Return to “Suggestions”

Post

Re: Procedural A.I.?

#16
Gazz wrote:One of the tricky bits about AI is to teach it to see.
Sure there's a hostile cruiser 10 km away. Now what does that mean? Is it a danger to me or something I care about? Will it be a danger if I continue my course? Space Is Big, after all. Can I tie up an entire enemy cruiser by having it chase my little patrol boat all across the sector for no reason?

Computers are great at finding facts. Connecting facts to meaning and goals... that's the tricky part.

Once the AI understands the situation, has dissected it into variables and local force concentrations, doing something about it is comparatively easy.
Many 4X games are reasonably good at this (massing fleets near enemy system? Expect war dec or at the very least a stern warning very soon). Most of the AI in those games are also smart enough to recognize that chasing a patrol boat with battleships is "probably a bad idea" (though that doesn't stop at least some of them from trying). Too bad this hasn't moved into 3D space games.
Post

Re: Procedural A.I.?

#17
If a fleet always has a current task and a policy, then those could be enough to determine behaviour. For example, the task could be to guard a region, and the policy to attack any intruder but not leave the region. That means they'll go for any small enemy craft that appears, but give up the chase if it manages to exit the delimited region. Alternatively the task could be to wait for an enemy, and the policy to aggressively hunt down any enemy it encounters (or unless it's too big, in which case run away). Then it will pursue the craft until it's destroyed or out of pursuit range.
Post

Re: Procedural A.I.?

#18
I found something the math nerds might be interested in: Build-Order Optimization in StarCraft
In recent years, real-time strategy (RTS) games have gained interest in the AI research community for their multitude of challenging subproblems — such as collaborative pathfinding, effective resource allocation and unit targeting, to name a few. In this paper we consider the build order problem in RTS games in which we need to find concurrent action sequences that, constrained by unit dependencies and resource availability, create a certain number of units and structures in the shortest possible time span.
Click the link to the PDF document and there are actual formulas of how the machine decisions are made. It's way over my head.
Post

Re: Procedural A.I.?

#19
mcsven wrote:The issue was that the AI had no way to assess you as a threat. Thus post campaign, when I was flying a Sabre or Titan with class 10 weaponry on every hardpoint, I'd find Liberty Rogues pilots in Bloodhounds taking me on. These guys were literally one-volley kills but their behaviour was only governed by my standing with their faction and so they would valiantly perish in large numbers.
I clearly remember some encounters with Liberty Rogues where they just scattered after seeing my Sabre, complete with the 'He's too strong for me to take on, retreat!' chatter, in the mod-free game. It didn't happen all the time, so I guess algorithms for that weren't optimised well or were too simple and not taking into account the amount of Nomad weaponry I've carried :)
Image
Survivor of the Josh Parnell Blackout of 2015.
Post

Re: Procedural A.I.?

#20
outlander4 wrote:
mcsven wrote:The issue was that the AI had no way to assess you as a threat. Thus post campaign, when I was flying a Sabre or Titan with class 10 weaponry on every hardpoint, I'd find Liberty Rogues pilots in Bloodhounds taking me on. These guys were literally one-volley kills but their behaviour was only governed by my standing with their faction and so they would valiantly perish in large numbers.
I clearly remember some encounters with Liberty Rogues where they just scattered after seeing my Sabre, complete with the 'He's too strong for me to take on, retreat!' chatter, in the mod-free game. It didn't happen all the time, so I guess algorithms for that weren't optimised well or were too simple and not taking into account the amount of Nomad weaponry I've carried :)
There were some places like that, yes, but they were few and far between.
Have a question? Send me a PM! || I have a Patreon page up for REKT now! || People talking in IRC over the past two hours: Image
Image
Image
Post

Re: Procedural A.I.?

#21
mcsven wrote:I was mulling over Freelancer today and remembered one thing about the AI that bugged me when I got to the higher levels. I'd like to turn it into a suggestion for LT, so I figured I'd add it here rather than add a new thread; if it's a duplicate of a previous suggestion then my apologies.

(Note: I never played any of the Freelancer mods, so what I'm about to describe may have been fixed in one of them. It was definitely a characteristic of the vanilla game however.)

The issue was that the AI had no way to assess you as a threat. Thus post campaign, when I was flying a Sabre or Titan with class 10 weaponry on every hardpoint, I'd find Liberty Rogues pilots in Bloodhounds taking me on. These guys were literally one-volley kills but their behaviour was only governed by my standing with their faction and so they would valiantly perish in large numbers.

My suggestion is that AI pilots have two levels of logic that activates when they detect you. The first layer assesses your standing with their faction. If it's good or neutral then no further action is necessary (leaving aside potential actions governed by their procedural AI or bounties on your head). The second layer is where they assess the total strength of you and your forces against themselves and their forces. Only if this calculation indicates a favorable chance of success will they attack.

As usual there are plenty of areas for interesting gameplay mechanics here. Perhaps your faction alignment is different to your reputation as a pilot? Or perhaps their total-force determination includes distance to friendlies, so that NPC pilots know if they're on their own. This may also allow you to STOP ongoing fights by offering protection to one side...
Having gotten to that point in Freelancer at least twice, i had similar issues. One i had included several waves of a factions System defense ships attempting to kill me. [Keeping in mind that i'm not in the middle of nowhere, so they're weak as hell]
Freelancer was just frustrating in that respect. It was equally frustrating when, outside of specific objectives and scripted events, nothing over the size of a transport or freighter could be destroyed, even temporarily, including Docking Rings, Space stations, Jump Gates, Trade lanes and Battleships. I mean, i can sort of understand it, considering how the game perceived time [inaccurately, and rapidly, when on stations or jumping between systems] if you can script a damn battleship/docking ring's death, i'm sure you can include it on every battleship/docking ring, not just one or two, and without gamebreaking results, as you can have them respawn [See: Rebuilt] within a set amount of ingame [or potentially out-of-game] time. It would also have been a lot more interesting to keep the in system Battleships on regular patrols around the system, rather than staying at one specific point permanently [i mean, they don't have an engine for decoration... :P].
But considering the amount of time and effort shown in the end result of the game, all of the above was apparently too difficult with the time/personnel constraints at the time.

Judging from Josh's progress in the Dev logs alone [i don't watch the clips he does too. I don't watch many videos at all online :monkey: ] he's well beyond any of these trivial concerns and progressing onto solving issues most of us haven't even realised are issues. Keen for LT Release, so i can haz this, and back Josh at the same time. :D

P.S. I goddamn hate some systems in Freelancer. Specifically, the ones that do Radiation damage at a set percentage of hull over time, rather than a set hull health amount. There should be a difference between going into one of these systems with a light fighter or Freighter, compared with flying through one in a Sabre/Titan...
Post

Re: Procedural A.I.?

#23
johncage wrote:and here we go with the rampant fantasizing and hope build-up. you do know you're just setting yourselves up for disappointment right?
That's why I posted that PDF with some formulas for AI strategy. It's enormously complicated.
mcsven wrote:I was mulling over Freelancer...The issue was that the AI had no way to assess you as a threat...
But at the same time, it's amazing to me something like this event got overlooked by a major game developer. Game Theory 101 is a payout matrix, which this flaw should've been discovered.
Post

Re: Procedural A.I.?

#24
johncage wrote:and here we go with the rampant fantasizing and hope build-up.
So? The entire game is one man's rampant fantasy.
With some input of the local residents to be sure but don't think for a second that we have a chance of significantly altering "the vision". If there was a Mule School of Stubbornness, Josh would be giving lectures.
There is no "I" in Tea. That would be gross.
Post

Re: Procedural A.I.?

#25
johncage wrote:and here we go with the rampant fantasizing and hope build-up. you do know you're just setting yourselves up for disappointment right?
Perhaps. Or we are just too used to the old way of making games - where everything has to be made by hand - that only now we can dream realize the full potential games can have?
Post

Re: Procedural A.I.?

#26
Gazz wrote:
johncage wrote:and here we go with the rampant fantasizing and hope build-up.
So? The entire game is one man's rampant fantasy.
With some input of the local residents to be sure but don't think for a second that we have a chance of significantly altering "the vision". If there was a Mule School of Stubbornness, Josh would be giving lectures.
:clap: Bravo, Gazz, that's a fine appraisal of the reality of Limit Theory. I thought every member was aware of these truths. :lol:
Post

Re: Procedural A.I.?

#27
Wow, this deffo wins the "longest wait between replies" award, given that my original post was made in March 2013. 8-)

I think that the intervening period between that post (which was one of my first) and now has indicated to everyone's satisfaction that the AI in LT is likely to be considerably more sophisticated than the one in Freelancer (as Nebuchadnezzer2 said). I would be surprised (and a little disappointed) if the kind of behaviour I complained about was exhibited by LT AI; or, if it was, I believe that the game should communicate with you why the AI took the action. I don't think anyone would complain about some AI that conducted a kamikaze attack on your vastly more capable ship or fleet because it turned out you killed his brother a while back and it was made clear that was the motivation behind his actions (in fact we'd be blown away - but you get the point).
Post

Re: Procedural A.I.?

#28
Well, Japanese performed kamikaze attacks with their fighters on USA for a reason... I did so too in War Thunder a couple of times. When your guns are jammed, you're damaged, have a bandit on you and a badly damaged enemy unit in front of you then performing this last act is a sound tactical decision. But I agree, I'd like to see the reasoning behind it and more complex behavior than "I am following orders without question".
Post

Re: Procedural A.I.?

#30
Since it seems to fit here:

What about simulating reaction times by introducing a bit of lag in the AI's reactions, and skill with guns or lack thereof by some random aiming error?

That would allow simulating anything from an extreme rookie (has trouble hitting a cruiser from 200 meters distance) to an ace pilot. Also, these stats could improve over time to simulate growing experience...

Online Now

Users browsing this forum: No registered users and 20 guests

cron