In another thread, there was some conversation about small kinetics, and the case of a medusa vs a hammerhead was used, so I was doing a little testing myself, and ended up encountering some AI behaviors that I find very frustrating and think could be improved.
The main issue is about committing to dissipating/venting.The biggest issue I encountered here with the AI is that it often decided it is at a flux disadvantage and tries to back off, but never really commits to fully venting, and ends up re-engagning at elevated flux. I think it's probably related to the significant range disadvantage as well (this is much less prevalent if the medusa has ITU), but the AI should be able to play effectively at a range disadvantage, and this happens even when then ship in question can win very easily under AI control with not other changes, if it doesn't do this behavior.
I recorded a video highlighting the issue. In the beginning, it's clear that the medusa has the advantage when slugging it out, but backs off when on a moderate amount of flux, which is a bit annoying but not totally unreasonable. However, it never actually vents, and repeatedly re-engages on ~30% flux, not achieving anything, before eventually (~1:15) ending up in a kiting cycle, where the hammerheads 0-flux boost is enough to keep it nicely in range. I let that cycle run for a while to show that the medusa will not get out of it (I've tested this 20+ times and this constantly happens). At about 2:30 I take manual control, and skim back to vent, then give the AI back control, after which it wins very easily, so the ship is totally capable of killing under AI control without any human intervention (it's not an issue of having to take some hull damage/risk to win), but these engagement issues prevent it from doing that.
I think having the AI commit to fully backing off and venting when it knows it is at a flux disadvantage, and more importantly,
not re-engage at elevated flux would completely eliminate this issues. Particularly for ships with unrestricted mobility systems, this should really not be a problem. Obviously sometimes when you are at a speed disadvantage, you can't do anything about kiting, but I see this in ships with speed advantages as well (scarabs a lot), where they keep re-engaging at elevated flux, or hover around inside the enemies weapon range but outside their own. I think the AI just needs to be more decisive about engaging/dis-engaging. If it's too dangerous//risky to fight at the current flux levels, then make sure you get out of enemy weapon range and vent. If you want to fight, then get into range ASAP.
I actually noticed after recording this that I forgot to remove my player officer for this video (I tested it yesterday without an officer and saw the same issues). The only skill I had was ordinance expertise (and no fleet-wide skills, I'm early in a campaign), so not a big change. When I removed the officer, the ship won easily with no issues which was confusing, until i realized I had set the fleet behavior to agressive after testing yesterday. When I changed it to steady, all these issues re-emerged, so I think it's particularly an issue with less agressive AI (which also reinforces my general preference for agressive AI). IMO, the steady AI should be decisive about the actual movements/mechanics of engaging/dis-engaging, it should just be more reserved about the decisions to engage. In other words, less likely to 'go for it', and more likely to back off when things are dicey, but still
actually going for it, or fully backing off when it decides to do those things.
In addition, I noticed a few other smaller things that I don't feel like making separate threads for:
Spoiler
-I noticed that it helps a lot if I give the medusa some momentum towards the enemy before handing back control, as it tends to result in the medusa ending up in phase lance/ion pulser range immediately. If I don't do that, it sometimes gets back into a kiting cycle, or at least takes a lot longer to kill. It would be cool if the AI would prioritize and use its mobility to get into shorter weapon ranges. It seems very willing to skim once into LDAC range, but never skims into ion pulser/phase lance range. Maybe this is an aggression level thing though. If the ship performance is so sensitive to AI aggression level though, then we really need some way to set it on a ship by ship basis.
- I noticed sometimes the AI fires the phase lance when it is in range of shields but not hull so that when the enemy dropping shields, the lance is wasted. I'm pretty sure this is related to auto-fire behavior. It would be nice if the auto fire would wait to get into hull range to fire. I think that behavior would be better overall, even if it was a minor disadvantage for kiting occasionally.
- I noticed there seems to be some bug/issue with the AI and finishing things off when they are overloaded. It's highlighted at the end of this video, but I've seen it many times in other contexts, the AI just doesn't fire stuff when ships are overloaded. You can see in the video when the hammerhead overloads, it backs off and plinks away with kinetics when it could easily kill with a phase lance or ion pulser. I've watched ships refuse to fire anti matter blasters or other strike weapons at overloaded ships in the sim as well, and at this point I'm almost convinced there's a proper bug somewhere.
- I think 'counter venting' is something the AI should do as well i.e. vent when your opponent vents to reset flux. It's not really highlighted in this video, but I frequently saw the medusa just sit out of range, or maybe slowly move forward/sit still with shields up while the hammerhead vented instead of venting for itself as well. I think sometimes, it was maybe reacting to missiles and backing off, but even then, it never vents after the missiles are gone and does the re-engage on elevated flux thing. It really feels to me like the AI should make sure it resets flux when the opponent vents unless it is trying to kill, otherwise it just ends up at a big flux disadvantage when the enemy finishes venting, which often leads to the noncommittal backing off type behavior I've been seeing as well.