Oolite Bulletins

kanthoney wrote:

Some of the combat AI design decisions do seem a bit odd. For example I had a go at the combat simulator recently and picked a harmless pilot in a Mk III Cobra, and got a bit disconcerted when he started tearing me to shreds with his side lasers.

I don't think I've ever met a ship with side lasers in the wild.

Nor I - and if the core ships were to start doing that, I'd have to use my own - mostly unused - side lasers!

Heh. I'd use side lasers more often if I could operate them independently of ship maneuvering like turrets As it is, I think they're only useful for mining lasers to free up the front and rear for combat.

mossfoot wrote:

I think they're only useful for mining lasers to free up the front and rear for combat.

For some OXP situations, four mil lasers are more or less essential (if memory serves).

Cody wrote:

mossfoot wrote:

I think they're only useful for mining lasers to free up the front and rear for combat.

For some OXP situations, four mil lasers are more or less essential (if memory serves).

Thargoid Battleships in particular comes to mind. Go stationary about 10-25 km from it, overheat ALL 4 Military Lasers on one...and barely kill it. Run like hell, as the rest of the Thargoids may have beat your shields down by this point...

The combat simulator ranks are approximate - the AI shoots better than a Harmless pilot at that level, but dodges far worse. (Also, I set the approximations in 1.77, so I may need to relabel some of them a little)

Hi, here I am again. I did remove some opxs from my installation, hoping the game will get easier after this.
But I was disappointed. It's still the same frustrating game experience. Two things are going wrong with 1.80 IMHO:
1. The attackers are much to numerous.
2. The AI is too aggressive. They laser precision is too good and their reactions are too fast. As I mentioned before missiles are useless now, either they got an ECM or they shoot the missiles down after a few seconds.

Sadly, I have to tell you that you lost a fan with this version.
Is there still a version 1.77 available for download?

Maybe 1.82 will do better....

aen32 wrote:

Is there still a version 1.77 available for download?

Version 1.77.1 can be downloaded from either of these pages:
https://github.com/OoliteProject/oolite ... ag/1.77.1a
http://sourceforge.net/projects/oolite. ... %201.77.1/.

Have we got anything planned combat AI-wise for 1.82?

My suggestions would be:

1) Split the attack behaviour into two separate behaviours - dogfighting, and sniping. I think the problem average players like me and the OP are having is that while some NPC's are much more accurate they haven't been slowed down at all to compensate, so they can quickly lock onto your position and take you down with highly accurate fire. At the same time, they're not accurate enough over long distances - if you've got a rear military laser you can just run away and pick them off one by one, and they've got no comeback to that (so you can win, but it's a bit dull fighting like that). So I'd have a dogfighting behaviour in which the NPC would quickly acquire target but the accuracy was dialled back to 1.77 levels, and a sniping behaviour which was much more accurate but in which the NPC was slowed down to mimic careful lining up on the target. Of course you could have some ships that could snipe quickly, but they'd typically be highly skilled assassins and mercenaries and not pirates.

2) Currently, the skill level of an NPC seems to boil down to just one number, the accuracy. It might be an idea to break that out into several different skills and apply skill levels to each independently, which would produce a more rounded set of AI's. So for example you could set DogfightAccuracy, SniperAccuracy, SniperSpeed and EvasionSkill all set independently. You could even have different skill levels for each gun - an NPC may be good at both dogfighting and sniping with the front gun, but may only be able to snipe effectively with the rear one.

3) I did have an idea a while back to slow down the NPC's reaction times, which I'll have to resurrect. Currently the NPC knows exactly where the target is and can therefore adjust instantly to any course changes. My idea was to record the positions of the target a few times a second, then fit a curve to the last three positions and use that to work out where the target was - the idea being that if the target flies in a predictable way the NPC should be able to hit it, but if the target jinks then the NPC should take a couple of "ticks" to fully adjust to the change. And, of course, the amount of time between ticks could be adjustable, giving us another skill parameter to play with.

kanthoney wrote:

Have we got anything planned combat AI-wise for 1.82?

Nothing yet.

Depending on how the equipment discussion goes, two things which might make 1.82 combat easier are
- reduced power beam lasers (or if that doesn't work out, just significantly reduce the number of ships with beam lasers at all)
- NPC equipment damage causing them to flee more often
But those need more discussion.

Taking missile fire out of its current control and putting into the JS AI might be a useful way of allowing new players to scare off a Gecko or a Cobra I without - if they then stop firing! - it feeling the need to use its missile on them.

I am also thinking of increasing the "base" threatAssessment of a ship from 1 to 2, to make it more likely that the small pirate groups will attack the player giving them a fight that they can win in the safe-ish systems, rather than anything the player might possibly stand a chance against staying well away: that's too much caution for pirates to get away with, at least in the disorganised groups.

kanthoney wrote:

3) I did have an idea a while back to slow down the NPC's reaction times, which I'll have to resurrect. Currently the NPC knows exactly where the target is and can therefore adjust instantly to any course changes. My idea was to record the positions of the target a few times a second, then fit a curve to the last three positions and use that to work out where the target was - the idea being that if the target flies in a predictable way the NPC should be able to hit it, but if the target jinks then the NPC should take a couple of "ticks" to fully adjust to the change. And, of course, the amount of time between ticks could be adjustable, giving us another skill parameter to play with.

I like this idea a lot. They need to be as accurate as they are to stand much of a chance of hitting a ship flying straight and level at anything beyond point-blank range, but unfortunately the very small course corrections they need to follow an evading target at ~10km are then also fairly easy for them (rolling so you're dodging away from rather than along their pitch axis helps a bit)

This would make things a lot more fun, I think.

cim wrote:

kanthoney wrote:

3) I did have an idea a while back to slow down the NPC's reaction times, which I'll have to resurrect. Currently the NPC knows exactly where the target is and can therefore adjust instantly to any course changes. My idea was to record the positions of the target a few times a second, then fit a curve to the last three positions and use that to work out where the target was - the idea being that if the target flies in a predictable way the NPC should be able to hit it, but if the target jinks then the NPC should take a couple of "ticks" to fully adjust to the change. And, of course, the amount of time between ticks could be adjustable, giving us another skill parameter to play with.

I like this idea a lot. They need to be as accurate as they are to stand much of a chance of hitting a ship flying straight and level at anything beyond point-blank range, but unfortunately the very small course corrections they need to follow an evading target at ~10km are then also fairly easy for them (rolling so you're dodging away from rather than along their pitch axis helps a bit)

This would make things a lot more fun, I think.

I like this idea too. Something to simulate human fallibility and adjustment time during close range combat. Otherwise it just feels like they always have a lock and you're just relying on their lack of accuracy to keep you alive. Granted at a distance it's much easier to keep a lock.

3) I did have an idea a while back to slow down the NPC's reaction times, which I'll have to resurrect. Currently the NPC knows exactly where the target is and can therefore adjust instantly to any course changes. My idea was to record the positions of the target a few times a second, then fit a curve to the last three positions and use that to work out where the target was - the idea being that if the target flies in a predictable way the NPC should be able to hit it, but if the target jinks then the NPC should take a couple of "ticks" to fully adjust to the change. And, of course, the amount of time between ticks could be adjustable, giving us another skill parameter to play with.

That's a very good idea!

I've written this now - I'm not sure it's making a huge difference, but then we don't want the game to be too easy, do we?

There was something about the way the vsync branch was set up that made me think I might be violating some etiquette by just pushing branches out to the repo like I have been doing, so I want to confirm it's OK before I push this one out?

kanthoney wrote:

There was something about the way the vsync branch was set up that made me think I might be violating some etiquette by just pushing branches out to the repo like I have been doing, so I want to confirm it's OK before I push this one out?

Go ahead - it's a lot easier to test if there's a branch for it.

OK, thanks. I didn't think it was a problem but just wanted to check. It's on the npc_reaction_times branch if you want a play with it.