I really like how much effort you've put into this over the years! I always wanted to do it but I just have too many projects already 😂 Regarding your bot decision making getting stuck walking forwards and backwards, behaviour trees and other AI algls are often used with an immediate 'brain' and an 'action' brain. Immediate is things such "I am being attacked, I need to defend" action is more long term and is actions like attacking. Not sure what your architecture looks like but maybe it could be an interesting idea Edit: I see the root cause of your issue is different but maybe this is also worth your time
what you describe happens to ALOT of bots especially as they get more complicated. Conflicting states was the first thing I looked for. I have a feeling as I buildout more of my threat detection this is something that will come up again. also you should def give making a shot, what's one more project ;)
The community runs their own ladder and matches them up direct so they don't compromise the integrity of the real ladder for players. Real is a bit rough of a term xD They're both imaginary digital ladders. Also I imagine for the most part, people would win against the vast majority of bots the vast majority of the time. They're not really designed to play against people and meta gaming is pretty huge. Some of the bots you see in the video aren't even competing to win, they're sub-goals of like - a miniZerg that Can win, some games, with the least code, so you'll get 12 pool all ins that beat some bots that haven't checked for cheese. A bot that's only goal is to optimize and perfectly execute a flawless X opener, if it wins or loses isn't important, but optimizing resource collection and getting the objective unit onto the battlefield as early as possible, sending it across the map and maybe it wins xD That kind of thing.
Hi, just recently discovered you, very interesting series! I think part of the problem is the name. If you are going to base a bot on PiG, then it better be called something like BaconBot or MetalPork. Otherwise the machines will not like it. Keep up the good work!
that's a good idea haha. The bot was a continuation from the last 2 episodes but I am going to see if they admins will let me rename the bot for sake of the show lol
Should work on re-optimizing the early game for the upcoming balance changes. 27 second gateway stalkers is huge. It's not finalized yet but having 3 seconds sooner stalker against cheese and a bulkier and ranged unit for moving out seems worth the time. It probably won't beat the highest ranked bots on the ladder, but you can be suboptimal in unexpected ways to protect key structures and production, dropping a pylon and tech structure out of the way, far enough that most of the time it's not worth checking for your opponent, and then surprise them with the additional technology, you can postpone your second base instead of canceling until you've detected cheese or not, if no cheese go ahead with building it, if cheese save the minerals lost on the cancel, and use that 100 minerals to fund your offmeta tech. I feel like Disruptors will give similar results to anti-cheese strategy, you'll rank above the bots that haven't learned to micro away from the novas, but you'll hit a wall against those that have, for content and seeing progress for the bot it's a solid 8/10, but as a strategy for winning it's like 2.4/10. Checking for proxy and building up an army with the equity you earn from your opponent crossing the map and defending well to invest in tempest or dark templar blink, you can do some slick micro with DTs blinking forward 1 second of walking distance away from the target so the attack cooldown is minimized and mobility is higher than a human player could reasonably preform while managing the base defense and infrastructure, would be the sort of happy medium of content play and moving up the ladder, and also give your bot a bit of character. Maybe have a trick for each matchup, PvP I imagine you'd run into observers more often and PvT they may or may not have a scanner sweep setup or missile turrets but you can scout those. The upside to tempest is you can relatively low investment get a scout phoenix and incorporate some lift micro into your main army once scouting becomes too dangerous. You might even be able to trick the opponent into thinking your phoenix is hallucinated if you get a Sentry early on and use that to scout. Their - Ohno, Air units - false alarm, they're using a Sentry flag might not get reset for the first real Phoenix, especially if you control it like a hallucination for the scouting mission. All while slowly accumulating a fleet of Tempest to spring on the enemy not preparing for a large airforce.
there is alot of here (added them to my idea bin) Disruptors is next on the line. Completely agree and have similar reasoning, that at lower levels Disruptors will cause alot of havoc for bots that aren't coded to dodge it. Hallucination is something I hadn't considered and playing around with it to mess around with other bot logic could be interesting. Though I can't recall if in the API a bot can tell if its a halluc without detection so I'll need to look deeper. If anythign they could at the minimum body block and have alot of fun with that. DTs are also on my roadmap-would love to incorporate Feardragon's build if I can make it work.
@@Vers-AI The bots have player info and typically they wouldn't know, but if you intentionally expire a hallucination - the enemy might pick up on the fact that most real air units - don't evaporate xD
Yea I like this idea, putting on my todo list. In the Ares Framework there is a `on_unit_took_damage` that I'll test for buildings see if it works. The thing is got to be careful it doesn't cause my whole army to return back to the base or something
Bot is finally on the ladder, if you've been holding off on putting one up. Telling me what's stopping you?
I'm here for this series, very exciting!
really cool video series!
I do love this channel though I don't watch all the time - just know I support the AI scene - you guys are truly creative geniuses
I was working on a sc2 bot back in 2017... keen to try it again lol
way back then!? What language was it in? Did you have it on the ladder?
I really like how much effort you've put into this over the years! I always wanted to do it but I just have too many projects already 😂
Regarding your bot decision making getting stuck walking forwards and backwards, behaviour trees and other AI algls are often used with an immediate 'brain' and an 'action' brain.
Immediate is things such "I am being attacked, I need to defend" action is more long term and is actions like attacking. Not sure what your architecture looks like but maybe it could be an interesting idea
Edit: I see the root cause of your issue is different but maybe this is also worth your time
what you describe happens to ALOT of bots especially as they get more complicated. Conflicting states was the first thing I looked for. I have a feeling as I buildout more of my threat detection this is something that will come up again.
also you should def give making a shot, what's one more project ;)
You can get the bot onto real ladder?? Didn't know that. That's so cool!
Edit: I get from the context of the video that it's referred to an AI ladder.
The community runs their own ladder and matches them up direct so they don't compromise the integrity of the real ladder for players. Real is a bit rough of a term xD They're both imaginary digital ladders. Also I imagine for the most part, people would win against the vast majority of bots the vast majority of the time. They're not really designed to play against people and meta gaming is pretty huge. Some of the bots you see in the video aren't even competing to win, they're sub-goals of like - a miniZerg that Can win, some games, with the least code, so you'll get 12 pool all ins that beat some bots that haven't checked for cheese. A bot that's only goal is to optimize and perfectly execute a flawless X opener, if it wins or loses isn't important, but optimizing resource collection and getting the objective unit onto the battlefield as early as possible, sending it across the map and maybe it wins xD That kind of thing.
Hi, just recently discovered you, very interesting series!
I think part of the problem is the name. If you are going to base a bot on PiG, then it better be called something like BaconBot or MetalPork. Otherwise the machines will not like it.
Keep up the good work!
that's a good idea haha. The bot was a continuation from the last 2 episodes but I am going to see if they admins will let me rename the bot for sake of the show lol
great video! informative content combined with entertainment
super cool!
Should work on re-optimizing the early game for the upcoming balance changes. 27 second gateway stalkers is huge. It's not finalized yet but having 3 seconds sooner stalker against cheese and a bulkier and ranged unit for moving out seems worth the time. It probably won't beat the highest ranked bots on the ladder, but you can be suboptimal in unexpected ways to protect key structures and production, dropping a pylon and tech structure out of the way, far enough that most of the time it's not worth checking for your opponent, and then surprise them with the additional technology, you can postpone your second base instead of canceling until you've detected cheese or not, if no cheese go ahead with building it, if cheese save the minerals lost on the cancel, and use that 100 minerals to fund your offmeta tech. I feel like Disruptors will give similar results to anti-cheese strategy, you'll rank above the bots that haven't learned to micro away from the novas, but you'll hit a wall against those that have, for content and seeing progress for the bot it's a solid 8/10, but as a strategy for winning it's like 2.4/10.
Checking for proxy and building up an army with the equity you earn from your opponent crossing the map and defending well to invest in tempest or dark templar blink, you can do some slick micro with DTs blinking forward 1 second of walking distance away from the target so the attack cooldown is minimized and mobility is higher than a human player could reasonably preform while managing the base defense and infrastructure, would be the sort of happy medium of content play and moving up the ladder, and also give your bot a bit of character. Maybe have a trick for each matchup, PvP I imagine you'd run into observers more often and PvT they may or may not have a scanner sweep setup or missile turrets but you can scout those. The upside to tempest is you can relatively low investment get a scout phoenix and incorporate some lift micro into your main army once scouting becomes too dangerous. You might even be able to trick the opponent into thinking your phoenix is hallucinated if you get a Sentry early on and use that to scout. Their - Ohno, Air units - false alarm, they're using a Sentry flag might not get reset for the first real Phoenix, especially if you control it like a hallucination for the scouting mission. All while slowly accumulating a fleet of Tempest to spring on the enemy not preparing for a large airforce.
there is alot of here (added them to my idea bin)
Disruptors is next on the line. Completely agree and have similar reasoning, that at lower levels Disruptors will cause alot of havoc for bots that aren't coded to dodge it.
Hallucination is something I hadn't considered and playing around with it to mess around with other bot logic could be interesting. Though I can't recall if in the API a bot can tell if its a halluc without detection so I'll need to look deeper. If anythign they could at the minimum body block and have alot of fun with that.
DTs are also on my roadmap-would love to incorporate Feardragon's build if I can make it work.
@@Vers-AI The bots have player info and typically they wouldn't know, but if you intentionally expire a hallucination - the enemy might pick up on the fact that most real air units - don't evaporate xD
@@nicholasiverson9784 haha for sure, will have to see what I can get away with then
The bot beat everyone afk, and also the games people insta quit
Oh i love this ❤
What song at 5:41?
Could you maybe do a dps check on your structure to based on it's current health and the units around it?
Yea I like this idea, putting on my todo list. In the Ares Framework there is a `on_unit_took_damage` that I'll test for buildings see if it works. The thing is got to be careful it doesn't cause my whole army to return back to the base or something
responding to early scouting seems like a hassle so my bot just builds defensively while trying not to fall too far behind in workers
Defensive is a good catch all. But wanted to take it to the offensive to give my build the best chance
It s so fucking cool