There have been a couple of discussions about how good/bad the bots are recently.

I know that a lot of people don't care about the bots but they do make the single player mode and for a new player to immediately start playing with real players online without having learnt the weapons and movement would be folly. I really don't see why so many people do that. If the bots were more realistic they could also be used to make empty servers less empty so that there's a point for real players joining them. Once one real player is in a game, there's much more chance of others joining.

The bots actually work quite well in the single player maps as these maps are generally set up well for the bots with a decent set of waypoints. The fact that most 3rd party maps do not have waypoints means that you forget the bots even remotely being able to join in the game properly. If mappers were to put in waypoints, the game becomes much better with bots. I think the waypoint format could actually be improved by having more than one kind of waypoint. Different marker types could be used to define places where a bot should jump or laser jump, or to define the defensive boundaries around a base in CTF. Prioritisation on different point type by situation would make them much better for CTF. Once a bot has a flag, it's number one priority must be to get the flag back to base.

The bots do not react that humanly currently. As an example, if a bot is on it's own and you move yourself to a position some way vertically above a bot, the bot can see you before you have opened fire. This is just not realistic. A real player would not see the player unless they were looking up and on the point of the player above opening fire, they would look up and dodge, perhaps taking some time to locate the player before opening fire. The observancy of a bot needs to take in to account things like level a lot more. Currently the bots 'look' up and down just as well as they do in plane. A good example of this done well is the Reaperbot for Quake which on easy or normal difficulty has reduced ability to look/aim up and down. This is in line with new human players who take time to get used to mouselook.

Another problem with the bots currently is that they are not individuals. Looking at the current bot code in bots.qc, the names of bots are taken from a simple list and are independent of the model or any kind of play characteristics.

What I though of as a first step to improve this area is to define the bots externally. Have, say, a .bot file that specifies the bot. Initially this could just be a [botname].bot file which is blank and would only be used for the bot code to find bot names. By renaming, deleting or creating new .bot files you would adjust the pool of bot names available. This would allow silly tricks like naming bots after contacts in your instant messaging client, or after tracks in the playlist of your media player or after the headlines in an RSS feed. It would also mean that mappers could make map specific bots.

The first bit of info to then use the .bot format for would be to list the player model to load for that bot. This would mean that bots names would become a character rather than just a random name.

The next things to specify would be play characteristics for a bot. Weapon preference, agility, aim, observancy, agression, suicidal tendencies, teamplay etc. All ability would still be scaled by the difficulty setting. This would require massive changes to the bot code so that these values are used. I can envisage things like making bots reply to firendly fire in different ways. Ones who are good at teamplay would immediately message their own team to say 'Same team' whereas really uncooperative ones would return fire in a fit of anger until the firing stops. Agility could also be used to determine whether a bot uses advanced movement like laser jumps or jump boosting to get around the map.

Any comments on this lot?

Might take a look at this for more ideas/guidance

http://leesvoer.com/zooi/projects/gladi ... .shtml.htm

Thats the Quake 2 Gladiator bot plugin. Its quite good as far as I remember, used to use it years ago. Allows you to customise the bots etc.

What a coincidence, I was just looking for a topic to report some bot AI issues and found this newly made one. I agree with all that was said, and I also do care about the bots and find them a very important part of the game. If no one is online on a server you can play on, or you want to train yourself or just play but not with other people, bots are the ones to turn to. Bots are even there for you when you can't win on a server and feel the need to go butcher someone who doesn't fight back too much Also, you might want to note this topic I made a while ago about bot AI issues in mission based gametypes.

On the side of bot editing, I was hoping there might be a bot editor in the game actually. It could be something just like the one in UT classic, where you popup a bot editing window, visit a certain bot then edit that bot's name, model, the two colors, and once that will exist combat / movement / intelligence preferences, just like you edit yourself from Player Setup. Also a bot rooster is another thing I encouraged and wished for, in which you could specify exactly with which bots to play with in a mach and organize teams of bots manually in team games rather then have a random bot join in.

I was wondering though... aren't there any open-source bot AI engines out there? Something like a pre-made bot system just like physics engines can be found and used separately. Something like what Sepelio just posted is the idea.

Anyway like I said in the first line I wanted to report 3 bot issues I found during Deathmatch games and found this topic the best. I know there are many problems with bots currently but these are some more severe issues because they cause bots to get stuck in certain places on the map and not move any more. These are the three things I seen:

1 - There are spots on certain maps where bots cannot grab the paths from, and in such situations the bot will camp somewhere and not know what to do. When I came across such a bot he hardly even looked at me and tried to shoot towards me... I was able to shoot that bot down in a second on the Advanced difficulty without him moving a muscle. That's probably a map issue too but in order to avoid this every bot should always have somewhere to try and go to from every point he may get to on the map.

2 - I noticed an issue in another map where a bot who was normally running around and picking up items suddenly bumped into a wall and continued trying to get through it. Basically, the bot kept sitting in place and running into that wall forever, of course until he seen me and went on to shooting me. To avoid this, I think there should be a check so if a bot tries to follow a certain path but gets stuck into a surface and tries to run through it for more then 1 second, the bot then returns and seeks another path to go through to avoid this situation.

3 - One of the most probably known bot camping issues; If a bot jumps on a jump pad which doesn't have a sideways destination (throw the bot / player on another surface rather then back on the jump pad / straight vertically), the bot will just sit there and jump up and down on that pad forever because he doesn't attempt to move while in the air. Bots here should normally go towards another path node / item while being thrown up by the jump pad, to avoid staying there forever and bouncing up and down.

I put waypoints in my maps. I wish the bots were smarter and more efficent with CPU though. On some of the bigger maps I have to set bot_ai_strategyinterval to 10000.

Also once a bot gets shot off a path, it gets stuck running into a wall forever trying to get back on. It doesn't understand the wall is not moving nor is it.

Hey guys, I'm making a list of tasks for improving the bot AI based on other messages in the forum and my own experience (I have trained my aim a lot using bots). I'm willing to contribute to the project by improving the bots reusing the existing codebase, so this is what I plan to do. However it would be great to discuss about this list so the tasks can be shared among other developers (hey I'm not a developer yet), and maybe updating the development wiki with this.

Weapon combos
It would be nice if bots with a skill lever higher than X can make weapon combos

Better weapon choosing AI
Now the bots are sticking too much to the rocket launcher, even for far distances, or at least for distance where it is obvious that a nex shot will make more damage. This should be improved.

Custom weapon priority
I like this idea a lot, it should allow us to fine tuning the bot behaviour without too much effort. Also this shouldn't be hard to do at code level. Basically we have a custom list of best weapons depending of the distance to the enemy. If this is set then the AI for choosing weapons is disabled. It should depend on a set of cvars like this:

bot_ai_distance_bounds "200 800"
bot_ai_weapon_prority_near "A B C D ... "
bot_ai_weapon_prority_mid "A B C D ... "
bot_ai_weapon_prority_far "A B C D ... "

Skill range concept
Even looking at the code it is not clear to me how to tune the "skill" variable. I'll try to make a better coverage of this variable at code and documentation level. Now if a set a skill level of 99 the bot can be aiming at me without firing at all. Please correct me if I'm making bad assumptions here.

Movement
Don't get stuck
Detect when a bot got stuck, then at least return to the path it was following previously.

Don't stand still
If the bot can't see any waypoint, at least choose any path and keep moving.

Don't stick to other players/bots at close distances

Bunny hoping / strafe jumping support

Laser jumping
Yes I know this is extremely hard to do.

Visibility
Now the bots can see enemies up to 10000 units of distance. This should be a cvar instead so mappers can limit bot visibility for large maps, or as a quick workaround for maps with fog.


Item Pickup
Pickup of dropped weapons
It seems that the bots can't see dropped weapons at all.

Teamworking
When playing in a team, bots should know what weapons/ammo its partners have, then pickup more or less items based on this.

Fun
Basic talking, taunting, laughing at enemy fall damage, weapon juggling

Other
Support for electro secondary + primary attack and stuff like this.

Some days ago I've started learning Quake C and looking at the Nexuiz codebase, now I I have some basic support for weapon combos, but still have a way to go for reaching production quality. Needless to say the bot code is very complex, but I think I can do most of this list on my little free time. I'll keep you posted about any progress on this area.

Please let me know your thoughts

Cheers

Well this is the first patch of this series of improvements: http://pastebin.com/f2210d1d4

It covers the following issues (explained):

Custom weapon priorities for bots:
* Customized bot behaviour
This allow players to customize a little more the bot behaviour. Also this can be used for campaign levels. It adds the following cvars:

Code: Select all

set bot_ai_custom_weapon_priority_distances "200 700" // Define close and far distances in any order, the bots will detect middle distances automatically.
set bot_ai_custom_weapon_priority_far "9 7 13 6 4 3 2 1" // Desired weapons in any order. Up to the max ammount of weapons available in the system
set bot_ai_custom_weapon_priority_mid "14 7 6 4 3 2 1" // same as previous
set bot_ai_custom_weapon_priority_close "7 14 3 2 1" // same as previous


I don't know if it needs a further explanation, if you have any question just shoot. However if this patch is fine I would be glad to document these variables in detail.
If these variables are set, this algorithm takes precedence over the previous one. If no weapon is chosen by this method then the bots will try to use the classic method.

* Trying out new weapons
This way you can get a bot using the seeker, hlac or any other weapon instantly without making any custom code for it.

Weapon combos:
You can enable weapon combos for the existing and custom weapon priority algorithms using the following cvars

Code: Select all

set bot_ai_weapon_combo 1 // Enable weapon combos
set bot_ai_weapon_combo_threshold 0.3 // Try to make a combo N seconds after the last attack


However the weapon combo support for the existing weapon choosing algorithm is a little limited, it only supports weapons with long, known refire times. I would like to have some feedback from the developers about my work before making it better.

Please tell me what you think about this.

I will upload a video showing the bots under these changes very soon

Cheers

These patches look like a nice idea. Hope they can be put in SVN for some broader testing.

How can the bots tell which weapons are comboable? Do they know that they can detonate rockets with nearby hagar or mortar blasts? Will they understand that electro spamming can give massive combos?

See the bots in action !

http://www.youtube.com/watch?v=yuE1DIj74oU

Ed wrote:How can the bots tell which weapons are comboable? Do they know that they can detonate rockets with nearby hagar or mortar blasts? Will they understand that electro spamming can give massive combos?

Hey Ed, any weapon with a refire time greather than bot_ai_weapon_combo_threshold can be used by the bots for doing a combo.

About rocket detonations I think the bots are not aware of which weapon the enemy has. About the electro spamming I'm not sure, I will check the code later.

Nice, they are looking / behaving a lot better now. The only other things I would add to your list is a heightened awareness to what their teammates are doing. They are currently very selfish, run in straight lines towards items and are often hard to predict. In terms of teammates, I wish they would be more passive and give the players more of a chance. Which sort of relates to spawning and taking items, annoying as hell.

Thanks [-z-]

I think the bots can't see weapons at all, they can walk very close to a weapon without ever taking it, so it seems to me more like a waypoints issue.

And yes they behave the same being in a team or not, I've updated the tasks list with your comments.

Waypoints are automatically added for maps by engine for jump pads, armor/health and guns/ammo.

[-z-] wrote:They are currently very selfish, run in straight lines towards items and are often hard to predict.

Just like a lot of real players it has to be said.

Waypoint prioritisation would be what causes not picking up items. They see the item but they also see something else which they go for instead.

What is that else? Any good player picks weapons and armor and bots do the same.

Alien wrote:What is that else? Any good player picks weapons and armor and bots do the same.

Sure, for example in deathmatch mode this makes perfect sense. But for team game modes, like ctf, tdm, etc, the bots still will be picking up items and weapons in front of you, without caring if you just respawned near that nex.
Like any good team player bots should know a bit about which weapons/ammo their partners have, and a bot just can't do a better job compared to a human for most cases, not before this series of improvements

I think this will be my next move after getting my previous code reviewed. LIke Ed said, it seems that other waypoints are getting more relevance than weapon waypoints, it should be fixed somehow.

However now I'm just reading code related to this, I think I can make weapons more desirable for bots, and at the same time put there some creative algorithm querying the weapons, ammo, position, etc of every teammate before consider picking up items.

Thanks for that waypoint tip Alien, I hope you never find the exit of this trap I have confirmed it later by reading code.

Speaking of way points, here's one thing I'd REALLY like to see prioritized >> personal waypoints. I think it would be amazing if bots would respond to "help me" or "danger" etc.

Yes it would be really nice to have, like some days ago Dib suggested that the bots should handle mega[health|armor] timings for deathmatch mode(s). You know things like this are extremely hard to do, because bots are like the Memento guy, not carrying too much information about the map. But well I think this is still possible, it would require a major rewrite in the bots code.

Aside of this I want to say that I'm feeling a little discouraged because I didn't get any developer word/comment about my work yet. Only steel said he will be reviewing the patch, but it seems he is very busy on this days, and I understand that.

Now I'm working on parts of the code that can change a lot depending if my code goes in or not, ie: changes in same functions/methods. And as I do this for fun I really don't want to spend hours merging code.
I might sound a little angry but is not the case at all, just telling what I think

Should I ask for SVN write access? (This would be really cool)

Awesome stuff, keep this up!

It would be great to have bots that are actually _fun_ to play against, not just whatever we have now: walking around randomily, facing the walls, standing there like a stupid in front of you even if you have it on godlike difficulty, the list goes on. My #1 wish is to have strafejumping bots when the difficulty is set a little higher. THAT wolud add some feeling that you are playing against a good player, and not only a stupid bot.

I did a little tinkering with bots.qc. It looks like the bots assign a travel cost to each waypoint based on its distance. This is fine for dm, but other gamemodes (ctf) require going from point a to point b. So, even though in ctf the bots have code for defence, midfielding, intercepting and offense, they will always just sit there and gobble up items (2 feet away) instead of getting the flag (really far away) even though the flag has a higher "pickup value" to an offending bot than anything else. If the cost that the bots assign to distance is reduced, you can actually get the bots to cap on most maps with bot waypoints. Of course, an across the board reduction of travel cost breaks the bots in other ways, so prehaps some game objects (such as flags and fcs) should be flagged as having a reduced travel cost.

Bots would require fuzzy logic, that is their goals would have different "I want that" values based on the current situation and not only get armor, get health, get gun thing.

Alien wrote:Bots would require fuzzy logic, that is their goals would have different "I want that" values based on the current situation and not only get armor, get health, get gun thing.

Yah, the bots don't seem to prioritize item pickups; they even try to take rocket packs when they have 999 ammo. But other stuff, like the flags in ctf, is treated seperately and is assigned a highter "base" priority (to bots that are in the offense role), but the bots calculate the best item to take next just based on its base value and the distance it would have to walk. So, yah, it's currently kindof (but not technically) like a finite state machine.

Well .. This is a message for all the Nexuiz maintainers and developer(s)

I want to contribute to the project with code, I made this patch and I have work in progress for making the bots play better when they are inside a team.

Why? Because I love this game, and enjoy coding. Also it feels good to give something back and getting involved with open source for the first time. I don't use to have a lot of free time, but on this days I have lots of free time, so thats why I've started this.

Also I think it would be nice to reach a point where we can get some fun playing against bots, not the rocket spamming whores we have right now. I think about this many times when I see the Campaign menu as one of the main menus of the game.

But the thing is, I still haven't received any feedback about my little contribution. I said this before, but I want to add some things:

- Its ok if you say something like "your code sucks dude", but please provide me with some advice/directions about how to improve it

- There are some lines in the code that looks really painful to read. Well .. that code *is* messed up already, it is the current weapon selection ai. I only added support for weapon combos to this code, making it just a little messier. I would like to clean that code later, adding the new SVN weapons, and making it understandable for a human being.

- Do we have the structure for supporting those wanting to become developers or make sporadic contributions ? If not I'm afraid that Nexuiz is just dying.

Regards

Well, who SHOULD be looking at it? I'm out of this, mostly, and simply have found no time to check out this huge patch. That leaves LordHavoc, and I haven't seen him for two weeks now. I am even less inclined to look at your patch after the master of arrogance and manipulativeness tried to push the issue on me.

Remember that his pressuring me to implement stuff I didn't even like, and then complaining that I did it wrong, was one of the main reasons for me leaving, as I refuse to work for an OPEN SOURCE PROJECT as a HOBBY when under PRESSURE by arrogant manipulative people.

And no, there is no such "structure", and never was, as such structure requires people to support it. The team is simply too small. It was too much for me to do everything on my own, so I had no choice but to leave.

Just wait, there's nothing else that you can do.

Has anyone of the "users" tried the patch?

Also add to this that the bot code is only really understood by LordHavoc. It is nice that you did get so far, but well, even I don't understand what that code is doing.

Maybe send your code to LordHavoc via e-mail, his address is to be found on http://icculus.org/twilight/darkplaces/email.html

Well, who SHOULD be looking at it? I'm out of this, mostly, and simply have found no time to check out this huge patch. That leaves LordHavoc, and I haven't seen him for two weeks now. I am even less inclined to look at your patch after the master of arrogance and manipulativeness tried to push the issue on me.

Gah .. I know about who you are talking about ... What can I say, you should know I was a little disgusted with him. I made clear my position in a post, like a lot of other people in that odd thread. I think the message was clear. But now I'm just waiting for a change, I think that most people can change, and not losing faith is the best way I know for making that change happen. Maybe I'm just too idealistic, maybe not. But this time I just don't know about what you are talking, I'm afraid I'm not part of that. All I can say is that he just asked me in good grade to talk with you about getting my code reviewed. Anyway I think this is between me and the developers/maintainers, I really don't want/expect to involve anyone else if it will not be for constructive feedback and good vibes.

Remember that his pressuring me to implement stuff I didn't even like, and then complaining that I did it wrong, was one of the main reasons for me leaving, as I refuse to work for an OPEN SOURCE PROJECT as a HOBBY when under PRESSURE by arrogant manipulative people.

Well I work mostly as DBA, but I'm a part-time developer too with many years of experience, and I think I understand how you feel. Since I've started lurking on the forum I saw a great deal of bad vibes against the developers, maybe the same kind of complains you get when you are maintaining any other system. You have there both extremes: retarded people and good people, you can choose to who do you want to listen, talk, interact. I'm afraid you took too much of the bad energy that is *always* (imho) floating around. My point is, you enjoy programming and enjoy playing this game, also I can bet that you have very good friends here; Can't you just ignore all the morons and the noise and just get what makes you feel good or what you can interact in a positive, constructive way?

And no, there is no such "structure", and never was, as such structure requires people to support it. The team is simply too small. It was too much for me to do everything on my own, so I had no choice but to leave.

Yes, I know. What you made and your level of understanding of Nexuiz really amazes me. And you resisted too long. I don't know what to think, maybe this is a side effect of not having a strong leadership, or maybe we should be a little more organized. It's clear that all the project direction, maintenance, bug fixing, presence on the forums just can't /shouldn't be handled by just one person.
Also I know that when someone is involved like you in a project it becomes really hard to leave from one day to another. I prefer to think that you want a change, instead of stepping back.

Just wait, there's nothing else that you can do.

I think I waited enough, if I don't get a review soon I will ask for SVN access, I feel strong enough for requesting it. If not ... well I think it will the right time for giving up.

Has anyone of the "users" tried the patch?

I know some friends that can help me with this. I'll try this way too. Anyway I'm confident enough about the quality of this patch, I really don't want to waste anyone's time.

Also add to this that the bot code is only really understood by LordHavoc. It is nice that you did get so far, but well, even I don't understand what that code is doing.

Yes, its a pain in the *ss. I had a hard time reading it before to start writing.

Maybe send your code to LordHavoc via e-mail, his address is to be found on http://icculus.org/twilight/darkplaces/email.html

Thank you very much. I'll try this way too.

In the worst case you could always provide a separate patched progs.dat to download.

I'd be more than happy to test your code, too!

I'll try to test too, but I have a few questions too:
I'm using SVN most of the time, so I'm building the game from the SV source. If I aply your patch, how'd it interract with the next version of SVn update? Could it fail? (I'm not afraid of it, but it's a hell lot of time to re-check 2 gigs, and I already have two different SVN checkouts for Nexuiz, one for playing, and one for mapping. Another 2 gigs... erm... would be rather "useless".)

C.Brutail wrote:I'll try to test too, but I have a few questions too:
I'm using SVN most of the time, so I'm building the game from the SV source. If I aply your patch, how'd it interract with the next version of SVn update? Could it fail? (I'm not afraid of it, but it's a hell lot of time to re-check 2 gigs, and I already have two different SVN checkouts for Nexuiz, one for playing, and one for mapping. Another 2 gigs... erm... would be rather "useless".)

SVN is smart. But if you get a conflict in the future you could just delete havocbot.qc and bots.qc then update again. I'll check this patch out also.

Use this: http://svn.collab.net/repos/svn/trunk/c ... /svn-clean
It will clean the source without the need to redownload everything.

Actually, I looked at the patch now, and found it doesn't even do much at all. Maybe it would even be wise to use it, and manually calculate the weapon priorities for the distances based on the weapon parameters (e.g. by a script, or an excel worksheet).

That would make the bots a bit less smart, but provide an easy way to add new weapons, and also it would allow different bot "personalities" in the future (like in UT or Q3A, that is, making one bot a "rocket whore", another a "laser camper", the next one an "annoying sniper", etc.).