EPTBot on ladder - Page 5
| Forum Index > BW General |
|
RowdierBob
Australia13537 Posts
| ||
|
Dante08
Singapore4203 Posts
On September 19 2026 21:21 TMNT wrote: I saw a game between Pluto and Paralyze on Odyssey. This is like their 6th or 7th game already so Paralyze fully knew what kind of "player" he was dealing with. Once again the bot dismantled him with MM and Tanks on 2 bases. I think the map hack does play a role though. The MM Tanks army was CONSTANTLY moving around the map at a level I dont even think human team melee is possible. Protoss couldn't find any good angle to fight. Surely knowing the movement of your opposition's army all the time plays a role in the bot's army movement? Yeah it'll be interesting to see how it does without maphack, but this is still quite impressive in its current form. But it's not too surprising if it wrecks humans given it has an insane APM and multi-tasking advantage. Would be more interested to see how it fares if you limit its APM and multi-tasking ability (e.g. it can only action items it sees on the screen like humans). | ||
|
vesicular
United States1311 Posts
Certainly if it's using some level of cheats that's lame, but watching some of these games it feels like it's basically just winning with micro. | ||
|
StRyKeR
United States1756 Posts
For the purpose of human vs AI, I wrote a post on the other thread in this forum about formalizing the rules that we expect the AI to follow. Obviously, not map hacking is one of them. Indefinite worker stack, money hacks, mineral walking without vision of minerals, etc. are some other obvious ones. A subtle one is z-ordering, which I believe is a crucial game constraint. It's what allows Zerg to buy time versus Irradiate by stacking Lurkers, as you can only target one of them at a time. It's also what allows Tanks to be hidden under Engineering Bays against the Spawn Broodlings spell. I highly agree with the competition organizer's restrictions as well: "While there is no limit to how fast a bot can run, we do place limits on how slowly a bot can run, after all it is a real-time strategy game. The main rule for the competition is that a bot is not allowed to have too many individual frames that exceed a specific amount of computational time." In effect, this limitation forces the AI to prioritize the actions it takes instead of just "do it all". It probably also naturally limits their APM. Very reasonable to me. | ||
|
TT1
Canada10105 Posts
it keeps the game organized/easier to manage by staying on 2 bases and bio + tank is a very strong comp to max out on, mainly cus bio is 1 supply and it's moving/attacking perfectly while locking down with ghosts (low cost utility vs tech which bio supposed to be weak on "paper" for humans if theyre microing that comp) but this was pretty easy to guess on our own too  , terran requires the fewest expos out of all the races so having them execute/iterate a bio tank ghost style as best as it can is the best bang for your buck in terms of basic/simple execution (for a bot moving and microing everything is simple) + ultimate cost efficient comp off of ~2 bases | ||
|
quaristice
177 Posts
it's also part of why it's so effective with wraith and ghost in tvp and it's putting a lot of stress on the robo build timer meaning less reavers | ||
|
Jealous
10416 Posts
On September 20 2026 01:59 quaristice wrote: it instantly seeing cloaked unit blurs mapwide and scanning them is very broken and feels like a form of cheating on its own it's also part of why it's so effective with wraith and ghost in tvp and it's putting a lot of stress on the robo build timer meaning less reavers Visual intelligence is a very complicated field. What would you suggest as an alternative? I'm struggling to think of anything reasonable or feasible. Feels like this could be abused, too. Move DT into vision, see scan, move it back, rinse repeat, waste all scans, then go kill. | ||
|
quaristice
177 Posts
| ||
|
earob84
Germany195 Posts
https://play.sooplive.com/kalmian/297251825 They even let Pluto play the other races. Didnt know he was able to | ||
|
namkraft
624 Posts
On September 19 2026 21:21 TMNT wrote: I saw a game between Pluto and Paralyze on Odyssey. This is like their 6th or 7th game already so Paralyze fully knew what kind of "player" he was dealing with. Once again the bot dismantled him with MM and Tanks on 2 bases. I think the map hack does play a role though. The MM Tanks army was CONSTANTLY moving around the map at a level I dont even think human team melee is possible. Protoss couldn't find any good angle to fight. Surely knowing the movement of your opposition's army all the time plays a role in the bot's army movement? I casted that Odyssey game. In that particular game, my take is that the bot utilizes peripheral vision of his furthest units to deduce the location, size and composition of Paralyze's army. You could see in some instances that the moment the bot sees some enemy's units, its army movements suddenly changes. But in the 2gate rush game on Attitude, the bot builds a instinctive bunker after seeing the probe coming into his base. While maphack is a possibility, it could be a intentional move to lower the risk of being cheesed when the bot sees a probe that early. Interestingly enough, in the PvZ game played 4 days ago, the bot displayed both signs of maphack and lack of maphack at the same time. He builds the perfect counter stargates / cannons, while sending out units to check for enemy's positions / expansions the way that a human would. | ||
|
quaristice
177 Posts
| ||
|
FaZ-
United States191 Posts
- vs. 빠른비상 (RapiDFlyUp): blocks its own main-natural choke (Octagon) with a Sieged Tank, is never able to push out - vs. OnlyWinWar: gets contained on 2 base (PvZ) against Ling/Lurker, goes Zealot+HT but never pushes - vs. llllllllll: can't figure out where its natural is (Odyssey), takes 3rd and never pushes. The bot (or its owner) CHATS!? - vs. ANSOO[iris]: same issue with Odyssey natural, gets run over by 1-base Zealot+DT into Arbiter+BCs, lol - vs. Shot1234: blocks own ramp with a Sieged Tank again (Knockout), never pushes. - vs. Ssak: badly misplaces natural CC, somehow gets out to a midgame supply lead (macro CC?!?), but fully contained VODs: - vs. Larva: https://vod.sooplive.com/player/207504015?change_second=3774 - vs Ssak: https://vod.sooplive.com/player/207470727?change_second=218 - vs Paralyze: https://vod.sooplive.com/player/207485853?change_second=9034 | ||
|
TMNT
3409 Posts
| ||
|
CostaCustoms
2 Posts
| ||
|
ArmadA[NaS]
United States350 Posts
| ||
|
prosatan
Romania9436 Posts
| ||
|
prosatan
Romania9436 Posts
| ||
|
Jealous
10416 Posts
On September 20 2026 07:55 ArmadA[NaS] wrote: Interesting that it does 1 base tech in pvz. I assume thats because of the maps used during training? Can't imagine 1 base would be optimal on most modern maps with 1-2 zeal gap forge gate wall. It doesn't know how to FFE/GFE wall. However, the degenerate 1 base tech PvZer in me wants to believe that I was right all along 🥲 | ||
|
seRapH
United States9830 Posts
| ||
|
iopq
United States1198 Posts
On September 20 2026 07:55 ArmadA[NaS] wrote: Interesting that it does 1 base tech in pvz. I assume thats because of the maps used during training? Can't imagine 1 base would be optimal on most modern maps with 1-2 zeal gap forge gate wall. I think it just hasn't "learned" to wall effectively since in its own play it needs to discover walling before it can use it in other words, you need to throw a bunch more compute for it to optimize its strats On September 20 2026 12:36 seRapH wrote: Yeah it seems like things developed through theory like wall-ins and build orders will take much longer to brute force in the hyperbolic time chamber. exactly my thoughts | ||
| ||
