@dark1202 said in Bot AI issues:
Metal Gear Solid 5 had fantastic AI
F.E.A.R. AI was brilliant...
The Last of Us
Deus-Ex
I made a list of posts I wrote on the subject and pasted them here for easy navigation. If you read these you may want to reevaluate the list you proposed, reply me back after if you want, so I avoid repeating the same information. Its a mixed text, but it is relevant and address molotov throwing, restricted areas, why A.I. is not better elsewhere, difference between single and coop etc.
HERE WE GO:
POST 1:
Yeah, I think AI or the bots algorithms could be improved in all games always, as I have never met good AI in any video game ever. (I have played video games since Wolf3d came out).
Certain games like half-life 1 had A.I that seemed to react in a smart way, but those situations were actually heavily scripted events from what I've read meaning they would perform certain tasks when you touch a point on the ground. The AI from NWI on the other hand have some dynamic behaviour (ruleset they can follow in different situations) so even if they seems dumber than for example in half- life 1 or 2, they are actually better in the games from NWI because you want the dynamic behavior to keep the game fun in repeated playthroughs.
My favourite AI is the librarians from the Metro series, meeting them was the weirdest experience as they felt so unpredictable in the theme of the situation. In shooters I really like the AI from the Misery mod for stalker call of pripryat, though that ai would kill you from 600 meters away with a single headshot, so it would not be fun without a save/load function.
Imrovements to coop from my point of view in Insurgency Sandstorm:
-There could be incorporated more bots. A lot more (though more bots = more strain on your system, so optimization first right)
-Different difficulty settings, and a lot higher difficulty settings as options (a setting higher than the old brutal named suicide or slaughter or along those lines would be cool, and have dedicated servers locked on top difficulty for the players yawning through the brutal setting)
Rule changes to the mode: If you play a game of coop as the state is now and just run from objective to objective only killing direct threats but trying to avoid most, you will end up on top of the scoreboard with the Ranking "Professional" "secured the most objectives, x objectives secured" or something alone those lines. Even if you are the reason the rest of the players must rush after you and adapt to the same playstyle to traverse the map, no tactics, no safety planning, no consideration of anythig you would brand as professional in reality. They could add team alive bonus to try promoting teamwork, lower the points considerably if you cap an objective alone when x teammates are alive, I don't know, and these changes would go against competitive, as in competitive, initiative and fast reflexes is key skills to get ahead of the opposing team. In light of competitive training it is therefore straight on point already^^ I guess they could have to different categories of coop, 1 category being competitive oriented, like promoting behavior that will help in PVP, and another coop mode that is focused on team strategy and promoting mechanics that ensures people play in a way that helps the whole team.
It is still great fun though, and I guess people rush because it works the best within the current ruleset + makes it faster to level up to get those nice clothing. I don't know but the points I save up gets erased all the time, so I can not save up 600 game currency to get shades to my soldier even if I am lvl 17. But devs are aware of these things I am sure, and with time iron those bugs out.
What is it about the current state of the AI that you find frustrating, could you provide an example? (Like do they kill too fast or too slow, have a moving pattern you mean is not optimal, stay in unnatural places when inactive/camping/guarding etc)
POST 2:
Yes, I believe they got a lot right with that AI in that type of game, some of the devs on metro are the same that worked on Stalker series, even one of the devs working on Escape from Tarkov now. Too bad the company was closed and the X-ray engine not updated anymore (the program the stalker games were built on, Metro used 4a engine and the new Metro will use a version of 4a from what I've heard) The AI there were also very nice in my opinion, but they had to make a trade off at least in Stalker. The A.I. had different states, like alerted or just patrolling, and I think the trade off was made in that when the bots were alerted they kind of knew where you was at all times, making it more fun to fight them if you wanted a challenge, but less different ways to approach after they got alerted, as they would all shot your direction with aggression at max.
The comparison can be difficult to make between games like Metro and Insurgency Sandstorm because of how different the size and type of games they are (single- vs multiplayer), and the scripting I mentioned earlier. Have a look at the A.I in Escape from Tarkov (a more similar game in comparison with Ins sandstorm), it does not seem more impressive? Though I don't own the game, only seen footage, the bots seem to be quite easy to take out and not being routed to play smarter moves. I guess that in multiplayer focused games there is not any other A.I. that actually plays a lot better than in Sandstom, not from what I have seen anyway.
Reaction time on bots is just one of the numbers they can adjust to change difficulty, and I have yet to see an example from a similar game with better routing for the bots - They could probably make the bots stay closer to cover and hide more in bushes, don't walk in groups, make them lean and such, but I don't know what the cost would be on the performance. All this kind of stuff costs more strain on the system when the operations get more complex, and they might have decided on current solution to have a balance between quality of bots and quality of gameplay. I have not seen any articles on the subject.
POST 3:
Not too difficult in singleplayer games, was done in the F.E.A.R game back in 2005, you can read about how they did it in this link named Three States and a Plan: The A.I. of F.E.A.R.
http://alumni.media.mit.edu/~jorkin/gdc2006_orkin_jeff_fear.pdf
In multiplayer it is apparently a lot more difficult to make it work well, as @Zafer states, it should already be implemented if it is present in the change log, but it will be difficult to notice when a lot of players attract the attention of the bots.
Also this article: (D)evolution Of AI In Video Games was interesting in regards to why bots in video games don't evolve that much.
https://www.pointsprizes.com/blog/174/devolution-of-ai-in-video-games
POST 4:
Hehe, the conga lines takes the challenge away, or lack of ability to spread out and act more like in guerilla warfare - This would be the most important behaviour to improve in my opinion if any improvements should be done.
The modded servers from Insurgency 2014, especially the pve BEF server gave me just the right challenge, and hopefully we will see similar mods in this game.
I do like that they often use melee attacks when up close even in smoke, which is a change in the right direction, though a small change. Hopefully they will leave out some of the noises/shouts the bots currently makes on higher difficulty in the finished version as well.
The molotov throwed directly at players from 100 meters could be a good thing. It would probably be easy to implement a behaviour where the bots randomly hit and miss with the throw. They always hit directly from such distance? The farther away a frag/molly is thrown from, the easier it is to predict and avoid - What would you see as an improvement in that regard?
Look at other pve multiplayer games, they often add difficulty by making enemies bullet sponges, having higher damage output or increase their number, but still acting like jackasses:)
I have yet to see anyone give examples of bots actually performing good in any multiplayer game though, "Poor AI" compared to a specific game which does this successfully would be interesting, as it would be cool to analyze that game and see what they did right, to get a feel of the actual situation regarding bots in multiplayer games in general. It would also create a platform to discuss what other developers in such cases have been able to achieve in the bot development department when also creating a multiplayer game.
The developers gotta balance their time to create a good game, and how much resources would it be logical to invest in pve when the main selling point is pvp, marketing wise.
After all, you get to play against real people and get all the challenge you could possibly need, and personally I find it cool that they even have bots in a game like this so I can enjoy the more relaxed experience of pve, practice reflexes and test weapons without competing.
From the last developer Q&A on youtube, it is mentioned only one person in NWI deals with A.I - which might be good enough regarding where the focus in a pvp game should be, though I respect that a lot of people see it differently.
As we have discussed earlier, it is difficult to compare a multiplayer game to a singleplayer game regarding bots, so that games like Metro and S.T.A.L.K.E.R - having both fun engagements with bots - Falls into a different category.
From the article I linked to about the (d)evolution of A.I. in video games it is also a point in my opinion that most people don't actually understands all the effort it is to make solid A.I. in video games (not referring to happybub or anyone in particular, just gamers in general according to the article).
As you probably know already (just for those interested) there is not much advanced A.I. in video games as the bots don't learn anything as far as I know and improve their own coding - That would probably be a bad idea and would make the bots exploit bugs and glitches as they just use the most optimal strategy disregarding game rules not clearly defined -Everyone would hate them and never win a game=D
Scripted behaviour, preset paths and "AI nodes" are all a lot easier to implement in singleplayer games where you can have triggers to activate behaviour because of player position or pushing a button, etc. More players change all this significantly.
I definitely agree with you though, just saying there is a lot of elements to consider both regarding resources spent on development, cost to performance on the system and actual reasons for a developer to invest in this regarding cost/gain of doing so.
I'd rather have the developers making the game the best pvp experience possible, but would of course welcome an improvement of the bots.
Specific suggestions to small changes the bots could do would from my guessing be the easiest way of getting the developers to actually change bot behaviour, as I believe the developers are aware of the current situation but need to prioritize their resource spending carefully.
So from my point of view this thread could be vitalized by:
Suggestions to small improvements in the bots in Sandstorm (easy to implement).
Examples of other games with multiple players that does the job better and being specific in what situation.
POST ABOUT DYNAMIC SPAWNS (kinda unrelated to our current topic).:
I have played a lot of coop as well in Insurgency 2014, but after a while I started using coop as a warm up game before entering PVP to be at the top of my game. I agree that coop could be tweaked - Most of the time I played custom coop servers (BEF was my favourite) because vanilla coop on brutal difficulty felt too easy and one reason for that is when you join a coop there is people only playing coop in the game - They use incendiary and frags + C4 on the poor stupid bots, taking away the gun play from the game and making vanilla servers mostly useless to join because of the low difficulty.
Now if you remove the restricted areas what will happen?
1- Players who play coop 24/7 will enter the spawn of the bots and putting down C4 removing the gun play from the game
2- The same players will use incendiary in choke points close to the bot spawn removing the gun play from the game.
3 - New players and/or players who don't like pvp because they think it is so difficult, prefer games that are easier so they can manage the situation. Restricted areas are probably good for inexperienced players as it makes the gameplay more linear, and should be kept at lower difficulties.
Maybe you are right that the restricted area in coop should be removed as the aspect of flanking and having free movement can be more fun, but it should not be done without thinking about the problems with the points I mentioned.
Possible soulutions:
• Make bot spawns dynamic: You never know where they will be, thus keeping the game fresh and the cheese -gameplay (abuse of mechanics to make game too easy) to a minimum.
• Remove most of the supply points on higher difficulty so players are forced to spend less points on explosives and other cheap options against bots ( I mean bots in every PC game are pretty bad because they are very easy to understand/recognize their pattern++) therefore explosives and other secondary items takes away the challenge when playing against bots.