For Backers Only: AI building stuff

Will there eventually be a few ai profiles that play less like a pvp opponent?

I don't play pvp and never have in rts, my interest in this and other rts games for the past two and a half decades has been singleplayer or co-op vs ai. The ai in PA so far puts all of its focus into attack, and nothing into defense.

Perhaps what I am describing would fit a 'turtle' profile, but in other rts games 'turtle' is more of a lazy ai who uses annoying weapons like nukes and long range artillery to win. I'm looking for an ai profile that focuses on self preservation; its goal would be to survive against player attacks for as long as possible by building defensive towers, walls and other forms of protection to guard its base against attack.

I've not found an ai that plays like that in several years. All the ones I've fought of late are okay at offense but their defenses suck. It's the most un-satisfying way to win, when ~45-50 minutes in I push an assault and flatten the enemies base in seconds. Like I spent all that time building up to a huge confrontation, only to find that had nothing to repel my assault.

I know the ai is far from done, but if I don't name what bothers me now there may never be change. Thanks for reading.

I played a game where the AI seemed to be pretty slow attacking, until I discovered that it had spawn in the middle of a lake and had amassed a ton of air fighter over a sea of mex/power. As it started expanding to the edges of the lake however it started building a few v/b factories and that's where it was attacking me from. If it had known to prioritize air offensive however I'm sure I'd have been crushed.

I think Sorian will give us enough possibilities for the AI, believe me

It's difficult to comment too much due to not knowing what's changed between the version we're playing and developement, and what elements sorian is putting in vs. those he intends to do but not until later when more of the mechanics are in place.

So here's my thoughts on the AI in 56516

1. AI always starts energy when it should start MEX
2. AI's balance of energy to MEX is poor. It seems to build far more energy than it needs or will need. It might be because it uses air fabbers which are energy greedy.
3. The AI needs to build fabbers in pairs. It's fine to build two fabbers then a couple of combat units (for defence and scouting on smaller maps), but it needs to be two fabbers not one.
4. AI needs to factor in the size of map and number of opponents when deciding on its fabber to combat unit ratio early on.
5. AI has fabbers building alone. A fabber should never build alone, they should be working in pairs minimum.
6. AI builds combat units early but doesn't do anything with them. In that case it shouldn't be building combat units. Either scout or build more fabbers.
7. Early combat units should move in pairs, they should be doxs and they should be scouting and raiding.
8. AI starts things other than bots. There's no reason to do this, always start bots because bot fabbers are the only fabbers worth building. They're more efficient and they're faster than vehicle fabbers. Vehicle and air fabbers should emerge only to build their tier 2 factory, and you should only build one and then assist with bot fabbers.
9. AI loves fabrication aircraft. No. Stop. They're terrible. Build scouts, build fighters. Right now air is for tier 2 bombers, and that's it.
10. AI goes air too early. It's great for scouting but you need two or three other factories first because air isn't going to defend against anything early, it's for scouting and annoying fabbers on the fringes. Even on a larger map you want those early bot factories because that's where you get your fabbers.
11. AI doesn't build enough fabbers early. Eight bot fabbers by six minutes is the goal.
12. AI builds too many lasers too early and too close together while under no pressure. It should be rapidly expanding its metal with fabbers working in twos and threes. These should be building missile turrets. Later on, or in metal heavy locations, it can build laser towers. But early on it should be expanding fast and hard.
13. AI doesn't build enough factories, though this is linked to it not building enough fabbers. At least four by six minutes, preferably with a fifth at least started.
14. AI will clump its scout planes reducing their effectiveness. They should be scouting in different directions and using long sweeping patterns. The AI uses short moves leading to them constantly slowing down and having to accelerate again. This leads to its scouting being less efficient than a human who will simply spam moves East or West moving in a North/South sweep.

Quitch brings up some good points.

Still, I'm mainly here to say well done Sorian. I've never been a big fan of playing AI as I always found them boring and samey in other RTS games. Today (mostly because of low player numbers online) I decided to have a go.

I've had a crack at the PA AI a couple of times now and its WIP nature has always been really apparent; I think previous to this one I've only had one game where the AI actually built combat units!

When I jumped in today, I didn't see anything for the first 15 mins and pretty much thought the AI had spawned badly and experienced a false start. I rushed orbital, but just as I got my advanced radar in the skies, I was attacked by a unit blob the size of which I've only ever seen a player produce.


They proceeded to pound down my defences and gut out the inside of my main base before my own army could intercept them and destroy them.

Long story short, it was one of the most satisfying games against the AI I've never had in any game. So thanks for all your hard work and I am seriously looking forward to what PA's AI will become in the future.

Two little things I noticed - I ended up nuking the enemy commander as my frame rate crashed while I was building my game ender army. After I nuked him, he seemed to develop a serious case of shaky legs. His legs would vibrate left and right and twitch in either rotation direction as he was walking.

Also has this happened to anyone else?

You pressed F10/F11 I think, which should display the pathing difficulty which units use to know where to avoid to get from A to B.

F11 doesn't cause the game to break though... ZaphodX uploaded a video (5 Vs 5 Realm, Water planet), post #103.

F11 just makes the planets passable terrain glow white. Impassable goes red.

Not quite. The red area surrounding water is beach or shallow water. You can build water building on it and land units can path though it. But you cant build land buildings or path water units over it.

The red and yellow border means ground units cant path though it.

It doesn't display properly on my system then. Never seen red and yellow.

IIRC an impassable obstacle is surrounded by a red line and then a yellow line. I didn't mean to imply that it is a red and yellow line.

*Sees fabber get killed while building orbital * *sees another fabber begin a new orbital despite the old one being almost finished *



We have a problem.

Fixed.

With newer AI out I think this is a good time for an observation update.

Threat response is fascinating. I had a blob of stuff between myself and the AI, and as soon as i moved it out of the way, the AI tryed attacking that very spot. I was impressed. Unfortunatly, its stretched out army got vaporized by my flanking blob. Seeing threat response was cool. Just got to get it a bit better awareness. Had it skirted furhter from my moving blob, it could have made a real mess of my base before I could chase it down.

More work for you sorian

Still seeing AIs sporadically entering suspended animation. Typically they get some eco and a factory up, and then idle. Don't use their units until they get attacked, and don't use their com again until it gets attacked, at which points it will repeat the above before again entering suspended animation. If it doesn't happen before the second factory, I find that it doesn't happen.

The AI had me on the ropes fairly well however after managing to resist it long enough, at some point it just seemed to give up sending mobile units and expanding, and instead concentrated on building Holkins, Catapults and Nuke Launchers, which in themselves wouldn't have been a bad idea but I had full anti-nuke coverage, and my borders were lined with Holkins myself. They definitely had territorial and economic advantage though, but never seemed to utilize this to continue to outpace me and eventually I managed to win as I amassed enough mobile units and surged them from multiple angles and accidentally popped their commander who can gone for a walk far outside the safety of their main base.

the best way to kill the commander is just to hunt down fabbers in the main base. sooner or later the fabbers job will pass to the commander, and he will walk to his death

Sorian needs group move implemented so that the AI would also have moved as a blob.

I think the AI's chief weakness is "defense against the major types of assassination".

Generally I just send T2 bombers at him and he blows up. Very little air cover.

He also tends to stand out in the open so I could probably nuke him.

Personally, late game, I set my commander to assisting a T2 air factory that's building anti-air planes which are set to patrol right around that area. Beside him is a deep space radar, an umbrella and an anti-nuke. This pretty well covers him from the major/easiest types of assassination methods.

Any commander caught wandering away from these types of protections is asking for one type of death or another.