open All Channels
seplocked EVE General Discussion
blankseplocked GIVE US NPC AI
 
This thread is older than 90 days and has been locked due to inactivity.


 
Pages: first : previous : ... 7 8 9 10 [11] 12 13 14 15 ... : last (21)

Author Topic

Bartholomeus Crane
Gallente
The Crane Family
Posted - 2008.12.02 12:04:00 - [301]
 

Edited by: Bartholomeus Crane on 02/12/2008 12:06:15
Just to lighten the mood a little:
Originally by: A Few Good Scripters

CCP: You want AI?
Bartholomeus Crane: I think I'm entitled to it.
CCP: You want AI?
Bartholomeus Crane: I want the truth!
CCP: You can't handle the truth! Son, we live in a world that has NPCs. And those NPCs have to be guarded by men with scripts. Who's gonna do it? You? You, J'Mkarr Soban? I have a greater responsibility than you can possibly fathom. You weep for Serpentis Rat #1101 and you curse the coders. You have that luxury. You have the luxury of not knowing what I know: that Serpentis Rat #1101's death, while tragic, probably saved person months. And my existence, while grotesque and incomprehensible to you, saves person months...You don't want the truth. Because deep down, in places you don't talk about at parties, you want me writing scripts. You need me writing scripts.
We use words like scripts, coding, loot-tables...we use these words as the backbone to a life spent writing scripts. You use 'em as a punchline. I have neither the time nor the inclination to explain myself to a man who rises and sleeps with the monetary gains of the very scripts I provide, then questions the manner in which I provide it! I'd rather you just said thank you and went on your way. Otherwise, I suggest you pick up a laptop and write a script. Either way, I don't give a damn what you think you're entitled to!
Bartholomeus Crane: Did you order the coding of scripts?
CCP: (quietly) I did the scripts you wanted me to do.
Bartholomeus Crane: Did you order the coding of scripts?
CCP: You're goddamn right I did!!


J'Mkarr Soban
Posted - 2008.12.02 12:37:00 - [302]
 

*smirk*

Also, I've got to say that missions in general are boring, even without the rats. All this talk of the time it would take to go through every single mission to see if the new AI works - why bother? Why not create dynamic mission creation on the fly? That's going to be a lot easier to balance, because you can balance it abstractly. An overhaul of missions and rats is desperately needed, but can be fixed in such a way that they'll never need to be looked at again.

Coriander Rinne
Federal Navy Academy
Posted - 2008.12.02 13:24:00 - [303]
 

Long topic is long and I just skimmed it, but the overwhelming focus of the topic seems to be related specifically to AI. The problem with introducing AI changes is that the PvE structure (specifically missioning) is still as stale as ever. Take a mission, look up every detail on it using the in-game browser, fit accordingly, and take it out solo according to the guide.

I don't think we necessarily need AI changes to make the PvE more interesting; the focus instead should be on unpredictability and a more dynamic and group-oriented experience than we have now.

This agent/mission system should remain in place for the solo player, but add to those agents the ability to participate in "patrols", which would then spawn a dungeon consisting of randomized levels and factions that offer more rewards as the group progresses further in.

Perhaps to make things even more interesting (let's make the PvE as hardcore as the PvP), the gate locks five minutes after you first activate it, forcing you to proceed through the entire dungeon without warping out. No changing your hardeners! >=D

Perhaps the gate locks itself to individual players after thay activate the gate, meaning that late arrivals can join a patrol in progress, as well as allowing intruding probers in on the fun.

Ideally, the random factions from room to room combined with the locking gate would require the group to prepare for every possible situation, making this an exceedingly difficult task for a solo player.

These rooms may include, but aren't limited to:
-triggers both obvious (single large bounty battleship)
-and unexpected (did that spider drone just call in reinforcements?)
-environmental effects (fighting in the middle of the Recon 3/3 cloud would be awesome and you KNOW IT)
-named spawns
-ambushes
-"impossible" spawns (perhaps destroying a nearby structure despawns them, or they show up upon clearing the room, forcing you to dash for the gate)
-industrial convoys (like the hauler spawns in asteroid belts)
-sharks with lasers
-swarms of a specific ship size/type
-and on rare occasions just to freak the player out, absolutely nothing.

A well prepared group should be able to crawl through the dungeon (maybe 6-10 rooms) in 2-4 hours, with a scaling reward depending on the deepest gate activated.

A fun and viable group PvE mechanic has been sorely lacking in this game, and I think such a thing would go a ways towards fixing it, while providing a challenge that forces Joe Dominix and Sally Raven to get a bit more creative with their fits.

The biggest problem with this is that if it's implemented in highsec, that's yet another reason never to leave for missioners. On the flip side, if it's implemented only in low/nullsec, risk/reward would be skewed so much that it won't be profitable enough to participate in for most people. L5's, anyone? THAT, I have no idea how to fix.

As for NPC AI: for the love of god don't implement fleeing NPCs. There's nothing more annoying than chasing that one idiot that bolts away faster than you can run after him.

The Tzar
FinFleet
Raiden.
Posted - 2008.12.02 13:27:00 - [304]
 

Originally by: Tobias Sjodin
These things are really important to make PvE fun:

1. Unpredictability

2. Risk.

3. Reward.




All of these exist within PvP already present in game.

Why play an MMO for the AI in it's PvE?

Strategy games will probably give you better AI than mission running in EVE.

Kenpachi Viktor
Gradient
Electus Matari
Posted - 2008.12.02 14:47:00 - [305]
 

(This is probably more for mission design than AI.)

Two things
1. No such thing as a PVE or a PVP fit, or at lest very little difference.
If there is a good combat fit for a ship, it should be very much the same for fighting NPCs or players. including needing to use warp scrams and/or warp core stabs.

2. My understanding of the game mechanics is that, all (or almost all) NPCs are NOT pod pilots. Therefore; they do less damage, have less tank, slower to respond, slower movement speed, longer locking times, etc. And they make up for this lack by using numbers. But they should still use the same ships, and the same modules, but with higher fitting requirements (Most have less SP than the new players.)

Mephistophilis
ANZAC ALLIANCE
IT Alliance
Posted - 2008.12.02 14:59:00 - [306]
 

Edited by: Mephistophilis on 02/12/2008 15:01:08
I've always had a beef with the npc A.I tbh, they don't know how to fit there ships, don't know how to work together and quite frankly have NO balls!

Make them tougher with the kind of fit you expect, more random ship types in spawns. And after a while of being killed constently you'd think they'd get together and hunt YOU down and camp YOUR station. Maybe even go on the rampageShocked

IMO Ratting would be more fun and not something you just did while making the dinner. You could increase the reward because most of the rats would have web and scram fitted like the rest of eve thus increasing risk. And it'd sort out those pesky ninja ratters in 0.0 you can NEVER EVER catch, you know the kind i meenWink

Mephistophilis
ANZAC ALLIANCE
IT Alliance
Posted - 2008.12.02 15:00:00 - [307]
 

Edited by: Mephistophilis on 02/12/2008 15:00:40
gah sorry

Bartholomeus Crane
Gallente
The Crane Family
Posted - 2008.12.02 17:14:00 - [308]
 

Originally by: Kenpachi Viktor
(This is probably more for mission design than AI.)

Two things
1. No such thing as a PVE or a PVP fit, or at lest very little difference.
If there is a good combat fit for a ship, it should be very much the same for fighting NPCs or players. including needing to use warp scrams and/or warp core stabs.


Sorry, but this is simply not the case for a lot of PvP. For one, PvP is usually very quickly over for the player involved. A couple of minutes tops. PvE, especially in missions, involves fitting for much much longer durations. Thus you see a lot of active tanking and cap stable fittings in PvE, and a lot less of that in PvP. DPS, ECCM, Alpha, all these things are much more important in PvP than in PvE. Also, the rats are very very stupid compared to players, and are therefore easy to counter. All these counters can and are incorporated into PvE fits. Which you can find all over the forum. Fit those in PvP and you'll be up for a big surprise. PvP and PvE fits are simply very different most of the time.

Quote:
2. My understanding of the game mechanics is that, all (or almost all) NPCs are NOT pod pilots. Therefore; they do less damage, have less tank, slower to respond, slower movement speed, longer locking times, etc. And they make up for this lack by using numbers. But they should still use the same ships, and the same modules, but with higher fitting requirements (Most have less SP than the new players.)


You are mixing up RP with reality. The fact that NPCs are not pod pilots is a (poor) excuse for why they are so poor. There is no viable reason why NPCs can't be pod pilots. In fact, there is no viable reason why NPC AI can't be comparable to pod pilot behaviour. You're mixing cause and effect here.

Bartholomeus Crane
Gallente
The Crane Family
Posted - 2008.12.02 17:22:00 - [309]
 

Edited by: Bartholomeus Crane on 02/12/2008 17:23:07
Originally by: Mephistophilis
I've always had a beef with the npc A.I tbh, they don't know how to fit there ships, don't know how to work together and quite frankly have NO balls!

Make them tougher with the kind of fit you expect, more random ship types in spawns. And after a while of being killed constently you'd think they'd get together and hunt YOU down and camp YOUR station. Maybe even go on the rampageShocked

IMO Ratting would be more fun and not something you just did while making the dinner. You could increase the reward because most of the rats would have web and scram fitted like the rest of eve thus increasing risk. And it'd sort out those pesky ninja ratters in 0.0 you can NEVER EVER catch, you know the kind i meenWink


The whole problem with NPC AI is that there is no AI. NPC rats are a set of parameters and one script telling them how to behave. The parameters determine how often they shoot, with what damage, and those things. The script just tells them to fly to a certain distance and start shooting or webbing, or whatever their parameters allow them to do. There is no intelligence, there is no thinking, there is no learning, there's just a script and a number of parameters. Then there are some rules on how to handle aggression and where they should circle in their mission, and that's that. There are even databases out there telling you exactly what each type of rat does, all the parameters are there. And these parameters have hardly changed over the years, hence the situation that rat BS still fly at the speed they did before the speednerf.

No wonder things are boring, nothing ever changes!

H Lecter
Gallente
The Black Rabbits Academy
The Gurlstas Associates
Posted - 2008.12.02 18:06:00 - [310]
 

Sorry that I did not read 11 pages of suggestions. No need to flame me if I repeated something as I'm not going to follow up on this thread. Just my 0.02 ISK:

- Clever rats would (try to) retreat when their tank fails, force people to keep a point on them. Of course you should be able to finish the mission by just chasing the rats away, but you would lose the tasty bounties. Twisted Evil
- Let them use MWD's and allow players to use MWD's in the deadspace as well.
- Make rats (in belts) appear on your onboard scanner - maybe more people would know how to use it then Laughing
- Let the rats vary the damage type. I always found it strange that you had to fit so differently for PVE compared to PVP.

To sum it up - do everything you can to make PVE feel more like PVP minus podkilling.

P.S.: Iirc Elite had rats that would demand your freight to let you go. I'm sure a pve ransoming mechanic would be feasible.

Chooch Chooch
Chooch Inc.
Twilight Federation
Posted - 2008.12.02 20:28:00 - [311]
 

I had to laugh reading this. To answer all of your points. go to 0.0. Not only do the rats change targets mid fight, warp scramble, web, disrupt and ecm you. They occasionally warp off and come back with reinforcements.

Also if you are looking for intelligence in a fight keep your eyes on local for reds will be around soon and you might actually fight something that thinks.

Oh wait you don't want that cause you want to win everytime while feeling like you accomplished something.

Thebro Nobrunder
Schrodinger's Renegades
Posted - 2008.12.02 20:48:00 - [312]
 

Edited by: Thebro Nobrunder on 02/12/2008 20:55:33
Originally by: The Tzar

>These things are really important to make PvE fun:
>1. Unpredictability
>2. Risk.
>3. Reward.

All of these exist within PvP already present in game.

Why play an MMO for the AI in it's PvE?

Strategy games will probably give you better AI than mission running in EVE.


actually pvp has lots of unpredictability.
It also has tons of risk.
not a whole lot of reward though.

Given that you can make 10-20 million an hour with very little risk... you would have to be able to make considerable more than that through pvp for it to validate on a risk vs reward basis.


Bartholomeus Crane
Gallente
The Crane Family
Posted - 2008.12.03 01:02:00 - [313]
 

Originally by: The Tzar
Originally by: Tobias Sjodin
These things are really important to make PvE fun:
1. Unpredictability
2. Risk.
3. Reward.


All of these exist within PvP already present in game.
Why play an MMO for the AI in it's PvE?
Strategy games will probably give you better AI than mission running in EVE.


Personally I do more PvP than I do PvE. I hardly do missions anymore (to boring), and rat only to make Isk for PvP. I don't want PvE to overtake PvP, nor do I think its possible. AI can be pretty creative, but currently humans can out-think it. However, what I do want is to blur the line between PvE and PvP, so that those masses of PvE-ers out there, will eventually be able to switch from one form of play to another without getting completely frustrated. I see no reason why this would be impossible, and I know a thing or two about AI. The idea that AI is only suited for strategy games is a fallacy and a myth.

CCP Incognito

Posted - 2008.12.03 14:50:00 - [314]
 

Edited by: CCP Incognito on 03/12/2008 14:51:36
Hi all.

Just a quick note, we finished our first sprint and we have a plan for what we are going to do.

Now for the bad news, we are not going to have time to revamp the entire AI, as a result we are going for small changes that have the biggest pay back. This means that there will be no adaptive AI, or other really cool technology introduced. Sad

We are aiming at to make some NPC more interesting and less predictable, as well as starting to lay the framework for the next release where we can add another group of NPC that wake up and get smarter. These new NPC will be scattered about in some of the existing content and some of the new content.

This will allow us to:
A) judge the impact this has on the mission running community.
B) Reduce the impact if we @$^# it all up.

Part of the changes we are planning is to allow us to break the hard coded connection between the the ship you see, and the AI that that ship uses. This is a important first step and will allow us to implement experimental AI systems for specific ship/groups of ships without affecting the rest of the AI in EvE.

There are several interesting things that we will be doing, and a couple of them you will look back and say, "That came from this thread" :)

Thanks for the great ideas and discussion on AI in general. If you have more, keep posting Very Happy Just because we know what we are doing this release doesn't mean they can't go on the backlog for next release.

CrayC
Gallente
CrayC Inc.
Posted - 2008.12.03 14:53:00 - [315]
 

Edited by: CrayC on 03/12/2008 14:56:10
Originally by: CCP Incognito
Just because we know what we are doing


Must............ resist.......... Shocked


Ah, but seriously, nice to get an update. Personally, I'd settle for NPCs changing targets in a better way than now.

Onyx Asablot
The Vicious Circle
Posted - 2008.12.03 15:04:00 - [316]
 

Thanks for the feedback Incognito, I have been watching this thread closely.

Your plan seems sound tbh, changing all of the existing NPC AI would cause such an epic ****storm as to be unworkable.

Please stay focused on improving the AI aspect of the game, I look forward to seeing new NPC behaviours in a belt near me soon.

J'Mkarr Soban
Posted - 2008.12.03 17:54:00 - [317]
 

Originally by: CCP Incognito
Edited by: CCP Incognito on 03/12/2008 14:51:36
Hi all.

Just a quick note, we finished our first sprint and we have a plan for what we are going to do.

Now for the bad news, we are not going to have time to revamp the entire AI, as a result we are going for small changes that have the biggest pay back. This means that there will be no adaptive AI, or other really cool technology introduced. Sad

We are aiming at to make some NPC more interesting and less predictable, as well as starting to lay the framework for the next release where we can add another group of NPC that wake up and get smarter. These new NPC will be scattered about in some of the existing content and some of the new content.

This will allow us to:
A) judge the impact this has on the mission running community.
B) Reduce the impact if we @$^# it all up.

Part of the changes we are planning is to allow us to break the hard coded connection between the the ship you see, and the AI that that ship uses. This is a important first step and will allow us to implement experimental AI systems for specific ship/groups of ships without affecting the rest of the AI in EvE.

There are several interesting things that we will be doing, and a couple of them you will look back and say, "That came from this thread" :)

Thanks for the great ideas and discussion on AI in general. If you have more, keep posting Very Happy Just because we know what we are doing this release doesn't mean they can't go on the backlog for next release.


Whilst I'm slightly disappointed, I am glad that at least some changes are being made. I have to ask though - have you thought of any longer term plans? Or are you just going to iterate through changes?

Bartholomeus Crane
Gallente
The Crane Family
Posted - 2008.12.03 18:49:00 - [318]
 

Thanks Incognito for coming back on this, and letting us know what you are planning to do.

Obviously I'm somewhat disappointed you decided not to implement all the nifty and cool things you could do, but I understand why you've made that decision. Small steps reduce the risk of a complete meltdown went you go life with them, and through player feedback you'll be better able to assess what more work needs to be done where. Given the amount of work I think you'll need to do for the next expansion, maybe this is the best we could hope for.

I hope this thread, and my contributions to it, have given you some ideas about what is possible and what can be useful for the future, or at least something to think about. I guess I can only hope that after this iteration you'll be able to use the feedback and some of the suggestions here to convince others in CCP what an opportunity AI can be for EVE and maybe make it more of a priority. Personally I'm quite sure that once you decouple the ship, its environment, and the 'AI', you'll find lots of hooks and opportunities to slot in some of the things we've been discussing here.

I'm going to try to keep an eye on this thread for if there are things I can maybe offer some advice on, but feel free to contact me through email in those cases where that is not possible. I'm sure my address will be available through some mailbox by now. I don't mind talking about games and AI, they're both my work and my hobby after all.

Thebro Nobrunder
Schrodinger's Renegades
Posted - 2008.12.03 19:09:00 - [319]
 

I still would very much like to see a python api for coding AI's. Let the community help construct AI's which are fun and interesting. (there are of course many issues here but it would give you some free manpower)

Another idea would be an AI tournament where scripts would be written to controll the ships and teams/individuals would compete against each other.

Might this be a possibility?


Thebro Nobrunder
Schrodinger's Renegades
Posted - 2008.12.03 19:27:00 - [320]
 

Here is another thought related to my previous post.

How about complex npc behaviors?
ie:
* a serpentis npc mining in a belt
* a couple guristas invading lowsec
* simple npc traveling between systems


Titas Agor
TITANS OF PEACE
Posted - 2008.12.05 08:52:00 - [321]
 

bit late in responding to this but figured i would anyway.. and i haven't read the entire post... but NPC AI would be nice if we ever get the ability to call in aid while running missions specificly. I've always thought it would be nice to use LPs or isk to spawn your own fleet of navy ravens or federation megathrons in lvl 4s and 5s. Just be a really kool thing and would get a more sense of the storylines that go on around eve if the missions really emersed you into the universe a bit more and being able to battle it out against NPC fleets with your own little NPC fleet of your own!

Using the right click commands when your in fleet mode, you can select which target to shoot first without the need of voice-chats and you would incorporate this feature to control what your small fleet of caldari ships are shooting at, for example.

CCP Incognito

Posted - 2008.12.05 10:26:00 - [322]
 

Edited by: CCP Incognito on 05/12/2008 12:03:32
Originally by: J'Mkarr Soban
Originally by: CCP Incognito
...

Whilst I'm slightly disappointed, I am glad that at least some changes are being made. I have to ask though - have you thought of any longer term plans? Or are you just going to iterate through changes?



Yes I am working up a multi-release plan that hopefully will take the AI into a much more robust state.

Originally by: Bartholomeus Crane
I hope this thread, and my contributions to it, have given you some ideas about what is possible and what can be useful for the future, or at least something to think about.


Yes you have given me several ideas and possible directions to go. A bunch of us from the AI group attended a seminar over lunch about AI in robotics at Reykjavik University. It was interesting but the level of AI was way beyond what we could do / need. But there was some interesting methods that could be applied in a limited form. For instance the AI the lecturer is working on at Honda's AI lab at MIT uses 3 computers on board the robot and 12 more connected by WiFi to give the level of AI that we saw. I can't see that scaling well to the eve cluster :) Still Asimo robot had a impressive level of human like capability from face/ body part recognition. and was able to play a simple memory game with kids. It also maintains a 3d model in memory of where it is and the objects around it.

I plan to keep positing in this thread going forward. I have gotten permission from Torfi to talk about some of the changes we are making, some I can't as it would be telling about unannounced stuff so I have to keep that quiet. But my plan is to talk about aspects of the AI as they are completed and added to the internal servers, and signed off as working as expected. This will hopefully prevent any Soon(tm)'s :) You will probably also see a Dev blog about the time the AI hits Sisi for public testing.

Like I said before, keep the ideas coming, you might not see them in this release or the next, but if the idea is thought of as good it will go on the backlog.

You might wonder what a backlog is, in this case it refers to how SCRUM software design works. It is a list of Features/ changes that the team feels are needed. They are then estimated and at some sprint in the future they are pulled on to the sprint and executed. For more information search for SCRUM on wikipedia.


Thebro Nobrunder
Schrodinger's Renegades
Posted - 2008.12.05 16:11:00 - [323]
 

Woot!
Someone else who knows what scrum is!!!

Ranger 1
Amarr
Ranger Corp
Posted - 2008.12.05 17:42:00 - [324]
 

Edited by: Ranger 1 on 05/12/2008 17:51:23
I've also always wanted to see something interesting done with convoys.

There once was a time you could eek out some money by attacking the convoys entereing and leaving stations in high sec, now it seems almost no one bothers.

Here would be some idea's to toy with.

Convoys in high sec that are actually worth hunting, but also have escorts for protection. Standings penalties to the corp would be the penalty, of course, as they are now. There would be some "imersion" aspects to be sorted out, as now its a bit odd that you can attack them but in most cases Concord does not immediately blow you up. It would be extremely nice if these convoys actually carried goods that would affect market prices at the buying and the selling end, and if interdiction of these convoys actually could have a benefit if you were trying to influence the price of these goods.

Convoys in low sec, particularly in faction warfare areas. Caldari faction convoys attempting to sneak through Gallante space, as an example. Again, these convoys should be able to defend themselves. Perhaps they could be used as bait. Their destruction could also add to your rank in some small way. By this same token, if this became possible there could also be patrols of hostile faction ships that would randomly roam contested area's.

Central to both of these points would be that the convoys are persistent, and actually use jump gates to get to systems other than their own. This also has the benefit of making EVE feel alive, with commerce and activity happening all around you regardless of what the pod pilots in the area are doing.

* Edit: If you really want to take this concept to extremes you could implement an Interbus sort of system. In otherwords, you could (for a fee) commision an NPC convoy to run your goods from one system to another within a certain range. You would have the option to pay more for more escorts, or skimp and take a larger risk. Goods would be picked up and deposited in your hanger at the other end. If defenses were reasonable for the npc convoys, it could be a cheap (if risky) way to transport goods from one area to another without having to haul it yourself. People don't seem to make much use of the contract options to hire players to move their goods for them (probably out of fear of getting scammed). This might serve as an alternative if it were cheap enough. I can see this opening up some interesting, and cut throat, game play options for industrialists.


Bartholomeus Crane
Gallente
The Crane Family
Posted - 2008.12.05 18:46:00 - [325]
 

I didn't know there was someone at Reykjavik University working on the Asimo. The whole robotics thing is all very Japanese ofcourse, but it's a very interesting project. I've always been very interested in robotics and cybernetics, even though it's not something I do research in.

Now I know that Eyjo has a background in doing research in the US, but I hope CCP will look beyond the US as well. MIT's AI lab is ofcourse well known, but Europe has a strong reputation in AI as well. Now I won't taut my own horn, but check out IRIDIA's website for work that's really interesting but in which I'm not personally involved. Also, the UK has the AI & Games Network, of which I happen to be a member. The Game AI field is really taking of, check out this IEEE article for example. And maybe you'll find this (older) article from the AAAI interesting as well. Both IEEE and AAAI are ofcourse US associations, but check out where a lot of the work is also done. In fact I happen to attend some of their conferences regularly. And there are many other websites, groups, and networks I could mention that also do a lot of interesting work.

In fact, within the current funding climate, it's a well known secret that getting European funding for AI and games is quite easy if you can find a business partner to cooperate with, even if the financial input from the business partner is very small. I have no idea why game developers are so hesitant to pick up on this. Wouldn't it be nice to find EVE and CCP in one of those articles?

But like you pointed out, the problems faced in robotics are very different than the ones you face with AI in games. Problems like spacial orientation, and facial recognition, you probably will not face. The environment in games is pretty well defined, unlike the real world, and that safes a lot of trouble.

Not that this limits things as most of the methods used in robotics for these problems can be used elsewhere as well. For example there is a famous study about using neural networks to do facial recognition, but these same networks have been used in games as well, for example in rtNEAT, which was originally developed for creating complex neural network topologies (of all things).

I'm glad you're allowed to talk a little bit about what is going on behind the scenes. It's personally and professionally quite interesting to see what you come up with and run into.

I am actually aware what scrum is (not an abbreviation for one Smile), I actually taught agile programming techniques last semester. I'm a bit bummed I'm only a chicken and not a pig (how many times do you read that in a post) but as a chicken, I do want to see a sashimi on the burn down chart that says: "most awesome AI in a game ever has now been successfully implemented". Maybe a bit vague, but if you can get a product owner to put that on the product backlog, I'll be a bit happier.

A question. You told us you were decoupling the ship from the AI in some circumstances. I'm reading the AI part here as some prescription or script of how the ships (actor in game AI parlance) should behave. Does that mean that you are going to move the parameters, as you probably have now, out of that prescription? And can I read that as that for certain rats (eventually?) these parameters can be replaced with things like real ships, real fittings, real skill-sheets, or a derivative for these?

CCP Incognito

Posted - 2008.12.05 23:02:00 - [326]
 

Originally by: Bartholomeus Crane

A question. You told us you were decoupling the ship from the AI in some circumstances. I'm reading the AI part here as some prescription or script of how the ships (actor in game AI parlance) should behave. Does that mean that you are going to move the parameters, as you probably have now, out of that prescription? And can I read that as that for certain rats (eventually?) these parameters can be replaced with things like real ships, real fittings, real skill-sheets, or a derivative for these?


Yes right now a ship in EvE has one class that backs it. This class has all the code that handles ships in combat ( I am simplifying things here ) So no mater what ship is in the game, except player ships, it has the same code behind it. The first step is to make it so that i can say this ship has the current Class backing it, but that ship has a new class backing it.

This will allow us to have ships in the game that behave differently, which is something that it can't do now. Well within reason, for instance a ship will check to see if it has a warp scrambler, if it does it uses it based on the rules outlined by the designers. The designer can set allot of values on a ship, speed, orbit distance, damage type, missile type,... you get the idea, but the rules for using that is fixed for every ship.

By making it so that we can assign a new class to a given type of ship, we can make this ship for instance spam local with "oh noz, my shi3ld5 are g0nz" when he starts taking armor damage. This gives character to a ship, while right now if we did that every ship in the game would do it.

So you see how this is an important first step in making the AI able to be different for different situations. It opens the door for use to put anything behind a ship that you fight. Even a trained monkey and we render the players ships as bananas for the monkey :)

I am not going to go into real details on how the data that defines a ship is stored, I would be embarrassed if you found out that it is written on bananas and a hyper active monkey serves them out) ooops :)

The data that governs the current behavior is stored in a generic way so it is able to be used without any modification by the new systems. The AI doesn't use modules per say, but modules are really a user mechanism for adjusting the values that govern a ship. I out fit a armor hardener (because Gallente rules) this is basically saying add a value that changes the armor resist variable. When a module on your ship changes the server goes over all your modules and sums up the changes to your armor resistances. The NPC just short circlet this by making the designer do all the work and record the final number. This saves CPU on the server for the npc's as they don't have to go through all a npc ships modules and figure out what the final number is ( read as less lag ).

Hope that answers the question.

Jarvis Hellstrom
Gallente
The Flying Tigers
United Front Alliance
Posted - 2008.12.05 23:06:00 - [327]
 

No time to read all that - but here are some ideas for smarter NPCs that give exciting fights.

1v1 or 1vN shouldn't much matter.

1) 1v1 the NPC needs a better ship fit. They need to use web, scram, nos, neut, and drones just as players do.

2) 1vN. Less good ship fits but still decent tactics. Ships which can fit drones should always use them. The fits should not be completely fail, just weaker. Less good guns, less expensive modules etc. If a player frig is flying in a mission with five or six enemy frigs they might all have completely basic guns and nothing ML or faction. However they should still fly intelligently.

Add to that:

Missions need to be less 'fixed'. "This room will have 7 of this kind of frig and 2 of that kind of cruiser." NO NEVER. Each room should be random with perhaps a +/- 25% of ship numbers and a similar randomness of fits. Yes, this might mean that some missions the player actually has to GET HELP. Or warp in, take one ship down and then flee to try again. Missions right now are grinding. That needs to stop and variable missions will mean that some will be cakewalks and others near deathtraps. Players SHOULD be able to lose in missions.

Target switching: Players who fire at something and find they cannot break the tank either disengage or switch targets. So should NPCs. If drones are launched there should be a chance that they are immediately targeted based on the results of the NPC's current status of target and own vessel/fleet.

Bringing missions a bit more in line with PvP as regards ship fits would be great too.

Missions should never, ever, be considered 'easy mode' or 'simple grinding where you can never lose'. Make them scary and challenging and PVE will be a LOT more fun as well as less ridiculed (eventually).

How are those for starting ideas?

MotherMoon
Huang Yinglong
Posted - 2008.12.06 01:51:00 - [328]
 

please don't use local for NPC communication anymore :(

I'll think up some good idea for other things you can do, but please don't let it be local or any other kind of chat channel. unless you can chat with them.

OK lets see.. you know in the resent trailers how they pop open a window that shows somethign else? like a news report, or a persons face? WEll what if you got little GFX popups along with voice acting?

You could even have this popup be up near the targeting window area to gives players control over where it would come up.


SO like you get a popup of a pirate or whatever face/faction he is from. You could even use a reduntant system, like you do for the art for the agents. And it could be black and wwhite with radio static sounds or something. and you hear something like...

"hey, come any closer and we'll open fire!"

"that's you had your warning, men! open fire!"

it woldn't have to be for all missions and could have a substitles option where it would show up in local or on yoru screen somewhere. But please, don't put it in the chat box :(

alle OfEVE
Posted - 2008.12.06 02:37:00 - [329]
 

Heya CCP Incognito

I have an idea. A recent project for USA DOD (some coworkers of mine were running it) creating a relational database for research paper publishing that would learn what people read and why and then create connections between the papers being researched to make the user aware of cross-referencing other people were doing and categorize the papers in the database for certain search criteria.

We were working on a way to adapt that to a game we have been kicking around.

What we have so far is a way for the program to watch and see what you do that is effective against the tactics you are using and to then attempt ways to get around it. If you used something that was similar to what it had used in the past it would remember what you had done against it and attempt similar manuevers to what had been effective against it in the past.

Using that idea in EVE if a certain style (such ass RR BS) were being used against NPCs and it was something they had seen before, they would over time learn what tactics were effective and which ones were not. That way they could fight more effectively. Obviously implementing this type of system would be difficult it is something that would be very different in an MMO.

J'Mkarr Soban
Posted - 2008.12.06 19:12:00 - [330]
 

Edited by: J''Mkarr Soban on 06/12/2008 19:17:33
Originally by: CCP Incognito
*snip*


Actually, one point of AI in games is that you minimize class use - one core class learns behaviour, and is able to adapt to whatever it is provided with. Multiple classes could be considered the antithesis of that. A perfect example is NERO, which uses rtNEAT behind it - once the player has trained the behaviour it wants, even though there are 50-odd operatives on the field with slightly vary neural nets behind them, the one with the greatest fitness is saved. Then when you use those operatives in combat - it could be 100, or 100,000 - they all use the same behaviour from the same neural net. The difference is that each of them will, at any point, have different inputs, and it's the different inputs that determine the behaviour shown. As an analogy to Eve, if the rats were taught 'if they have a warp scrambler, then use it', then even within one class that could be done - there is no need for that to be put into separate classes. If you are heading towards the direction of using the same ship types for rats that players have, and randomly generating the modules on that ship and the skills of the artificial pilot behind it (something I was working on as a private project last year - as an aid to truly randomly generated missions and spawns), then you want as general a set of rules as possible, and one class that can be run that takes stock of its surroundings and makes decisions on its actions based on that information. You seem to be heading in completely the opposite direction o.o

EDIT: I understand why though - you're coming at it from a software engineering point of view, which says that if you want different things to happen, you need different classes. It's just unfortunate that that isn't how it needs to work, and may cause problems in the future. It sounds as if you need a proper plan in place from a professional perspective - maybe you should take Bartholomeus Crane up on his offer to talk to you? He seems to know what he's talking about.


Pages: first : previous : ... 7 8 9 10 [11] 12 13 14 15 ... : last (21)

This thread is older than 90 days and has been locked due to inactivity.


 


The new forums are live

Please adjust your bookmarks to https://forums.eveonline.com

These forums are archived and read-only