There are lots of situations which can be accounted for, and others that can't. A single trace for a projectile along it's velocity to find the projected path is about as good as it gets. It could be done at shorter steps to create more of a sweeping test, allowing the bot to know when it's safe up until the very last ms it needs to start moving, not really realistic though.

a frj isn't exactly a critical rj (in terms of being essential to obtain an item, at least not in a traditional sense - although of course you could argue it IS critical, as if the bot is too slow it's not going to get any items of importance), it can basically be done anywhere and is an option, this still requires a bit of work to determine how much initial space there is ahead on the current path.

Right now though, projectile avoidance, bunny hopping etc. It's all so far down the chain of development comparatively speaking. I cannot stress that enough. Like right now I'm making the node creation tools (both point nodes and navmesh), both editable realtime in-game. General navigation is the biggest hurdle right now, drastically reducing the number of nodes to typical bots and making them aware of spaces they can move within is going to be of much benefit later on when needing to choose where a safe spot is in avoidance cases, as well as where it's safe to arc for bunnyhopping. It also means better performance than traditional node/waypoint only based bots and faster pathfinding result times.

Are you going to release test builds of this? How is it coming along? Not being a jerk but it's been two months and it's one thing to do mental masturbation on a forum and another to actually make a bot. I'd love to see a better QW bot as I play bots more than people and as QW gets older and older that will probably be the only thing left to play against.

The things I don't like about frogbots are:

1. No bunnying and little to no RJing.
2. Perfect item timing, you can abuse them with it but they also do it really well even if they haven't seen the items taken. It's annoying combined with the fact they often can only move around the maps certain ways and are slow as shit compared to a player.
3. Don't really follow QW physics, weird movement in the air especially.
4. Hard to customize them. I find skill 10s the most balanced but they still do certain things really well and others very badly. Be nice if you could tweak all abilities of them or at least give them preferences.

I don't know what you mean by above with node/navmesh but generally there are set routes all players use and bunny almost the same on each map. Certain tricks like bunnying to RA on Aerowalk from the steps are done the same by almost every player. Be cool if you could build those for each map to make the bots more human like, you could even give the bot knowledge on how long each route takes or whatever. It'd also be cool if you could build different routes or preferences into different bots. Just some ideas.

Shockbot, which looked very promising frmo what I saw, is the predecessor to Electro's current qw bot efforts.

http://www.youtube.com/watch?v=qhUl0eRZ1ZY
http://www.youtube.com/watch?v=bIBPrCvxAl4
http://www.youtube.com/watch?v=zycwWMylaTA
http://www.bendarling.net/index.php?page=mods

If there is anyone with the requisite skill, and will to donate his 'spare' time to make an amazing qw bot it is he. Even if it were the case, which it's not, that all that was happening was "mental masturbation on a forum", that would be infinitely better than nothing, which, to make an observation in the same trivially obvious category as yours pg, is what has been before and would be occuring. Were you hoping to remind Electro of his duties, in essentially saying: 'shush don't get our hopes up with cheap talk, so just get to work and deliver the product or stfu' ?

Is it true that frogbot doesn't obey qw physics? or is it just that it exploits them in a mechanical, inhuman way?

Yeah, I hate people getting my hopes up. Based on what I saw in your links it does seem he made a bot (looks pretty much like frogbots) but as far as I can tell nothing was ever released? I also found these links:

http://forums.inside3d.com/viewtopic.php?f=4&t=846
http://forums.inside3d.com/viewtopic.php?p=17663

Which are dated back to 2007/2008 so I doubt anything will ever come of this project but I'd love to be wrong.

yes even more customizable than the UT99 bots would be quite nice .

The bot should prioritise item control according to how it last saw the opponent ( whether it was attacked or glimpsed to a certain direction) and how long it would take the player running away to get an item to get there, and either challenge for it or avoid and go to the next item

I'm looking forward to see this bot!

I pretty much agree with everything everyone has said here. The most annoying part of frog bot to me, on level 20 it happens the most and is scaled down accordingly when skill is scaled down, is the ridiculous air movement that they have. Older versions seemingly didn't have this problem as bad, perhaps it was due to less computing power for the bot to utilise to calculate its ridiculous twirling.

Althought I disagree about the lack of spam, frogbot on level 20 has BRUTAL 'spam', it can predict the spawn patterns correctly more often then not and get 3 or more i spawn frags in a row (like any good player) and of course with its ability to hear sounds it can put a rocket in your face just as you turn a corner whilst bunny hopping. But of course, it does this even if you are silent, this needs to be addressed, you can really never surprise a level 20 bot. Unless you are absolutely raping it, the skill level seems to drop if you are continually killing it, but if you give it a chance back, it will, and so will the skill increasing. But this doesn't happen very often, if it was possible ot make this happen often, that would add to the personality of it. Albeit more prediction spam is needed, not just shooting at the sounds it is hearing or the spawn it thinks (well knows) you are spawning at.

There seems to be an aggression factor in frogbot, this can be observed in pre game, shooting at or near bots causes them to react by shooting back most of the time, Yet some bots don't care until you acutally hit them. Which can lead to a funny chain of events, you shooting at a bot, it shoots back, you dodge, it hits another bot, he shoots back at the bot and they just keep going like it's an actual game. I watched this happen non top of dm3 warmup for 15-20 minutes until the game was started!!! Anyway, if this could be controlled so it could be applied for different personalities and playing styles for bots that would be excellent.

It seems the bots use sound A LOT to determine where you are, with the higher the skill level, the better they are at it. A level 20 bot just knows where you are after a few sounds and seemingly is able to just tell where you are going, which is good and bad. It is good as they can predict where you are trying to go but bad because they generally never make a mistake, but are just too slow (need speed jumping) to ambush or intercept you.

There needs to be a way to stop bots taking items constantly in team play, if I am quaded, with rl and lg on dm6quad and I need ra, the freaking bot takes it every time, you have to shoot him with sg or kill him to stop him taking it. Same goes with weapons and ammo.

I remmeber playing podbot in counter strike a lot, and it had a feature of being able to 'learn' certain paths by playing. So I would issue the command, follow me, and they would (if they said affirmative) would follow me around the map. I would show them all the rushing lanes as ct and t for every map that I played. Eventually some of these files that were made from me 'teaching' the bots waypoints because a few hundred kilobytes big, and there was a noticable difference between vanilla bots and bots that had learned waypoints. For example, if I rushed with bots 10 times on each path way in a map, and 1 path proved (by winning the most rounds going that way) to be more successful, the bots learnt this, as well as the mediocre way and the 'bad' way. This led to more 'intelligent' rushing by the bots, they didn't go the same way everytime, nor every second or third time, but if you shut them down x amount of times, eventually they will get triggered by the file saying, GO THIS WAY WE HAVE WON THIS % OF TIMES GOING THAT WAY, and off they go. Something like this would be good, so that bots can actually make somewhat of an 'informed' decision about which waypoint they will use in what situation. Although frogbot does display some good waypoint decision making, like on dm4, it knows to stand in the middle room and wait for spawns, unless it thinks it can kill you easier by hunting you down. More of this would be great!

Again, coming from counter strike pod bot, the bot would respond to radio calls, perhaps in quake they could look for team chat binds, like Sync at RA, the bot would 'hear' Sync and RA and come to RA (in podbot the bots would say yes or no depending on a few things, one they couldn't get there, under fire etc, or two the last time you said come here they died following you so they just aren't listening anymore). If that is possible, it would make coordination with bots very good.

Weapon switching would be good, instead of just using LG constantly, it could shoot say 6-10 cells and then a rocket at the feet. Or (if prediction shooting was possible) spam some nails at a doorway then shoot a rocket, or spam some pipes and then a rocket etc etc.

Another thing, more KTX commands are needed to be compiled with the bot, frogbot does have a variety of commands, but not enough. For example, you can't turn off the damage in the middle of end.bsp in frogbot, making playing end annoying with bots.

/rant over.

Any news on the development of this bot? =)

Hah.. When it's done (c) //don't know about Electro.. I am making small steps towards ultimate bot once in a few months )
up to now i have a tool that get's players moves out of demo and gather some info about where the hell he was.. like that
info_player_deathmatch 0.1994 (coords are also known)
info_player_deathmatch 40.5
info_player_deathmatch 331.563
light 757.25
light 1058.46
...
light 15963.3
light 17421.5
light_torch_small_walltorch 14761
light_torch_small_walltorch 11225.8
light_torch_small_walltorch 8589.13
light_torch_small_walltorch 7421.62
light 5672.46
....
light 12920.5
light 15007.1
weapon_rocketlauncher 17017.4
weapon_rocketlauncher 13123.5
weapon_rocketlauncher 9384.02 (numbers are high because it's dist^2 not dist)
so it's completely understandable even to computer that it's a way from spot to rl.
that info could be extracted from any mvd demo, any key entities could be added so it could help to built moving base out of best demos.
up to now i didn't even spawn a bot, but i will sure look for it in a week if there would be enough time (personal life and other shit..)
(need to look closely to spaw,setmodel, and other bot_phys stuff and how to pass the move with given coords, angles and maybe user key angles to quake.

Very little but non-zero progress indeed )). My work is still all about understanding players moves now, not working model, for example looking at projection of acceleration (x'') to z-axis you can see where the player jumps of slides or rides on rocket.. in connection to .ent file it gives a complete picture of what moves players did in demos. That's all i've done up to now. so much to do so little free time.

by the way creating an unpredictable-competitive-humanlike bot seem to me still an actual task, because quake gives ultimate perfomance so quakers feel latencies like no other players and internet can never satisfy their needs due to speed of light. so bot dev is actual as before..
hope that Electro would tell a word about his creature.

Didn't read, too long, no patience.
Just that bots are annoying in general and just another hunt for me.
Only good thing is that they mostly are quiet.