Post news RSS LOUD AI - March 2018 - Notes from Sprouto

A March 2018 update on things going on with The LOUD AI Project for Supreme Commander: Forged Alliance.

Posted by on

March 2018

As we move thru March now, things are still very busy in getting the public release of LOUD settled down. We're getting lots of data on the many ways that people play the game, and as we see new things, we try to make LOUD adaptable to them. This takes time and repeated observations to get it right. Here's an example;

With respect to the naval AI and making it more efficient, we needed a way to make LOUD responsive to particular situations - such as a water map that has very little water, or more importantly, a map which has a decent amount of water - but the enemy is not bothering to play a naval game. Both of those are important considerations - as otherwise LOUD will build a sizable naval force that he can't use or doesn't need. Three metrics are used for this - the water map percentage - the unit cap - and the naval ratio.

The concept is rather straight ahead. If a map has a low water percentage, or a low unit cap, we want LOUD to really curtail his naval investment. That means two things to him, less naval factories, and lower minimum pool requirements for him to build. As for the other situation, where LOUD cannot find any enemy naval presence, we use the naval ratio to tell him to cease construction of any new assets - other than his minimum requirements.

These changes will require some observation to tune them better - so we'll be watching naval replays with great interest for the next while. As for other naval progress, all the unit category changes are now in place for BOMBARDMENT. This means it will now be possible to create a platoon definition that requires only those kind of units and build a behavior to go along with it. That's a short sentence - but a mouthful of work. Platoon definitions are easy - but behaviors take a lot of rinse and repeat to get 'just right'.

Another development for naval features this month includes two other minor, but interesting changes. First, all T3 Naval Factories gain the ability to build Hovercraft and Torpedo Bombers. Secondly, the Economic threat for ALL factories goes up substantially. One is a feature for our players that won't initially be used by LOUD, the other is important to LOUD as it will cause him to pay much greater interest to any factories that he finds. Be wary - the goal here is to make LOUD almost rabid about your factories - you've been forewarned.

As promised last month, the Stealth Field generators for all factions now gain T2 level teleport jamming. The addition of factional support 'aura' will come later along with the promised radar tower additions. We're engaging in a wider discussion of radar and the addition of new intelligence assets so for now, we're not rushing ahead with the implementation of T4 radar structures.

We've added another very minor but requested mod - Realistic Nukes. This one is dead simple - all static and sub based nukes simply do more damage - quite a bit more - and have a slightly larger radius.

We've added two new maps to LOUD. CityWars III - we released this before, but it had some texture issues which we've resolved. This is a fine team play scenario where you're stuck in the middle against overwhelming odds - best used with a 2 hour time limit. The Pass - our second 10k map addition, is a very simple tactical problem, but we've provided 3 different configurations. And, Hills of Havoc Wet. This is another 'wet' version of an existing map - but with a twist. While there is a lot of water surface area, none of it should be deep enough for Naval yards. This one should feature amphibious operations. Aside from the standard 4 v 4 configuration, we'll include a 7 v 4 configuration for our comp-stomp fans.

Some unit tweaks upcoming include increased mass and energy storage for factories - nothing huge - but enough to notice.

Started looking into why amphibious actions are so lacklustre. Generally, when faced with an amphibious challenge, LOUD seems somewhat ho-hum. Partly it's because he's not concentrating his amphibious assets, and he's not focusing enough effort on building amphibious units. I've addressed both of those this month - LOUD will now move his amphibious forces to the best forward postion - until now - it was random - it could be either of his primary positions. As for production, I've streamlined his production on water maps - this should get him building somewhat more amphib units.

With the introduction of the Hills of Havoc Wet map, I started running some sims with an increased focus on amphibious work. Right away, I was able to pickup a bit of data on why the amphib work can be so unfocused, and aside from the concentration issues I mentioned above, his production needed some tuning as well. It was also clear that without a naval component, he wasn't interested in creating any Naval DP's and that had to be addressed. The Naval DP is a valuable position for LOUD, even where naval units are not possible, simply because it gives him a staging point that may be much better than any he could find on land. So now those work well.

Found some issues with land scouts. For some time now - when LOUD sends out land scouts, he'll send them to a point that needs scouting - and then he'll abandon the platoon there - rather than trying to pick them up and move them. This works very well - as it frees up the platoon for LOUD to create another -- and the premise of this was great - he'll get to a point where he's got all the positions he needs to scout being observed all the time. Sounds great. It's not doing that. Why ? Well, it has nothing to do with the platoons or the intel behaviors - it has to do with the army pool.

The army pool is a global 'platoon' where every unit that isn't in a platoon goes to when it isn't in any other platoon. Using the Army Pool to do unit counts is really light on cycles so I use it a fair bit - for example - to make sure I don't overbuild scouts. Guess what. It sees all those abandoned scouts as part of the pool - and as a result - I don't build any more. Duh. Had to fix that - I put abandoned scouts into an alternative pool now - the Structure Pool - where they'll never get seen again. Fixed ! Not quite. Now LOUD is making all kinds of scout platoons again - but - he's making them so fast they tend to go to the same location - that's also not good. So - I reduce the number of active scout platoons to one - so he's only got one underway at a time, and that's good - less plato memory reserved. We'll see how effective it is.

Fixed a nasty little bug that was causing some AI issues. It seems that the Black Ops ACU has the 'DIRECTFIRE' category. Nothing wrong with that except that Bob tries to get into a fighting platoon and get into a transport - bad bad bad. Bob must not fly away - Bob must stay and work his base. The stock ACU does not have this category, and in hindsight, this could have been a problem for quite some time - only recently revealed when I was streamlining combat platoon compositions. Also, for the mindful, I've decided to re-introduce the inability for T1 transports to lift the ACU. This is especially evident when you look at the Black Ops ACU with his 40K+ HP. We don't let anything with that kind of awesomeness get into a transport at all - let alone T1. There is a case to be made for preventing the ACU from being transported at all on that line of thinking -- but hey - I'm not a radical.

Looking at the T1 > T2 transition. It's really pretty cheap all things considered. Most LOUD players acknowledge that there is a 'bite' when you go from T2 > T3, and that 'bite' is partly responsible for the lifespan of T2, but the T1 > T2 transition is almost trivial - being that it can be supported by about 10-12 T1 power gens. I'll be looking at a more expensive Energy cost for that move - both for the factories and the ACU.

I think I'm finally happy with the physical behavior of transports. After the last pass, while we got their flight characteristics settled up for normal things, it was clear that we didn't quite understand the loading and landing behavior very well. So, after some careful focus on that, we were able to restore the fairly efficient load/unload sequence for transports. This should make it possible for them to drop their troops fairly quickly and get away in a smooth transition.

We've made an exciting new contact. The author of BrewLAN has joined us several times on our Discord server, and he has energized the entire team in a number of ways. At this point it looks likely we'll be making an adaptation of BrewLAN specifically for the LOUD AI Project in the very very near future. There is a lot of support for this - not just because BrewLAN is a popular mod unto itself, but because there are many units which either fill a void in the LOUD arsenal, or offer a better looking and working model of a unit we already have. I'm really pleased that everyone is getting involved in the discussion. It won't happen overnight - but users should start seeing something fairly soon.


great work!

Reply Good karma Bad karma0 votes
Sprouto Author

Thank you - let us know how some of your games go - we thrive, and LOUD improves, on that kind of feedback.

Reply Good karma+1 vote
Post a comment
Sign in or join with:

Only registered members can share their thoughts. So come on! Join the community today (totally free - or sign in with your social account on the right) and join in the conversation.