Best posts made by relentless

Could try phantom. There is some strange appeal to playing happily with then betraying friends.

such professionalism Jip!

p.s nice work on todays patch!

Add a reclaim metric to the post game graph. Actually I've never checked if its recorded during the game, I just remember it from the supreme score board during replays.

Thanks for organising Swkoll! and the players for providing some interesting replays to ponder over (and whoever did the twitch cast as they pointed out where the AI's were failing by comparison to the humans they were fighting).

This is great reading. Especially about the auth woes. Makes you appreciate just how much unpaid time is put in.

svenni_badbwoi has done some great work with some of the older maps and has really improved them, I can see some of them are already in this list. It would be great to replace some of the other maps with those versions on ladder, forbidden pass being one of them.

Awesome work on this Brutus! Takes a brave man to dive into legacy code.

The replays were ok for me,
To be fair Mylaur anyone above a 300 rank playing against a non cheating AI is going to have an easy time.
For giggles you could try RNGAI (self promotion, just make sure you read the description for setup tips) with a 1.5 cheat/build multiplier on that same map and see if your experience changes.

Getting the acu to be remotely in the realm of a human is damn near impossible(have spent days/weeks/months trying), an (almost) single threaded game just doesn't have enough clock cycles to do complex calculations at the speed to capture the data required to react and make decisions, you can do fake micro but like you said the acu doesn't really know 'what' to do.

Example for reference : getting aeon auroras the try and maintain weapon range on a target during combat takes approximately 7 function calls per second per unit plus supporting logic, if there are 50 T1 auroras on the field fighting that's 350 per second and 50 units is a small number.

In an effort to help identify areas of focus and with the upcoming changes to the default AI I'd like to get community feedback. For those of you that don't usually interact with the community but play against the AI this is your oportunity to influence its direction. I've prepared some questions to help guide discussion.

Who am I
I'm a rather low key member of the community, creator of the RNGAI mod and having done some work on fixing some of the obviously broken parts of the Default AI. I would consider myself a senior citizen in the gaming community having started in the days when a Commodore 64 was the height of graphical fidelity.

How important is the default AI to you?
With 3rd party AI mods available in the vault that have more development freedom and quicker release cycles that provide a challenging experience for many. Do we believe its still important for those that pick up SCFA on sale to have a great AI experience. If so where should the difficulty to be aimed at?

What type of play styles do you want to see from the default AI?
People have different opinions on what they want from an AIs play style, some want something that can assist in preparing them for ladder, some want an AI that is focused on team games, others want something that will crash units into their firebases, others want 30 minutes of quiet time to build a city.

Do you want an AI that is adapting to what you are doing?
If an AI is seeing a weakness do you want it to be exploited. Or do you want it to maintain a predictable base state. There are currently a number of AI 'personalities', which is a smoke n mirrors way of saying there are some different build orders and configurations that the AI use to build in a way that is desired by the player.

How important is efficiency of the AI?
Gaps in AI efficiency have historically been covered by increasing multipliers. Is it important to you that the AI not cheat or are you happy for it to waste resources and aim for doing some of everything rather than focusing on efficient gameplay.

Do you want the AI to communicate with you?
Some AI's make use of taunts. Is this something that improves your AI experience or does it quickly annoy you? Do you want the AI to point out intel oportunities and request help during team games.

Is the AI's ability with the ACU important to you?
The default AI's ACU capability is weak, but its also one of the most difficult things to get right. Some of us have spent over 100 hours just on this one aspect of the AIs development. Do you care that the AI isn't using its ACU effectively?

What map sizes do you want the AI to perform best on?
The balance of FAF lends itself towards maps that are less than or equal to 20kms. Do you believe future AI development should focus on specific map sizes or types? If so which ones? Please don't say Astro.

What things about FAFs AI is most frustrating to you currently?
Do they steal your mass points all the time? Did they ignore you while you were being killed by gunships when they had a massive air force sitting doing nothing?

Do you see an obvious development focus for the default AI?
Is there something that you believe would give the most bang for buck in how the AI plays if it was improved/added?

How important is mod support to you for AI?
There are plenty of mods but not much mod support for AI. Is this something that the community feels strongly about? If so which mods do you feel are not supported at all that would be most beneficial.

Which game phase/Theatre is most important to you?
Alot of development effort goes into unit functionality for AI. Do you care about the T1 phase of the game when it comes to AI? Do you feel that the T4 phase or Naval is the most important to improve?

Other things that are important to you or you would like the AI to do better/ differently?
Open forum.

A note on the realities of FAF AI development.
The SCFA game engine is VERY old and people can only put in the time they have available after their commitments in life. So plugging openai into it or developing a neural net is not something that is going to happen, nor is it realistic to expect someone to to dump 20 hours a week into AI development. There are engine limitations that make some very obvious things very difficult or computationally expensive to work around. An AI being flabberghasted by a wall comes to mind.

It depends on if your going to enforce restrictions.

If your not playing with 100 mods and are happy to play against the AI like you would in a normal pvp game then the list of AIs goes something like.

M27AI - This is an efficiency king and at least 10% better than any other AI out there right now.
RNGAI - This is a much less efficient AI but can still give you a fun game.
UvesoAI - This is a more compatible with mods AI but is also turtle focused.
DillidalliAI - This is good fun on open land maps.

The important thing is that these all have different play styles.
There are a few others I didn't mention but you can track them down in the ai section of the faf discord if you still want more.

I'd say if you want to get a feel for which AI is good for your play style. Test them in 1v1 scenarios a few times on a classic map like open palms or something. Then you can explore team games.

Keep in mind that these AI's are not frozen in time. They are improving over time.

Enforcing restrictions is not really top of mind when someone is developing an AI so they don't support it well. If you want to win, just play a worse AI.

The details person in me wants the blinking lights to match the faction colour if they are static. But I couldn't see a blue blinky light.

I'm keen to have my AI playing in it. Though I'm in a terrible timezone so will will need to recruit someone for the support side of it.

I did a quick round of map testing and found issues with these two maps
Open Palms
Regor VI Highlands

Open Palms has land pathing issues. The default path markers and the Uveso generator make the AI think they can walk to the two side mass point areas but they can't. Would need adjusted pathing markers done.
Regor VI Highlands has 2 mass points that are too close to the map border so the AI can't build mass points on that. (lower right half of the map and upper left half of the map is where the two mass points live).

Both are great maps though from a gameplay perspective.

Setting up AI's to fight each other is ok to get a base number.
Good to have as side information side (another tab on your spreadsheet maybe? its easier to setup an AI game and walk away if your low on time). Small note here, recently the skirmish game mode was made to allow the human to become observer so you could run these on the offline executable without spamming replays to the replay server.

But I kinda agree with tatsu in that its not representative of how those same AI's will fair against humans and there needs to be a 'feel' indicator, especially when it comes to how they play with different human play styles, map preferences, cheat multipliers and even ai game settings. One might do ok against a turtle but garbage against a pressure player and vice versa. Its all 'eye of the beholder' sort of thing, but as someone whos stared at AI games far too much its worth something.

Curious to know what your after from the mod here Rogue_Price. Are you interested in improved enemy AI (harder? smarter?) or improved supporting AI (the ones that assist you etc).

Uveso AI has a marker generator that will help alot of these sorts of maps (rng still requires markers but if uveso is enabled then it will be fine).

There will be an update within the next 6 months that will help the ai marker situation significantly.

Though Sorian won't be getting much love as its going to be pretty redundant compared to the other AI available.

oh nice! didn't know his voice, in which case 2x thanks!

and Utopian Uveso would have done better but there was a little communication break down as to which variant of Uveso was used, the one selected was more suited to turtle game play than his rush one is. Next time it'll do better.

Having an option for min.max team spawn distances would be nice. Often it will place players really really close together when its not desirable.

Also agree with the Naval comments. It takes quite a few regens to get something that is playable and damn near impossible to get something similar to say white fire.

In your example you are using the unit id's in that function, its like saying you want to put someones name into a car instead of a person and wondering why they are not in the car.

What you would want to do is create the units first, for example. Actually I'm not sure how the units are being created but you get the idea.

local unit1 = armybrain:CreateUnitNearSpot('UEB0101', posX, posY)
local unit2 = armybrain:CreateUnitNearSpot('UEB0101', posX, posY)

ScenarioFramework.AttachUnitsToTransports({unit1, unit2}, {transport})

p.s just to clarify your confusion. The function takes a table of objects rather than the table being an object itself.

The blueprints-ai.lua file that Nomander referenced is what you want to look at.

Your obviously motivated to do something positive here. I don't believe we can dynamically adjust threat during the game as the blueprints are read once on startup sort of thing. BUT, I think the development team would be more than happy to see someone put time into improving the existing threat calculations that we are creating. The current calculations are based on a formula that Balthazar did however long ago and some adjustments since then and while it is ALOT better than what we had originally there is room for improvement.

You'll eventually hit limitations since threat also depends on situational factors (a sniper bot in dangerous unless an enemy Obsidian tank is sitting next to it sort of thing). So why not rather than making a mod look at improving what we are doing now? We're happy to collaborate and provide feedback on the impact it makes on AI games.

I think we need to remove those AI log statements so they don't confuse users. Also I think the transport debug is still enabled.