Pivot normals is the only good option lol, I don’t know why the foxes even tried up air. The problem is that any aerial can be Sdi’d to the ground before the attacker can land at tas levels
@@fiberjack645 SmashBot is far from TAS perfect. Any bot will always have something a player like Axe can gimp. Maybe real-time machine learning will change that in the future, but for now that's the way it is. If we're just talking TAS vs TAS, one of them will always JV5 the other I suspect.
@@RipleySawzen Yea I was just poking fun since it made no sense, same with tas. Tas assumes that you control the opponent's next move, what people really mean is perfect execution, and even with perfect execution there isn't an option that trumps everything
3:06 White Fox: **Smash DI's out of this plain of existance** Also White Fox: **DI's all the way back into uptilt's hitbox range** O P T I M A L M E L E E
@@bluespy4050 considering the development time. And probably even lower time for play testing.. no. just like people in the beginning of melee thought Zelda was high tier and puff somewhere in low mid tier.
I can totally see SmashBot competitions being a thing in the future. Where instead of having 2 people play, you have 2 people make their own bot and then see which one will win! ;)
tbh mine was when fox stole the ledge from fox and fox died but fox came back and then used shine on fox pushing fox off the stage and forcing fox to recover and thats when fox stole the ledge from fox making the game even
I think the best way to not have the stalemate at the start is just to have the starting position not be neutral. This would solve it (kinda). Because the problem is any time they dash at each other it'll just stale into 20XX again. So I think the randomness (in some manner) is the best way to avoid any stalemate (I'm sure there are others).
@@shiverian5970 simply having non-neutral starting positions would not be enough since there are a lot of situations that lead back into the stalemate. for example, ay 1:02 they have been playing for a bit so their states are obviously not perfectly symmetrical and yet they go into the stalemate again. the only reason they break out of the stalemate at 1:02 is becasue the same anti-stalemate code from the start is still operational.
Provides some interesting insights into how prediction-based the game is. Couple low% combos that I feel like humans would normally be able to follow up on.
Randomly saw the smash bot vs players thing because Ludwig was there any anything Ludwig related blows up in the RUclips algorithm. Decided to check your channel to see if there was an update on smash bot and saw there hadn’t been one in like a year +. Felt mildly bummed, forgot about it for a week, and then suddenly see this video out of nowhere.
If you implement the yoshi stage more in the AI the red bot wouldnt have died at 5:00 because he could have easily landed on that cloud there, that would be way more interesting implenting those moving objects
Truly this is the world imagined in 20xx where you have to read the seed and use that to determine which controller port you want and consider possible last-frame sudden death draws too.
I'm curious, does smash bot work for other characters? If so I wonder if it could be used to make a "perfect" tier list by having all characters fight each other.
Apparently the Slippi guys are getting a file of all inputs from every match. If you took all that data, you could feed the data into a machine learning AI to read the inputs of all matches storing which character was successful. Then feed it live data of someone playing and it react based on the best techniques of all Slippi players. It would just need to take into account the characters and stage.
I had JUST rewatched the original today! Actually, is its sudden popularity why you're doing this? The duel is more varied, but that doesn't save either from their fates being sealed through near endless wavedashing and shines!
It has no mind to know where or when to SDI to or to weave lasers into its neutral game to bait pressure or and sometimes drops easy tech chase situations. But since its now flawed, it cant rely on pure blinding perfect techskill to infinitely bully its opponent. It only keeps itself safe enough to punish and try and gimp with shine where possible. Its strangely some of the coolest gameplay, but its like watching a flowchart ken. You know whats about to happen, but you cant help but keep watching
Awesome video. I noticed a couple issues in the matches with failing to avoid avoidable moves--are enemy attacks just not particularly factored into the bot's play? ie, is it more about aggression and punishes into kill-confirms? Will future development be focused more on defense and the neutral game? The openings per kill seems to be pretty damn near optimized at this point.
I don't play melee, but do you think you'd ever design an AI for one of the bottom-tier melee character? And if you did, would it still dominate the world's best players? I know the gods of melee got JV 5 stocked by smashbot, but what would it be like if smashbot was Bowser or Kirby?
@FlashFire1999 Oh, I do see he managed to beat it with Marth. What was going on there though? Alt F4 in his video getting the pros to fight smashbot said that grabs don't work on smashbot because it will always just spotdodge and punish. But in Axe's video, it looked like he was able to run up and grab sometimes.
@FlashFire1999 @FlashFire1999 Oooh, you're right, I slowed the video down and axe fairs with Marth to get smash bot to shield, the fair hits the shield, Smashbot dashes back and Axe run up grabs. But I'm still a bit confused, because like I said, after Axe hits Smashbots shield, smashbot dashes back and Axe dashes forward and grabs him successfully. Can't smashbot spot dodge after dashing back? I don't know enough about melee to know, but I assumed you could spot dodge right after a dash, and spot dodge comes out quite a few frames before grab. Btw, I'm talking about 17:00 mins in Axe's video, "Axe vs smashbot incredible TAS CPU fox."
@FlashFire1999 Ahh okay, that makes sense then. I didn't know you couldn't spot dodge immediately out of a dash. Clever of Axe to be able to win against this beast.
I think what might be slightly better than injecting code to create slight errors would be to add code to randomly alter Smashbot's approach. One thing that makes computer AI so mediocre is it lacks adaptability. While you can't recode AI on the fly, having the computer choose different approaches at random would make it harder to "lame" the AI, since you would have to be ready for a mix-up. This means Smashbot would play suboptimal when compared to the safest options, but that is something that can work against human opponents. 5:00 AI needs to have Randall in the code. :P
I feel like there would be more interesting videos with this concept if you used bots that functioned with different characters and also did things other than mirror matches. Less than technically optimal but more entertaining for sure.
What would a free for all look like with 4 Smash Bots? Also are you planning on making bots for other characters? I assume this program only works with Fox currently.
It's really interesting that the DI and recovery is so good that even neutral wins don't result in 0 to deaths like half the time, because clearly it's not anticipating frame perfect SDI, and it just chooses neutral options that would let it 0 to death a human player.
But the most fascinating is 1:15 where vanilla fox picks a recovery option that won't work because the other options are upB, which gets shine spiked, and air dodge, which also gets killed by the orange AI.
Man, I love how it kinda behaves like a melee player and does BM/salt shines after grabbing a stock, and the little anime battles where they just dash in their faces and parry each others shines with their own, fucking hilarious.
there should be a code for SmashBot for Yoshi's Story that should check for Randal, seeing as in game 3 P1 could have survived on the first stock seeing as they just missed Randal when recovering
I’ve always wondered who would be top tier if military grade AI took control of characters. Imagine this in the newer smash games with all the characters too.
I saw your video about matches against pro player. It was clear that the smashbot is unbeatable by a human. But have you tried to limit the number of inputs the bot can do each second, in order to compare the strategic skill of the bot, compared to human skills ? Thank you for all of your work!
does smashbot only work with fox? if not, how about some smashbot bowser for some koopa backdash hoverwalk moonslide shenanigans (if Im remembering correctly)
You should do smashbots for the other characters too and see what they’d do against each other in different matchups, like fox vs jiggly puff, or falco
I think it would be way, WAY harder to program a bot for any character other than fox. Fox has something resembling an "optimal playstyle" that is more or less what the Smashbot is trying to do here: use shines priority to win every trade on the first frame of contact, use perfect multishine to stack damage, and then ledgeguard with shine and fire. Even though it's the most theoretically possible smash AI to make "perfect" and it's been in development for years we are still seeing some of the gaps in this very video, the AI doesn't recognize Randall as a recovery option, has predictable DI, frequently tanks damage it does not need to, and seemingly hasn't really optimized much other than aggressive neutral and follow up ledgeguarding.
Imagine a world where only shine is viable. just using a grab would be a MASSIVE callout and the crowd would go wild.
Not just shine but like any followup. I don't think a single aerial was used after an up-tilt or throw.
Pivot normals is the only good option lol, I don’t know why the foxes even tried up air. The problem is that any aerial can be Sdi’d to the ground before the attacker can land at tas levels
@@davidolinger3948 I think it's more based on fighting humans rather than players with TAS SDI.
I feel like if the bot could figure out how to perfectly space up air so only the second hit actually hits then it would use it a lot more.
20XX IS NEAR MAH DUDE
The little moment they have together at 4:43 is beautiful
It's kinda like "Yo, you ready?"
They became sentient for a second then realized they were happier as melee bots
Neutral game be like
Natural beauty.
“I have always loved you”
Someone needs to program the perfect Marth to see if marth truly beats fox 60-40.
yeah
@scar ruclips.net/video/3AvdIWDn7uw/видео.html&ab_channel=AZaxe
Perfect fox loses to axe's marth, therefore it truly is 60-40
@@fiberjack645 SmashBot is far from TAS perfect. Any bot will always have something a player like Axe can gimp. Maybe real-time machine learning will change that in the future, but for now that's the way it is.
If we're just talking TAS vs TAS, one of them will always JV5 the other I suspect.
@@RipleySawzen Yea I was just poking fun since it made no sense, same with tas.
Tas assumes that you control the opponent's next move, what people really mean is perfect execution, and even with perfect execution there isn't an option that trumps everything
Marth fox is 50-50 fox players are just used to winning in every other matchup
3:06
White Fox: **Smash DI's out of this plain of existance**
Also White Fox: **DI's all the way back into uptilt's hitbox range**
O P T I M A L M E L E E
Plane*
@@colinmurphy2214 My B. Thnx
amazing
I watched the first ditto literally yesterday for the first time
Same lol
I did too. It just popped into my recommended yesterday and was like, oh dope, and then this comes out literally today. Spooky.
Yeah true
Same
What’s spooky is that i was just talking about the bot for the first time in a while and then suddenly this pops up?!!! Wo
2 Fox's dash dancing and multishining repeatedly until one of them wins is peak Melee
"Two developers at HAL Laboratories playtest their upcoming title Super Smash Bros. Melee" c. January, 2001
I doubt any of them were even any good at the game honestly 😂😂😂
@@bluespy4050 considering the development time. And probably even lower time for play testing.. no. just like people in the beginning of melee thought Zelda was high tier and puff somewhere in low mid tier.
People used to use Zelda sooo much in the beginning
Fox started out pretty rough, but picked up momentum until fox snatched it from fox at the last second.
Spoilers
Dude why did you spoil it I wanted fox to grab that win
Fox got his ass whooped
The shine teabagging while the opponent dies is hilarious to me for some reason
Funny to me that you have to make them just imperfect enough for the game to not be a stalemate
He programmed random tripping into his own bot
@@satoruriolu6132 To great surprise of fans everywhere, we find out that the peak of Melee is, in fact, Brawl.
@@YetAnotherScrub Never doubt Sakurai
@@YetAnotherScrub My favorite melee is ultimate honestly
The bot is nowhere near "perfect" lmao it just has to have an element of randomness to break the symmetry of the starting position.
I can totally see SmashBot competitions being a thing in the future. Where instead of having 2 people play, you have 2 people make their own bot and then see which one will win! ;)
Kinda like chessbot tournaments?
So SaltyBets but for melee. Lol.
...What you're describing is Amiibo's for ultimate bro.
That smashbot guy so good he can play himself with 2 controllers at the same time :o
Like Desk
Or sakurai
Sakurai is better at that
5:30 Red Fox becomes conscious of itself and reads the code of the other bot
Imagine showing this to Sakurai in 2002
LOL
He would've gone door to door and smashed all the NTSC copies
Imagine showing this to Sakurai now
imagine showing this to a native american dying of smallpox in 1598
@@itsbarkweek6116 hold up
Love to see that AIs BM each other after a kill like people do.
omg i just watched the original again this morning and was thinking "dang I really want an updated version of this"
Same, so im glad it showed up in my recomendations right now
u on the wavelength
Huh, you gotta get Randall usage in there. Perfect bots would know the timing. Was weird seeing fox Firefox and not drift randall
Mabey Randal isn't optimal
Wow I literally got your previous video on this recommended yesterday and then this is released
My favorite part was when Fox used shine.
tbh mine was when fox stole the ledge from fox and fox died but fox came back and then used shine on fox pushing fox off the stage and forcing fox to recover and thats when fox stole the ledge from fox making the game even
If syrox and relno had a baby and wizzrobe blessed it with his smash DI
20XX has truly reached it's peak
It shine teabags when it wins, it's perfect
I PISSED MYSELF AT THE MULTISHINE TEABAG, THE FUCKING DISRESPECT TO ITSELF IS IMMENSE
would be cool if you could add the delay only when it rebounds shine on shine so that they don't mess up regular tech chases and stuff
I think you’re right. Want to break the stalemate but still see the machine be as perfect as it can be.
Wouldn't it then just become whichever bot hits the delay on the shine on shine loses a stock?
I think the best way to not have the stalemate at the start is just to have the starting position not be neutral. This would solve it (kinda). Because the problem is any time they dash at each other it'll just stale into 20XX again. So I think the randomness (in some manner) is the best way to avoid any stalemate (I'm sure there are others).
@@shiverian5970 simply having non-neutral starting positions would not be enough since there are a lot of situations that lead back into the stalemate. for example, ay 1:02 they have been playing for a bit so their states are obviously not perfectly symmetrical and yet they go into the stalemate again. the only reason they break out of the stalemate at 1:02 is becasue the same anti-stalemate code from the start is still operational.
@@shiverian5970
items on let’s go
This is it. We have reached 20XX.
So cool how they've finally learned how to play on stages other than FD
This looks like a reenactment of a Terminal Montage video
“Unfocus your eyes Gohan, you’ll be able to see them better.”
Provides some interesting insights into how prediction-based the game is. Couple low% combos that I feel like humans would normally be able to follow up on.
those TerminalMontage "Melee Fox" animations are 100% accurate
"you know what would make this fun? Random trip ups" - AltF4 and Sakurai
2:20 Anakin vs Obi-Wan, colorized
Why did you pick this specific moment? lol
@@Alienrun because it looks like that part of the fight when they’re both swinging their lightsabers but not actually hitting anything.
This is just what Melee used to look like before I actually started watching it enough.
Also, LET'S GOOOO!!! MY BOY, PORT 2
Randomly saw the smash bot vs players thing because Ludwig was there any anything Ludwig related blows up in the RUclips algorithm. Decided to check your channel to see if there was an update on smash bot and saw there hadn’t been one in like a year +. Felt mildly bummed, forgot about it for a week, and then suddenly see this video out of nowhere.
the best armada vs leffen set that will never happened
Leffen SDs a lot.
If you implement the yoshi stage more in the AI the red bot wouldnt have died at 5:00 because he could have easily landed on that cloud there, that would be way more interesting implenting those moving objects
I could watch this for hours. Its so precise and zany, good stuff.
The funniest thing to me is when one Fox up-tilts the other at
When Sakurai uses two controllers at once in a dlc character presentation
Such an even match it’s almost like they are identical...
It's really cool that the bot works with platforms now too, that must have taken a lot of work
Truly this is the world imagined in 20xx where you have to read the seed and use that to determine which controller port you want and consider possible last-frame sudden death draws too.
Man this is crazy. I never would have imagined what optimal D A N C I N G looked like
I'm curious, does smash bot work for other characters? If so I wonder if it could be used to make a "perfect" tier list by having all characters fight each other.
Apparently the Slippi guys are getting a file of all inputs from every match. If you took all that data, you could feed the data into a machine learning AI to read the inputs of all matches storing which character was successful. Then feed it live data of someone playing and it react based on the best techniques of all Slippi players. It would just need to take into account the characters and stage.
After AlphaZero for Chess there is SmashBot for Melee.
Welcome to the 3rd decade.
6:43 that tech tho
Toby Fox: If I win you will put Sans in smash
Sakurai: Bet
The battle:
SDIing down from the up smash to wall tech and recover from otherwise guaranteed death is insane. Now that's a good bot.
I had JUST rewatched the original today! Actually, is its sudden popularity why you're doing this?
The duel is more varied, but that doesn't save either from their fates being sealed through near endless wavedashing and shines!
Great vid! I love Smashbot shenanigans because it's so interesting and peculiar ;)
This is hilarious to watch lmao.
Curious to see more bots in the future. Marth, Sheik... Falco? :P
Heck, I'd even pay to see a Peach one kick ass lol
God, imagine the connected down smashes.
Awesome!
Could red fox have landed on randall at 5:00?
would be interesting to see a tournament where people fight with their AI they have made
This was starting to look like a terminal montage animation
the moment we've all been waiting for
It's nice to see how far the bot has come since the days of using only 3 moves on FD.
It's like the bots flex when the other dies. *Spams reflect*
It has no mind to know where or when to SDI to or to weave lasers into its neutral game to bait pressure or and sometimes drops easy tech chase situations. But since its now flawed, it cant rely on pure blinding perfect techskill to infinitely bully its opponent. It only keeps itself safe enough to punish and try and gimp with shine where possible. Its strangely some of the coolest gameplay, but its like watching a flowchart ken. You know whats about to happen, but you cant help but keep watching
Bro the shine flexes after every stock are hilarious
The Leffen vs Armada matchup has advanced so aggressively in the past 3 years.
I know it would be way more work, but it would also be way more interesting to make different smashbots for different characters and face those off.
Some DI effects are interesting to notice, but have to be so quick to get the inputs :p
The year is 2064, the last 2 melee players playing a best of 3 visualized
At this ascended stage of melee, is fox or falco better.
Awesome video. I noticed a couple issues in the matches with failing to avoid avoidable moves--are enemy attacks just not particularly factored into the bot's play? ie, is it more about aggression and punishes into kill-confirms?
Will future development be focused more on defense and the neutral game? The openings per kill seems to be pretty damn near optimized at this point.
It would be interesting to have different machine learning approaches for a smash bot.
Would you be able to make a bot play like a top player, for example, a Falco smashbot that plays like Mang0 or BBB?
How does SmashBot react to stage hazards? Could it complete the 1-P modes?
Can we stop to appreciate how long it took the smashbot at the beginning of the video to randomize into Fox
This is the 20XX Hax$ dreamed of, just two fox's shine clashing until the timer runs out and the one with port priority wins
Drinking challenge! Take a shot every time a Fox gets himself killed from SDI-ing shine offstage.
PLEASE MAKE A GUIDE ON SETTING IT UP!!! I can't figure out how to make it run
You literally main bowser
@@MrNapkino I'm not allowed to face smashbot then?
@@SkeletonSSBM nope sorry
@@MrNapkino Rats! Some guy on the internet told me I can't do something. Well that's unfortunate.
Ggs I feel like red has a better understanding of neutral, but White's punish game is cracked.
I don't play melee, but do you think you'd ever design an AI for one of the bottom-tier melee character? And if you did, would it still dominate the world's best players? I know the gods of melee got JV 5 stocked by smashbot, but what would it be like if smashbot was Bowser or Kirby?
@FlashFire1999 Does Axe beat him with the characters smashbot is programmed against though? Isn't he only optimally programmed against the top tiers?
@FlashFire1999 Oh, I do see he managed to beat it with Marth. What was going on there though? Alt F4 in his video getting the pros to fight smashbot said that grabs don't work on smashbot because it will always just spotdodge and punish. But in Axe's video, it looked like he was able to run up and grab sometimes.
@FlashFire1999 @FlashFire1999 Oooh, you're right, I slowed the video down and axe fairs with Marth to get smash bot to shield, the fair hits the shield, Smashbot dashes back and Axe run up grabs. But I'm still a bit confused, because like I said, after Axe hits Smashbots shield, smashbot dashes back and Axe dashes forward and grabs him successfully. Can't smashbot spot dodge after dashing back? I don't know enough about melee to know, but I assumed you could spot dodge right after a dash, and spot dodge comes out quite a few frames before grab. Btw, I'm talking about 17:00 mins in Axe's video, "Axe vs smashbot incredible TAS CPU fox."
@FlashFire1999 Ahh okay, that makes sense then. I didn't know you couldn't spot dodge immediately out of a dash. Clever of Axe to be able to win against this beast.
When Fox fell into the platform and the other Fox just chose to dash dance made me sad.
Can we play against the SmashBot?
Smasbot Vs. Smashbot: Who can shine harder?
I think what might be slightly better than injecting code to create slight errors would be to add code to randomly alter Smashbot's approach. One thing that makes computer AI so mediocre is it lacks adaptability. While you can't recode AI on the fly, having the computer choose different approaches at random would make it harder to "lame" the AI, since you would have to be ready for a mix-up. This means Smashbot would play suboptimal when compared to the safest options, but that is something that can work against human opponents.
5:00 AI needs to have Randall in the code. :P
I feel like there would be more interesting videos with this concept if you used bots that functioned with different characters and also did things other than mirror matches. Less than technically optimal but more entertaining for sure.
What would a free for all look like with 4 Smash Bots? Also are you planning on making bots for other characters? I assume this program only works with Fox currently.
These bots be disrespectful as fuck, I love it
Could you make a Marth bot? It would be interesting to see who would actually win more often.
It's really interesting that the DI and recovery is so good that even neutral wins don't result in 0 to deaths like half the time, because clearly it's not anticipating frame perfect SDI, and it just chooses neutral options that would let it 0 to death a human player.
But the most fascinating is 1:15 where vanilla fox picks a recovery option that won't work because the other options are upB, which gets shine spiked, and air dodge, which also gets killed by the orange AI.
Also points out the obvious flaw that it doesn't run away when it's opponent has invincibility spawn frames.
I was just watching the first video and the sequel came out today?
Man, I love how it kinda behaves like a melee player and does BM/salt shines after grabbing a stock, and the little anime battles where they just dash in their faces and parry each others shines with their own, fucking hilarious.
I loved this a little too much, the bots SDI got me going wild!
For some reason I found myself laughing while watching this, especially during those shine neutral trades.
It would be so cool to have bots of all the top tiers go at it.
I love that their idea of taunting is spamming the fuck out of shine
My favorite episode of dbz
can you give them an artificial reaction time? so they have to read a few frames ahead
Instead of making one "mess up" couldn't you make it so there's an rng chance of a double shine since that would be a mixup for the waveshine clank
might be interesting to see current smash bot play against a previous iteration of smash bot
1:25 is it actually possible to multishine just pressing down b at tas level or are the controller inputs wrong
there should be a code for SmashBot for Yoshi's Story that should check for Randal, seeing as in game 3 P1 could have survived on the first stock seeing as they just missed Randal when recovering
Damn. I always thought that shine was the best move but holy shit.
I’ve always wondered who would be top tier if military grade AI took control of characters. Imagine this in the newer smash games with all the characters too.
I saw your video about matches against pro player. It was clear that the smashbot is unbeatable by a human. But have you tried to limit the number of inputs the bot can do each second, in order to compare the strategic skill of the bot, compared to human skills ?
Thank you for all of your work!
are you able to use the same like "optimal option" for other characters like marth or yoshi?
does smashbot only work with fox? if not, how about some smashbot bowser for some koopa backdash hoverwalk moonslide shenanigans (if Im remembering correctly)
You should do smashbots for the other characters too and see what they’d do against each other in different matchups, like fox vs jiggly puff, or falco
I think it would be way, WAY harder to program a bot for any character other than fox. Fox has something resembling an "optimal playstyle" that is more or less what the Smashbot is trying to do here: use shines priority to win every trade on the first frame of contact, use perfect multishine to stack damage, and then ledgeguard with shine and fire. Even though it's the most theoretically possible smash AI to make "perfect" and it's been in development for years we are still seeing some of the gaps in this very video, the AI doesn't recognize Randall as a recovery option, has predictable DI, frequently tanks damage it does not need to, and seemingly hasn't really optimized much other than aggressive neutral and follow up ledgeguarding.