AIs don't seem to know about space elevators

[Tualha]

I've been experimenting with some very sparse layouts on 0.4.4. Over 700 systems, but only 40-odd with stars or planets, and all the planets are huge, hence bad supply range. Low starlanes, too. You pretty much have to build space elevators before you can get anywhere.

The AIs can't figure this out. They never build space elevators, so they never prosper. They colonize whatever's close enough, and then they stagnate for hundreds of turns.

I think they should gain the ability to figure out that they desperately need more supply range, calculate how much extra they'll get from a space elevator on any given planet, and do the necessary research and building to solve the problem.

[Vezzra]

I've been experimenting with some very sparse layouts on 0.4.4. Over 700 systems, but only 40-odd with stars or planets, and all the planets are huge

How did you manage that? To get that, you have to tweak the universe generation scripts - did you do that?

The AIs can't figure this out. They never build space elevators, so they never prosper. They colonize whatever's close enough, and then they stagnate for hundreds of turns.

I think they should gain the ability to figure out that they desperately need more supply range, calculate how much extra they'll get from a space elevator on any given planet, and do the necessary research and building to solve the problem.

Although the AI is already quite capable and will put up a somewhat decent fight, it's still very much WIP. There are a lot of things the AI can't do or doesn't get right, our main AI dev (Dilvish) still has long road ahead. We're still in Alpha, after all

[Tualha]

How did you manage that? To get that, you have to tweak the universe generation scripts - did you do that?

Yes, of course. I've tried all kinds of variations. I like this one because it gives you a nice long game, but without having to deal with the tedium of managing a zillion planets.

Although the AI is already quite capable and will put up a somewhat decent fight, it's still very much WIP. There are a lot of things the AI can't do or doesn't get right, our main AI dev (Dilvish) still has long road ahead. We're still in Alpha, after all

Ah, okay. Well, something to add to the to-do list, if people agree this is worthwhile. Even if this particular situation wouldn't come up with the standard universe table, they should at least know what a space elevator is and be able to assess whether one would be useful.

Actually, I think it does come up. There was a map in 0.4.3 where the AI would get stuck if you used low starlanes. I'll try to reproduce it and post the details.

[Vezzra]

How did you manage that? To get that, you have to tweak the universe generation scripts - did you do that?

Yes, of course. I've tried all kinds of variations. I like this one because it gives you a nice long game, but without having to deal with the tedium of managing a zillion planets.

Ah, I see Well, in that case don't be surprised if the AI can't handle the kind of map you get. It might still have some problems on some of the maps you can generate when you try certain more extreme game settings, so even more if you tweak the universe generation scripts to get something outside the scope it's expecting

Did you only tweak the universe tables, or did you actually play around with the universe generation Python scripts? In case you did the latter, how easy/difficult has it been to understand and tweak them? I'm asking because I'm the one who implemented scripted universe generation and haven't had much feedback on how "user-friendly" they really are...

Well, something to add to the to-do list, if people agree this is worthwhile. Even if this particular situation wouldn't come up with the standard universe table, they should at least know what a space elevator is and be able to assess whether one would be useful.

I guess Dilvish intends to add handling of space elevators at some point (if he hasn't already done so). It's just that his AI to-do-list seems to be a rather hefty one AFAIK, so it's a question of priorities

[Dilvish]

I guess Dilvish intends to add handling of space elevators at some point (if he hasn't already done so). It's just that his AI to-do-list seems to be a rather hefty one AFAIK, so it's a question of priorities

Correct; I haven't yet, but it is on my list. There are a few ways that the AI's handling of supply issues is still on the rudimentary side. To handle this issue well will require a better treatment of supply in some other respects, for which I've got ideas percolating but haven't gotten to implementation yet. I'll try to keep it in mind to put in place a preliminary handling of space elevators that will at least help the AI some in the kind of extreme cases raised here without making it waste resources on them in the more typical scenarios.

[Tualha]

Did you only tweak the universe tables, or did you actually play around with the universe generation Python scripts? In case you did the latter, how easy/difficult has it been to understand and tweak them? I'm asking because I'm the one who implemented scripted universe generation and haven't had much feedback on how "user-friendly" they really are...

Only universe_tables.txt. I looked at the scripts a little but they looked too hairy. I didn't try too hard, though, so this doesn't mean much.

[Tualha]

Actually, I think it does come up. There was a map in 0.4.3 where the AI would get stuck if you used low starlanes. I'll try to reproduce it and post the details.

I've tried to get that situation in 0.4.4. I was able to make it happen with low starlanes and low planets, but not with high planets. Perhaps the AI, faced with the map I saw in 0.4.3, would now cope better; universe generation has also changed, so I can't reproduce that map.

[MatGB]

Improving the way the AI works with low supply has been something that has been worked on over the last cycle, quite a fair bit, I tested a number of patches Dilvish worked on, and both of us generally prefer to work on Low planets (and frequently low starlanes as well) in order to support larger maps without the lag caused by huge numbers of planets.

So yes, it's likely the AI would work better on a map that previously caused it to freeze, plus of course some of the new Supply rules changes have made it easier for the AI to expand their supply (they use outposts on small/tiny planets a lot instead of space elevators, I also tend to prefer the latter but they are harder to code for in many respects).

[Tualha]

(they use outposts on small/tiny planets a lot instead of space elevators

Yes! I noticed they were using outposts that way. Indeed, they often ignored perfectly good large planets that I would have snapped up ASAP. More concerned about supply range than productivity, apparently.

I guess this is related to another problem: they don't seem to have realized that scouts are cheap and it's good strategy to send them out past your supply lines until they run out of fuel, then scrap them. They act as though every scout is sacred (every scout is good, every scout is needed, in your neighborhood). So they need that supply range for scouting.

Of course the whole outpost-on-small-planet tactic fails completely in the highly artificial situation I set up where there are no small planets at all. Tsk. Poor AIs, they don't have cable.