[b]They have strafe jumping, item timing, awareness of stack and opponent stack, better map navigation, weapon selection is really good (sometimes too good with Spite and not the other bots), awareness of opponent's weapon, it rocket jumps for items, RJ for items, RJ for position. It has awareness of low health and tries to avoid opponent. It really is basically CPMA Bots but works in VQ3 or CPM mode within the mod. There is just not a Green Armor, so that kind of sucks.

Because of it's fluidity in decision making, it is very close to playing a real person but some maps make it feel more like that because the AI sometimes has less path obstruction.

so the bots require the custom maps with mod enabled to work properly, with hardcoded waypoints and such?

I guess there's no way to integrate the bots in QL

It uses the standard Q3 AAS system, so that's not an issue.

More of an issue.... I've not really been paying attention, but the QL source code wasn't ever released, was it? So QL + mods still isn't a thing, right?

You might be right, but people do seem to make mods still on the QL Workshop somehow. Usually retextures and things of that sort. You're probably right about the mod support though. This is why OpenArena and Quake 3 Arena are superior to QL, these games released source code and people are able to mod. Unfortunately, there are enough engine limitations that you can't play some of these old mods in EDawn or CPMA or the AI can't be completely reprogrammed.

I've thought of trying my hand at a 'Duel Arena' for UE5 and have a main theme of offline play. QL\Q3A and UT maps would be available through something like MODDB or Epic Store but separate from the project to avoid obvious copyright. Bots in duel would operate like the system of Alien Isolation, using a data collection engine within the game that pays attention to stacks, positions and weapons. It would collect and relay the CPU opponent that data relevant in order to simulate a thinking player who has map awareness and positioning, whether it might choose to play sneaky or to play full on aggressive depends on that data.

I think probably with the knowledge we all have today of duel, my objective would be to program the AI to think more about positions, which in the case of positions they would be like preplaced zones around map geometry. And to have nodes or paths that link them to one another and different items. But the purpose of this would be to allow the AI more flexibility on deciding where to go, where not to go based on conditions relevant to what players usually think about. Virtually every time the CPU goes into weak positions\zones I would try and program it to move fast and only grab what is directly in it's path or what has the highest value and split. Say a player walks into a zone with higher value within a radius or a section of the map, the zone that the player trips into gives the AI knowledge that the CPU's position is weak and it needs to prioritize reaching a better place. This gives the human player a chance to make cutoffs and traps like they normally would, but not be able to always find free damage. Usually poor rated players just walk around in the open grabbing items and looking to fight. Highground > Items usually in duel, especially out of control, and sometimes in control it is better to land damage than take items based on how much stack was already in store.

Most likely, putting some of the AI functionality into the hands of the mapper would provide larger options for most of the challenges programmers have had in Q3A. I think there would be ways today to actually make computer AI operate stronger in Arena FPS, just from playing SpiterBot and CPMA Bots I get the sense that if not for engine limitations in the past it would have been achievable beyond question, just it really depends mainly on what the programmer knows about duel or if someone is providing feedback on it.


If I were a decent enough programmer I would have already given it a go. Obviously there is the option of just rewritting Q3A code into UE5 for specific things like weapons and movement. But that isn't really my forte either. Still, I think it isn't out of question whether it is achievable to create a realistic AI opponent to play duel against, it is just a matter of how hard someone is willing to work to achieve it, skill, time and resources. You can look to CPMA bots for instance to see how close that actually is, the only downside is how they move around and attack sometimes.

Personally, if I were to have a go at making an advanced bot now, I quite like the idea of trying to apply modern day AI techniques to the problem, rather than going the spitterbot route of pre-programmed behaviours. I think you could probably do something interesting by using demos as a training set to teach it very advanced play. I'm not likely to find any time to play with this stuff myself though.

Deepmind did something interesting with CTF: https://www.deepmind.com/blog/capture-the-fla...ive-agents (for the purposes of making a good Q3 bot you don't need to go as far as they do and require the inputs are solely the pixels on screen + win/loss result...)

No they work without the custom maps, that was just advertising that they do play with QL ported maps. It might be possible to port this mod to QL, but some of the functionality will be messed up. For instance the custom hud and menu settings won't work because QL has a different menu. However, if Arqon is still around in the community and still has the source code to SpiterBot or still keeps in touch with the original authors, maybe it would be a good idea to talk to him and see if this can get ported or at least acquire the source files. That would be awesome if that actually could happen.

The original author is qrealka - I've not kept in touch with him, but I guess he's more likely to still be around than arQon.

Fun fact: qrealka started off doing the Q4Max bots but he tired of Q4 before they were ready, so I finished them off. The Q4Max bots did know how to dodge projectiles like Spitter did though.

I remember playing this mod in somewhere around 2006-2007. In the German Quake Forum, there was a challenge: Who could beat spite(100) in a 15 minute duel, on pro-tourney 4. (His best map)

Everyone failed :D

https://esreality.com/post/536273/re-spiterbot-beaten/

Quite a few people on this site were able to, but it is difficult.

Alright, so the version in your SpiterBot competition is supposed to be RC9 which I am pretty sure is applied to the March 6th 2003 update, this is it right here. I played it on T4 and I lost the match the first time, and won the second time :D. It was pretty hard. Okay so I have the RC9 Update applied to March 6th version of SpiterBot v.50, this should be the base version needed for RC9 to work properly. Enjoy! Also please let me know if this is the version that you played on T4 because I am almost certain it probably is. Look at the dodges this thing does lol.


https://fastupload.io/HUV3gypvdaxp9Op/file

Here is my score against the RC9 Version on Q3Tourney4. It was pretty challenging.

https://imgur.com/yDGtb6a

You can also try playing March 6th Update without VM01.pk3, this might actually be the toughest version. I think people misunderstood that RC9 was an irrelevant patch. So far, there is a significant difficulty difference between these. To do this, go into the folder you installed RC9 with and remove VM01.pk3 but leave VM.pk3 alone.

Note that you need to play it on pro-q3tourney4. q3tourney4 is a different level that the pro version is based off.

Unfortunately I won't have the time to set it up and try it out, my life is very different compared to then, spending a few hours playing quake sounds insane lol.

Although I would be interested in how I would do today, honestly I would probably wreck it. My aim continually got better over time, I'm sure my aim now is still better than it was in 2003, even though I haven't played quake in years.

Would just play careful, deny it rail and time an armor.

I played it on pro-q3tourney4 or verticalvengeance, it has the teleporter at center map with two ya's, a mega that spawns every 2 minutes and the railgun that respawns every 15 seconds. I can pretty much guarantee that anyone even today if they played the RC9 version or just the March version would likely still lose xD. RC9 might be the March version for all I know, some old comments allude to this. But it is a straight aimbot, and moves very fast, so even careful play doesn't work sometimes. I died a bunch because of how accurate it is with PG and other weapons as well. On average I lost more than half of my health trying to close out a frag, and prior to trying to close out a frag I had typically about 100 armor. At moments where I wanted to spawn frag I was too weak to commit to it from the previous frag.

Idk how anyone even today would fair against it, you'd have to be a top level dueler to double the computer's score, and most likely they wouldn't even do that, unless they spam raped the spawns fast enough to avoid damage. v51, v40 and v52 don't really match v50 RC9, as you might have read in the comments around your contest.

Yes I am familiar from the bot, and remember that that was literally 20 years ago. The increase in skill in players from 2003 to 2013 was insane. If you play random joe on a server now its not uncommon to see 70% rail 50% lg etc. That would have been very unusual in 2003.

someone put a good record on youtube, since we are not in 2003 anymore

This happened once, was hard.

I can see it being done with spawn frags, but you gotta absolutely rape the spawn perfectly or the cpu will MG you down in seconds. 16 Spawn Frags, that's good effort. I don't actually know the spawns well on T4 since I never played it much in QL, all I did to win was keep the highground and only drop down for items\hp and attacking close range. Its hard to hit any rockets from above since the computer just lunges around like in CPMA, AI will always hit a rail so I died a few times trying to hit some chip damage.

Yeah, that sounds about right, the key is to not let him get the rail out, then it's a long streak of death :) So, it is kind of random, depending on the spawns.

mid air rail

nice

I still have v0.52 on my hdd if anyone interested.

Thanks.

If you have g_cpm 1 set g_weaponrespawn doesn't work so you're stuck with 15 second weapon respawns for cpm duel, lame.

Also you can't pick up items by jumping over them (g_cpm 1).

Otherwise, I like the mod and will continue to use it.

I pitted a 100 spiter bot against a nightmare q3 bot and an open arena bot on hektik_b3 (5 min.). The results were spiter won 18:6 against the oa bot(kyonshi) and 17:1 against the q3 bot (xaero).

I just noticed you uploaded an older version of spiterbot that is supposedly better than the 2023 final version. This might be the case for VQ3, but with g_cpm 1 set "smooth stairs" doesn't work. So when you go up stairs your view will be jarred every step (roughly 20 times on aerowalk RA stairs for example). This is basically unplayable. So I'll stick with the 2023 version.

Also in case anyone didn't read the readme, when you set g_cpm 1 it automatically adds 100 ups to the rl speed so keep it at 900.

EDIT: too buggy, not even gonna use spiterbot.

I apologize I am just now reading your comments. Yeah, so I haven't tested cpm mode yet, tbh I would just recommend CPMA for that, which I find to be more of a challenge than SpiterBot even in VQ3. Apheleon at lvl 100 is actually probably the best bot experience I've had in any Quake to date. I put Spiterbot v.51 and v.50 RC9 out because I wanted to provide a similar experience to CPMA bots for VQ3 and perhaps a less buggy experience than that of v.52 which everyone had.

Though it is extremely unlikely that anyone will update the VQ3 bots in Pro Mode Arena, I still sort of fantasize about being able to play bots that play all of the maps with better pathing than even the current ones that I uploaded. CPMA bots also have better decision making, but they move around pretty weird, no rocket jumping which is big no go for VQ3. Spiterbot was an earlier version of CPMA bots, but strangely the rocket jumping was removed for CPMA.

There are also a few things v.52 does better than v.51 and v.50 RC9, there is better decision making on attacking or fleeing, perhaps significantly better ZTN play. But v.52 fails on a lot of various custom maps, especially the QL ported maps, for instance on Furious Heights it falls instead of jumps onto the Mega after standing idol for a long period of time. On sinister, the bot can't go up the spiral staircase and has no other path to the red. The bot struggles to make gaps. Smaller maps like Toxicity and Aerowalk are 100% unplayable because of pathing. Hektik on v.52 is great.

The versions I had linked do a pretty good job with making gap jumps, rocket jumps, and seem to produce a variation of paths across the maps though I've noticed when I steamroll the computer it becomes significantly more predictable.

it is possible that with v.50 and v.51 cpm mode on aerowalk might work better than that Pro Mode version of the bot, since these bots rocket jump. but there might be the lack of green armor, which is a game breaker.


After some time of playing these bots in VQ3, I figure that while it is like playing a human in some ways, there are ways that make it too easy if you have any kind of experience. The bots don't think about positioning on the map, they are always rotating to items, often times from the low ground which offers the least control of what path to take and makes the players job significantly easier to zone. The bots don't really camp either when they get into good positions, so a form of map control is almost completely non existent when you play them. That is big problem on ZTN, DM6 and Furious Heights.

Even on some maps like Delirium where basically you don't want to ever be on the lowest part of the map since there is minimal visibility of opponents, and less navigation options. The bot is practically glued to those lower areas when it leaves RA room. But one thing that is good about these bots is it can teach you the difference between playing a good player and a bad player, against a good player they won't take the low ground that much and so you'll have to learn to adjust your decision making to that.

I am probably the number one player versus spitebot in term of time spend versus him in the past.

From what I remember.
Spiterbot does not strafejump, only do huge jumps or leaps without innertia. It does not help much for a training simulator since humans does not strafe nor move that way! It also would shoot at you without turning. With his back. When he exit the teleporter he would instashoot you. You cannot fight him in open spaces because he will eat all your stack in 1 sec so you cannot learn angles and play strategical in any way. You only need to hide all the time and use sneaky positioning and angles. No freedom at all.
Only trying to trap him, finding some odd behaviours that you could exploit and timing to death items especially the rail you can beat him.
At some point I played only spitebot on 100 for about 10 months only spitebot and after I played humans I did worst than I did when I trained versus regular bots or when I duel low/medium players.

Cheating movement, cheating aim and wallhack all in one. This is all about the bot.

Not good at all for anything only if you want to feed your curiosity. I don't recommend to anyone if you want to sharped your skills. Don't waste your time with this garbage, like I did.
This is my sincere advice.

later edit
SInce I see some lauding his navigation.
If you really want to see his true "map depth" or navigation "cleverness" try him on dm6, since you cannot get away there with just pure aim, if you are a retard. See how well he'll do there -whisper -not a lot!

The uploads I provided is mainly an appeasement to custom map support. I am going to eventually upload my final patch for the mod soon which seems to change the variance of pathing on DM6 (it will now strafe jump the pillars to get to the Heavy Armor, improves a lot about how the original mod interacts with RC9 (so now you won't get inhuman aim from the bot, it will still dodge and it will still use the new paths and rocket jumps). The AI now will silent walk in your proximity, more especially if you're trying to sneak up on it. As far as the crazy jumps that it does, I need to playtest it again to see if it still does that but I think that might have been fixed in most situations. I find it too easy for me, but it's more balanced overall.

Also, I've gotten a whole lot of comments that the mod doesn't work on sourceports, and I tried the mod on the original quake 3 engine and it seems to restore functionality even in the last version the original author made. So I would give that a shot if you feel like messing with it.

I was participating in planning/testing spiterbot in early versions. I was also author/admin at progamer.ru where all of the dev talk took place. Qrealka was the dev and there were tons of interesting challenges that he had to resolve to make bot act like a human while seeing , hearing, remembering through bsp and aas.

The only person who beat the bot back then was daler (right around he was #2 in qcon)

That is awesome btw, I just noticed your name in the credits an hour ago when I uploaded the hotfix. So I am basically just trying to work out what content could be missing for RC9 to work. Was RC 7-8 needed in order for it to work optimally?

I haven't got all of the release candidates so I am pretty much just tinkering with various files till they work. RC7-8 was the only content I don't have, which might have updated parts of the VM. I am suspecting this to be the case because in the main menu it says an earlier date than in the match, in the menu it dates 'March 29th 2003'. In the game it dates 'June 30th 2003'. So I just figure there may be outdated files in the VM, RC9 just came with a cgame.c update.

The versions I linked on this forum post are actually to v.51, I'll link the .50 RC9 prototype below.

https://www.moddb.com/mods/spiterbot-v50-rc9u...9-hotfix-1