zippyzee said:In general terms, if an NPC is hungry, it should try to find some food. It might look in the normal area where it has a food supply. If there is none there, it might travel around a bit looking for food. If it can't find any, it might look for an animal to kill that would provide the food, and attack it. If we want to get really crazy, if a set of NPCs can't find food, they might migrate to another location, and perhaps displace the population there through a protracted battle. In a "learning" system the prey might realize that standing in that area is not safe, and migrate away as well.
My son says if the NPCs have a cook, and a rogue travels into the dungeon and kills the cook, and the NPCs can't eat, they may start to venture out of the cave looking for food. Or, players might place some food near the entrance of a cave, and that food source would be closer than their typical source, and they might venture out to eat. These are the kinds of gameplay elements that I'd like to see.
Now, I have no idea if we'll make hungry NPCs. If we did, we could have a hunger scale of 1-10 that is their disposition for needing to eat. At any given time they might have other more pressing priorities to attend to. But occasionally the "Hunger" priority will trigger and they will respond. And as NPCs spawn, they might have a range of say 4-6 for that particular behavior, and might act slightly differently than the last time they spawned.
Substitute any particular disposition for "Hungry" and imagine the same type of process, and you get the general idea of how I'd like AI to work. I want behavior to be predictable to some degree, in order to be a better player through experience, but for surprising behaviors and reactions to occasionally take place, and for players to experiment with ways to alter behaviors in their favor.
I can not tell you how much I am excited to see something like this in action.
zippyzee said:---snip---
I will get into more details as we go along, but wanted to make sure you knew this was getting serious attention. My personal belief is that AI has been really ignored in most MMORPGs and if done correctly can really make a game stand out.
Sounds bloody awesome! Looking foward to learn more about it mate!
No cook = they eat man flesh... their aggro goes up.
meats back on the menu...
Deadshade said:At best there is a script for a raid end boss fight but that has nothing to do with AI, it is merely a prescribed scenario which is always the same (f.ex at x% of HP the boss calls adds) .
I have always wondered why it is so and suspect that it is simply because the MMO developpers are not familiar with AI research and AI software that already exists so that they are stuck in reproducing what was invented 20 years ago so that no progress is done .
With regards to raid bosses, there are several things you have to consider with regards to AI. If you make the mob too "clever" it becomes all but impossible to form any kind of strategy. In an open world where every pull is at a premium (because a queue is forming behind you), that would be a disaster. It would also turn raid bosses into a game of luck. Ideally, a mix of the two should be considered - scripting, but with multiple possible "branches" of behavior the mob can take based on certain circumstances. That way, you could have several permutations of successful strategies, and it would force people to really understand the encounter instead of trying to zerg it. You also have to consider the computational power required to account for so many variables. In a single player game, this is doable, but on an MMO with a mob potentially having to react to hundreds of player inputs, the complexity raises exponentially and with it the required processing power.
Rest assured that I am sincere about doing this in a way that I think we all will like. I don't want it to be a simulation, but a game, but I'd also love to sit back over the years and watch players find new ways to approach problems and new strategies that might have been completely unexpected. I like the term "emergent gameplay" and that would be a goal. I'd also like to have the ability to "turn on" or add something to NPC behaviors from time to time that will shake up what they do. This is all taking place right now as I work on making the system as open-ended as possible.
For background I have a Master's degree in Computer Science from Indiana University, which I completed in 2012. I focused on Artificial Intelligence courses, some of which had serious gameplay AI to design and develop, often pitting my creations against classmates in daylong tournaments. I have worked extensively in the computer vision world, developing object recognition AI that has been patented and is used commercially. I'm a problem-solver and also pretty creative.
I'm also a lifelong gamer, a huge Everquest player (and Vanguard, and EQ2, and just about anything) and I want to see most of the things you want to see. There will be some obstacles and limitations, some from computing resources (but not a ton) and some from overall design. In the end this dev team is all about fun, interesting gameplay with the same social emphasis that many of us remember from back in the day when we had a second life inside an MMORPG.
There will be scripted "boss battles", and I agree that they don't need to have a bullet list of things to do in order. They might include a few decision points, or even some randomness to make each encounter a little different.
One other aspect that I did not mention is that by assigning certain dispositions to NPCs, or even a list of priorities in their lives, we don't necessarily have to hide those from the players. Some classes might be able to either learn or innately see what those NPCs are thinking about, or planning to do. So depending on the makeup of a group, you might have a warrior that can detect what the NPC's next attack priority will be, or a scout that can see that an NPC is an alarmist and will run to get help if someone gets within range. There is the possibility of studying to learn dispositions and behaviors as well. In this way NPC AI can have a second layer that influences gameplay in a different way. That is another goal of mine.
So, zippyzee, your name wouldn't happen be Robert Ford, or Arnold would it? :p
I see where this is going...
Hey zippyzee It's cool reading your goals I truly hope it works out. It would definately be fun if NPCs, friend and foe, were somewhat unpredictable (so long as it makes logical sense atleast haha).I for one am completely behind actual AI as I learned about...from this thread. Hurray forum education!
zippyzee said:One other aspect that I did not mention is that by assigning certain dispositions to NPCs, or even a list of priorities in their lives, we don't necessarily have to hide those from the players. Some classes might be able to either learn or innately see what those NPCs are thinking about, or planning to do. So depending on the makeup of a group, you might have a warrior that can detect what the NPC's next attack priority will be, or a scout that can see that an NPC is an alarmist and will run to get help if someone gets within range. There is the possibility of studying to learn dispositions and behaviors as well. In this way NPC AI can have a second layer that influences gameplay in a different way. That is another goal of mine.
*Cough* Lorekeeper *Cough Cough*
https://www.pantheonmmo.com/content/forums/topic/4252/random-thought-lorekeeper-skill
Great posts Zip. You're making it this much easier to be a fan.
Manouk said:zippyzee said:My son says if the NPCs have a cook, and a rogue travels into the dungeon and kills the cook,
...... WHO might one be reffering to when speaking of this rogue, I mean anyone specifically....
Lol, he couldn't possibly mean me, I am the best behaved Rogue in all of Terminus! /shifty eyes ;)
zippyzee said:
I will say the design takes into account the possibility of "learning" over time through weighing positive/negative outcomes but that may be overkill for an MMORPG and it would be fairly easy to make the AI just more difficult over time rather than more interesting, at some computing expense as well that might not be necessary.
Thanks for these infos Zippyzee, it gave me reason to hope that my wishes are not completely irrealistic .
Yes I understand what you mean . For playability reasons it is not possible to push the AI too far .
An example .
Take the Sarnaks in Chardok and give them JUST the intelligence of an ant .
They would look for food, breed and defend their home . They would "learn" to distinguish friends from foes and apply adequate behaviours to help the former and kill the latter .
But even such a primitive AI would be enough to immediately destroy every and any group which would try to penetrate the place to get the Prince's head .
So an MMORPG's AI needs to be dumber than ants what puts the priority more on the diversity and a moderate unpredictability of behaviours than on truly learning and adaptative behaviours . It would be an AI "light" but even that would be a huge progress compared to what has been done in every MMORPG during the last 20 years .
Just out of curiosity Zippyzee, speaking of hunger and AI "light", did you read my proposal here : https://www.pantheonmmo.com/content/forums/topic/1663/animal-ai-specifications ?
Yes, I've read every post I can find on here about AI. Most are very well thought out and I can't disagree with them.
The funny thing is that MMORPG AI is not that difficult to design. We can all come up with a number of good ideas really quickly, and we can all agree that there is a limit to how much we really need in order for it to be enjoyable rather than stifling. Really the biggest issue is finding the best solution from a programming standpoint in Unity Engine (and C#) that allows flexibility and scalability, and offers designers easy-to-use tools to assign behaviors to NPCs. There's a number of ways to skin that cat and that's exactly where I am right now--putting some things in and seeing what I like/don't like about them.
AI in an MMORPG is as much about the illusion of intelligence than actual intelligence. We are all creating our own stories when we play, and we will automatically assign 'thought' to NPC behaviors, especially when they aren't expected. I think concentrating on basic needs like you suggested above is a really good start. Companionship, safety, and those other life necessities alone lead to interesting actions over time.
I don't understand why this has been seemingly ignored in past MMORPGs. It doesn't take a lot of computation time to do just a few things that would make NPCs seem at least basically intelligent.
zippyzee said:...
The funny thing is that MMORPG AI is not that difficult to design.
...
I don't understand why this has been seemingly ignored in past MMORPGs. It doesn't take a lot of computation time to do just a few things that would make NPCs seem at least basically intelligent.
Your posts have really been awesome to read! To be honest I'm surprised. I have no experience in the field, but I assumed from past games that AI of this sort would actually be quite difficult to implement. Some single player games seem highly geared towards AI (rimworld comes to mind), but I figured that it only really worked if the AI drove the game design, rather than having AI added in to improve the game. It would be super cool at some point for you to do a newsletter type thing describing how the intelligence is implemented in layman's terms and the possibilities/limitations with it.
What would be killer is if you were in an open area, be it a great room in a dungeon or outside area, and tried to pull a mob, all of their friends in VISIBLE range would aggro (think cone-based line of sight)... not just close proximity, social aggro. This would be kept within reason, of course. In every single MMO you can split and break packs, I get it, but across the room no more than 50 feet away another pack just waits their turn. Surely they can see you cracking their buddies' skulls! This would make a CC player or puller really think about how to defuse and break down a whole area with calming spells/sneak-pulling so that the group doesn't get trained. This could be challenging to code at times if you wanted to have fog, rain, etc. affect the visibility.
can we have NPC's react to the PC if the PC Insults them.
1. start attacking the PC.
2. the NPC offers to Duel the PC if declined walks away laughing and insults them back.
3. starts crying.
4. runs away.
5. calls for fellow villagers to intimidate the PC.
6. locks the PC in Stocks for 5 to 10 mins or some variation.
7. the NPC will not talk to you again untill you help them with a simple task IE: Time waisting for insulting them.
i like No 2 the best.
Kilaen said:What would be killer is if you were in an open area, be it a great room in a dungeon or outside area, and tried to pull a mob, all of their friends in VISIBLE range would aggro (think cone-based line of sight)... not just close proximity, social aggro. This would be kept within reason, of course. In every single MMO you can split and break packs, I get it, but across the room no more than 50 feet away another pack just waits their turn. Surely they can see you cracking their buddies' skulls! This would make a CC player or puller really think about how to defuse and break down a whole area with calming spells/sneak-pulling so that the group doesn't get trained. This could be challenging to code at times if you wanted to have fog, rain, etc. affect the visibility.
I agree. I'm often left scratching my head in most MMO's where a simple radius determines whether or not you get an add. Like, this guy's friend is basically just watching as I beat the crap about him, but if I were 3 feet closer, then he'd help. It's so unrealistic that even the sound of a battle wouldn't trigger the rest of the camp to at least come investigate. And if they (visually) find what's going on, then they may aggro. And pretty much any intelligent creature would call for help if their camp is within earshot. This would definitely make it challenging (but not impossible) and therefore require a lot of really fun strategy. But I also think this is where perfect realism is probably not really condusive to MMO's. If one group gets the attention of an entire camp every time, there's not gonna be enough mobs to share between the playerbase, unless the world is just littered with camps. Neither situation is favorable. Really curious to see how VR handles this.
That said, I'm really starting to like this Zip guy! His responses are very interesting, informative, and keep the discussion focused. This is what we need.
Really great read Zippyzee . It is rare to get this kind of interesting informations (without betraying trade secrets of course :)) directly from the developpers .
If only that, it is what sets VR apart from other companies who shout perhaps louder that they put priority on communication with their customer basis but who do it much less if at all .
Be sure that it is very appreciated !
From your last reply I guess that you are working with that : http://www.opsive.com/assets/BehaviorDesigner/ .
Right ? Or have you some more customized software ?
Deadshade said:Really great read Zippyzee . It is rare to get this kind of interesting informations (without betraying trade secrets of course :)) directly from the developpers .
If only that, it is what sets VR apart from other companies who shout perhaps louder that they put priority on communication with their customer basis but who do it much less if at all .
Be sure that it is very appreciated !
From your last reply I guess that you are working with that : http://www.opsive.com/assets/BehaviorDesigner/ .
Right ? Or have you some more customized software ?
No. I'm custom writing all of our AI in C# within Unity to do it in the way that I want, which gives the most control. If we run into issues there's always the possibility of integrating 3rd party utilities.