targets based almost purely on positional factors. (Basically, "which ship is best to pivot around so that doing so will put all the enemies on the opposite side of that ship so I don't get flanked" is the general target selection algorithm.)

[TheWetFish]

Ah-ha!  Thank you very much for clarifying that

I understand effective feedback of this nature can be quite difficult, particularly if it's not already built for it.  From community experience this is one of the largest friction points about learning the game; learning how to predict how the AI will react in the absence of effective feedback.  Please do let us know if there is any way the community can assist with this

[Thaago]

I think the AI could do with more feedback to the player.

A lot of confusion arises due to insufficient feedback as to why the AI is doing something.  For example if the AI ship provided some form of feedback to the effect of "I want to attack my target but I am low on flux and scared of being flanked".  This would both improve player expectations and improve player experiences with the AI.

...

I think some of the Combat Chatter dialog is keyed off of the AI ships having certain AI flags happen? I'm not entirely sure. If not, thats something that could be done rather easily I think to give better player feedback.

[TheWetFish]

I think some of the Combat Chatter dialog is keyed off of the AI ships having certain AI flags happen? I'm not entirely sure. If not, thats something that could be done rather easily I think to give better player feedback.

Correct, there are some exposed AI flags which I believe are indeed used in Combat Chatter or in other mods such as Blothorn's AI Flag Tool which does get a decent amount of use in testing

More flags would potentially enable the community to experiment with various ways of presenting the data, which may potentially eventually inform vanilla solutions

[Alex]

Please do let us know if there is any way the community can assist with this

Thank you!

I think some of the Combat Chatter dialog is keyed off of the AI ships having certain AI flags happen? I'm not entirely sure. If not, thats something that could be done rather easily I think to give better player feedback.

Correct, there are some exposed AI flags which I believe are indeed used in Combat Chatter or in other mods such as Blothorn's AI Flag Tool which does get a decent amount of use in testing

More flags would potentially enable the community to experiment with various ways of presenting the data, which may potentially eventually inform vanilla solutions

Right, that's workable, but those flags really aren't meant to be player-facing, so it can get pretty iffy. That is, these flags aren't so much indicators of intended courses of action or long-term intent, but rather ways for different modules of the AI to communicate in the moment. For example whether a "back off" flag would be unset in the next few frames or not, you can't really tell from looking at the flags (yes, flags have a "duration", but they can be unset before that expires).

And further, the AI doesn't generally operate off of long-term plans, but rather off of the current conditions. So "I'm doing X" is just fundamentally "I'm doing X this instant, but may or may not keep doing it" so it's... complicated. I think the only way/reason "barks" like this can work is due to the human propensity to falsely assign meaning to the AI's actions if the AI "tells" you what it's doing, even if it's not actually doing that. Which, fair enough, it's a reasonable trick to pull out... but then that limits the usefulness of this as a learning aid. Still, may be worth messing with at some point, since it would be quite good if pulled off reasonably well.

[Thaago]

The most important Combat Chatter ones for me are for ships calling out in distress when the are under heavy fire/high flux/taking hull damage because that gives me an alert that I should check out the situation on the tactical screen. Those seem fairly reasonable as they are more acting as status reports than plans.

[Alex]

Ah - yeah, that makes a lot of sense!

[TheWetFish]

Great point on status vs plans

If we want to narrow it down then major friction points sound like they occur when the player issues a specific order and the AI appears to not follow it (such as the OP of this thread).  If we can get some soft detection on when that is happening and any status feedback then I think that'd go a long way

For example not moving towards an eliminate target somewhat quickly or not being able to fire on designated target for some amount of time then trigger status information of some form

[sector_terror]

I am in no way a developer but it does seem that some builds causes the AI to be more aggressive than other builds. No matter what you do there is always 1 ship that always goes suicide.  It almost seems like some weapons on ships make them go in close. Like if you have a tac lazer the AI will stay at a distance but if you mix them they will not go in close even if you have weapons that are all close except 1 tac lazer.

I can confirm this in my own experience as well. The AI handles carriers differently than an enforcer which plays differently than a wolf. This does mean figuring out what the game really plays off of is pretty critical. If you really want to cap the AIs operation, youd have to really dig into how it measures its design, assuming I'm correct(and I think I am.) Ex: The AI will stay fairly passive on an eagle equipped with 2 assault chainguns and a hypervelocity driver, if you put 3 graviton beams on it. But swap the beams for pulse lasers or especially ion pulsars and it changes the combat range to be much closer. Again, just my experience so I might be imagining it, but my point remains.

Honestly, this is making me appreciate the AI. It's acting "intellgently" enough that I can't figure out what it's doing, but never feel it's making a truly stupid move. I'm getting mental uncanny valley vibes and thats truely an impressive feat to have created.

[TheWetFish]

Starsector AI is generally best in class for this sort of game.  The downside to this sort of AI is that it takes a very long time to fine tune it - which is just fine for Starsector's very long development time

Check out the AI Behaviour wiki page if you want to know more about the dynamic weapon range behaviour, in particular how officer personality interacts with it

[Tartiflette]

What we need in 0.95 is the possibility to assign an AI core as an officer that will mindlessly follow the players orders! And then we can watch those complains about AI multiply after those same player got their ships killed because they mindlessly followed their orders.

[Grievous69]

I was just about to comment that. Not trying to be dismissive but people really don't realise AI tries its best to do these things in order:

• Try not to die
• Stayin alive
• Stayin alive
• Follow some orders

If AI behaved just like RTS units where you simply click, and the ship just follows that order (without falling back or doing anything else unless you directly command it), you would have the most frustrating experience this game would ever offer to you. I know AI is not the best thing in the world and has its flaws buy it does its job damn well.

[IonDragonX]

I re-read Alex's patch notes. There's two AI changes that he lists that do affect the combat AI. He mentions that the AI has new code to "do what the player wants ", whatever that may mean. He also mentions that the subroutine that tries to get superior positioning (or just not get flanked) kicks in earlier.

[intrinsic_parity]

One kind of off topic thing I've noticed is that the AI sometimes gets into trouble because it seemingly doesn't know that the enemies can deploy new ships from the top of the map. Especially for fast ships, they can sometimes go on wild flanks and get surrounded by new reinforcements. I'm not actually sure if the AI addresses this at all? If not, it would make sense (to me) if there was a fake enemy presence in the deployment at all times that the AI considered in its positioning.

[Megas]

One kind of off topic thing I've noticed is that the AI sometimes gets into trouble because it seemingly doesn't know that the enemies can deploy new ships from the top of the map. Especially for fast ships, they can sometimes go on wild flanks and get surrounded by new reinforcements. I'm not actually sure if the AI addresses this at all? If not, it would make sense (to me) if there was a fake enemy presence in the deployment at all times that the AI considered in its positioning.

This is another reason why I make my fleet wall-camp (by having them escort my ship while my ship camps at the edge - thanks no-edge ralllies) during the cusp of endgame, when my fleet is not quite strong enough to mow down endgame fleets.  Not only to quickly retreat ships that run out of PPT, but also to trap the cowardly enemy AI ships in an ambush and kill them before they get away.