Counter-Strike bots are renowned for being garbage. But don’t tell that to BOT Connor.
Quick primer for non-CSGO folk. Whenever someone quits a competitive match, their position is replaced by a bot. You can control the bot, but only after you die. And that usually happens pretty quickly.
But not this time. Not Connor.
I’ve put thousands of hours into CSGO. I’ve risen to Global Elite twice, fallen, and gotten back again. I’ve never seen a CSGO bot get more than two kills in a round without someone having taken control.
Step aside, AlphaGo. Connor is here.
Thanks, PC Gamer.
Comments
8 responses to “BOT Connor Is Here To Slay”
My nemesis bot is in Half Life. His name is Toshi
I was surprised by how bad the bots are in CSGO (I only recently started playing it) compared to CSS bots and even the original CS bots. Even at the highest setting they’re pretty terrible and their aim is really random. I did find though, they’d go on streaks at times where one bot was in god mode and never missed.
Does anyone remember REAPER bot for Quake? That thing was off the charts, and learnt the map on the fly. If there was such advanced AI 21 years ago, created by a fan, what the hell has gone wrong in the future?
It wasn’t advanced, pathing in Quake was always pretty simple and the AI always knows where you are, it just lies to give you a fighting chance. The difference today is that games work differently in terms of geometry and mechanically.
The easiest was to explain is the Arena shooters of yesteryear largely played out on flat platforms that did not require taking cover as much as moving very fast, whereas new games have to take into account a whole lot more, like;
-Slow movement speed
-Taking cover due to short TTK
-Bullet penetration (in some games)
-Wildly varied inventories (a lot of old arena shooters lacked recoil, spray patterns, alternate fire configurations and bullet drop)
-Regenerating health
-Enemy numbers and their impact on cpu
The problem with AI is not that the people making AI haven’t gotten better, it is that what is required of the AI has become exponentially more complex. The days of “point and click games” are largely over and replaced with games that have wide array of mechanics that have to be taken into account when making an AI.
If only we would accept relatively simple environments that games like F.E.A.R used, we could perhaps get away with level designers pulling another army of clones out of their asses, but it doesn’t work that way anymore.
Obviously some games have no excuse for how garbage tier their AI is, but largely the problem is that making tough, yet satisfying; punishing, yet hard AI is ridiculously difficult.
Tldr; Pathing is easy, it is everything else that makes AI development the dumpster fire that it is today.
Didn’t the guy who made the Reaper bot get hired by Epic to work on the bots for UT99?
AFAIK yes he did.
Vaegrand basically covered it, but I’ll add a bit more (used to make loads of maps, dabbled in Unity). It’s easy to make a bot that is exceptionally difficult to beat, and ReaperBot was hard to beat because it was programmed to be fast. It’s just math, and the computer can never miss if you want it to be that way. The bot always knows where you are and can be 100% accurate. It’s harder to make bots act human though – and that’s what people actually want.
The ReaperBots actually learned by following the player – the player plays the map per normal, and drops ‘breadcrumbs’ which act as waypoints. It observes how players move and act in the map, and adds on top some combat routines to help with aiming and dodging. It doesn’t actually do any real map analysis on its own, it’s all player-driven; which is clever, but it’s not like the bot figures everything out on its own. Many bots worked like this although they didn’t hide it like ReaperBot did – they often had specific console commands to allow players to drop waypoints on an unfamiliar map.
As Vaegrand states the simple level geometry of Quake made it easy to figure out pathing, and they just have to move fast. Today’s AI are expected to take cover and cope with much more complicated level geometry, and that’s harder. I remember HL2 maps have specific AI nodes to help the Combine AI figure out the geometry – these are manually placed by the mapper and direct them to positions they can take up, how to move through the area, where to take cover, etc.
Super AI is easy. Convincing human-like AI opponents on the other hand are another matter.
Everyone praises F.E.A.R, but what they don’t know is that it was really just an ingenious relationship between the AI team, sound and the level designers that made the clones seem so much more alive. The AI was smart enough to know that it needed to stay in squads, flank and flush the player out with grenades; however it was the level designers brilliant placement of environment interactables, coupled with some great sound that made the enemies feel alive.
If we were to have amazing AI come out in a great game it would have to be something kind of like Overwatch, where the maps are relatively simple and there are no moving parts. There would need to be a lot of well placed miniature chokepoints where AI would just happen to find themselves in the right place to use interactive cover and the sound team would need to really dig deep to not create the ever so generic lines we hear in most shooters today.
I think what impresses me most with F.E.A.R was that when I think back, I can only think of like seven enemy types (though they had varied weapon loadouts) throughout the entire game, yet never once did they bore me on my first playthrough.
God I miss that game.