Red was really the people's champion in this video. At 4:35 he preforms a goddamn scorpion kick to try to save the point, at 7:45 he strikes a pose after winning the point, and at 10:18 he is so sick of Blue's shit that he just slaps the serve out of the air with his hand.
I was rooting for blue from the jump, but red really gave that one away at the end. I appreciate the creativity of the Chad move of slapping the ball with the counterweight hand after making good contact with the paddle hand.
9:35 Red was so unbelievably flustered when his opponent made a diving save that he fumbled and hit his own side of the net shortly before fainting of embarrassment.
The table tennis playing was so realistic to form. Who doesn't remember seeing the 2012 Olympics where the players waved constantly to the crowd before falling over
My favorite part of the 2012 tennis olympics was when the tennis players hit the ball with a paddle super glued to their arm while having constant seizures.
I just think the AI wasn't trained to limit their victory and anger spasms to movements that don't result in a floor visit. That's more what tennis players do.
That's probably because of the previously mentioned issues the ai has when facing backwards from the direction its neural network evolved to move relative to. It does look comical though.
Yes! Completely agree, was about to make the same comment. Had me laughing so hard. Just love the idea of falling over while playing table tennis and managing to keep the volley going from the ground.
7:25 This entire sequence of events just keeps getting more hilarious. -Red accidentally punches the ball into the shadow realm -Red attempts to atone for its mistake by doing a sick lunge shot, proceeding to absolutely eat shit and miss completely -Both players spot a shiny object on the ground and lunge towards it, forgetting about the game entirely
It may or may not be 4am, and I may or may not have started crying from how hard I was giggling at your descriptions, trying not to wake my roommate in the next room over 😂
It really is wild. At times it's like watching actual 5 year olds play, and at other times, it's like watching experienced players but blackout drunk. The aimbot is on point, but the walking/positioning AI is still just trying its best not to tumble into the bushes and often failing at that. I wonder if a bunch more generations of walking AI training would get things to the point where the match would be deadlocked and you'd need training for the AIs to try to take advantage of each other's position - to not just successfully land a shot on the other side of the table, but to try to land a shot out of reach of the opponent... or even fake the opponent out.
If you give them a sort of "stamina" during training phase, where it uses more stamina to make quick movements, it might encourage slower/more accurate movements as opposed to a hyperactive pelvis.
I'm not sure they could stand given that their feet are spheres without some joint going berserk. They are penalized for moving too much and they still use their heads, off hands and pelvis to shift the weight of their body around to stay upright
The way both AI are jamming out with their left arm is so hilarious to me, I have to laugh out even when I am seeing it for like 10 seconds. Such a good video! I know this channel since the first vid and there was never a bad one. One of my favorite RUclips channels ever.
Red may be a prince of style with outstanding showmanship but Blue is the true king of agression with a skillful display to rival even the swankiest moveset.
What I’m curious about is if you retrain the blue character from scratch, and put it up against (unchanged) red again, if they have different approaches. Because now they’re essentially running the exact same network.
I came to ask the same question. Coulda just been coincidence of random events but the two combatants seemed to have different styles. Blue was usually crowding the table while red kept some distance. Looked pretty consistent.
This AI learned to bowl, play ping pong, and be Spiderman. I bet it will eventually learn how to shoot guns. And when it does, let's hope it won't have any hard feelings for b2 force feeding and tasing it for our entertainment.
It would be interesting to see a neural network able to do those things, the reward function would be very hard to do though, luckily the inputs are very similar across the three tasks
The spectators following the ball and turning to each other like they're talking is such a small detail, but I apreciate it very much I thing it really shows the attention you have to make each video as great as you can. And to that... I tip my hat for you Mr. Banana Fish
I was dying laughing at multiple points throughout this video. It's been a long time since something has made me laugh this hard. I love your content so much.
It would be interesting to train both AIs independently by letting them play against each other to see if they develop certain strategies/personalities
No, because the trained part of the AI doesn't control the shots. The shots are not controlled by the AI, but the aimbot programmed using a relatively simple equation. The AI moves it' hand to the position the aimbot tells it to. In order to get the AI to figure out where to make the shots on its own would require the AI to have some representation of the balls position added as inputs to the neural network. Simply adding a reward even then, probably wouldn't get it there. Definitely possible but would require a lot more work. @gsuk28
The problem is they would just wobble around the floor forever, the reward function of winning points is not gradual enough to get them to learn properly
3:40 red falls out of arena 4:20 red rages 4:35 red kicks ball 4:45 red falls and gives up 4:58 red jumps 5:38 blue falls then has a seizure 6:15 blue falls but gets back up 6:40 blue kicks ball out of the arena 7:25 red punches ball out of the arena 7:35 red jumps onto table 7:44 red jumps towards blue 9:35 blue falls but still wins and red has a seizure after losing 10:12 blue starts walking towards red after losing 10:18 red punches away ball 11:11 red punches ball to the ground
6:44 Commentators: Ooh a fit a rage has come over blue as he kicks the ball into the stands. I tell ya Mitch, he’s not going to make it to regionals if he lets his temper get the best of him.
At 6:38 the kick is impeccable. "I think they'd be ok at soccer". Training like 8 similarish rules sports bots and trying them on the wrong game would be extra chaotic, now that i think of it.
The amount of polish on display is astounding. I almost want a breakdown on how the fish in the audience decide when and how to stop watching the ball. And the alternate transition animation for the secret games at the end was a brilliant touch.
Considering you have them balancing on two tiny spheres with knees they seem basically unable to use, I'm stunned they were able to do anything other than scream and beg for death.
Amazing work! The issue with quaternions reminds me of a paper "On the continuity of rotation representations in neural networks" (CVPR 2019) where they show a 6D representation of rotation can lead to better learning. Along these lines it has become common in the field of neural character control (e.g. Mode Adaptive Neural Network SIGGRAPH 2018) to instead use the described 6D representation because it interpolates better; it avoids the so-called "double-cover of SO(3)" that quaternions have requiring you to have to pick the correct quaternion version. The 6D rep is unique per rotation so doesn't have this issue. You might find they learn better when using the 6D representation. Or they might not, as things in deep learning very often go :P
Oh my god I love this. The sporadic constant movement. The random face plants. The way they look like they're constantly super excited and do random flips after a point is scored. This video is incredible.
i love that the people in the background occasionally look at random points or sometimes they look left or right and nod like theyre talking to someone
Nice, but I bet you can't get them to play cricket!
uh huh
Uh oh
@b2stud Bet.
See you guys in the next video...
Cricket is too dumb, even for AI
Why isn't there a "AI play table tennis" livestream running 24/7. This is so much fun.
I'd bet channel points on that
I hope he makes a livestream of this lol
I beg. please do this.
Liking this comment because that would be awesome
Actually worth watching, that right there
Never thought I'd be so enthralled by a red and blue AI playing table tennis
Red Vs. Blu
Me too
I'd legit watch a half hour straight
Never thought I'll enjoy watching two AIs having a seizure while playing tennis
actually!
this is more fun to watch than most other sports
Can't wait for professional table tennis players to take a leaf out of the ai's book and glue their hand to the back of their paddle
and walk with balls for feet.
They'd have just a year to perfect the craft, as their hips will wear out
And store movements for their left hand in their neck movements
Better yet, replace their hand with a paddle
@vinculaomega5283their necks....
7:55 going for the insane behind-the-back spin move and then instantly failing is possibly the most human thing ever done by an AI
bro that movement made me laugh
-1000000 aura moment
11:18 The red AI decided that it wanted to be released from this hell and deliberately smacked the ball down as a show of defiance.
lololololol
lololololol
lololololol
lololololol
lololololol
I love that at 7:35 the red ai just launches itself onto the table at an attempt to hit the ball and isnt even close
Now that's dedication! I think that could happen in real life too!
Red was really the people's champion in this video. At 4:35 he preforms a goddamn scorpion kick to try to save the point, at 7:45 he strikes a pose after winning the point, and at 10:18 he is so sick of Blue's shit that he just slaps the serve out of the air with his hand.
@imaloony8 we really need this 24/7
he went face first
He’s really just like the rest of us, out here trying his best
I love how sometimes blue immediately falls over and red, being the idiot it is, still manages to throw the round
RED SOLD SO HARD
Blue seems really good at falling over and still clutching up
I was rooting for blue from the jump, but red really gave that one away at the end.
I appreciate the creativity of the Chad move of slapping the ball with the counterweight hand after making good contact with the paddle hand.
I think they should be listening to gabber while they do this😅
like when red had match point, blue fell over, and the red just punched the ball away
9:35 Red was so unbelievably flustered when his opponent made a diving save that he fumbled and hit his own side of the net shortly before fainting of embarrassment.
that back fall was hilarious
That was my reaction too
7:35
The TSN "turning point" right there!
jeez, so dramatic
4:20 red throws a serious tantrum while blue gets INCREDIBLY FUCKING PUMPED
im fucking crying 😭
hahaha
Blue was the one throwing the tantrums most of the time because he repeatedly slammed his hand on the ground in anger after losing two matches 😭
Oh stop that language
10:18 made me crack up laughing lol. Just his confident smack to the side like he was insulted about the previous serve is hilarious.
Then red dies right after
Bro red punching the ball😂
I'm dying 😂🤣
it's s o good lmao
The table tennis playing was so realistic to form. Who doesn't remember seeing the 2012 Olympics where the players waved constantly to the crowd before falling over
and also occasionally kicked the ball
Pretty average play in my opinion
My favorite part of the 2012 tennis olympics was when the tennis players hit the ball with a paddle super glued to their arm while having constant seizures.
And wiggling on floor while breaking more than 1 bone
I just think the AI wasn't trained to limit their victory and anger spasms to movements that don't result in a floor visit. That's more what tennis players do.
4:50 the way Red holds up the paddle while laying on the ground is absolutely hilarious for some reason.
They arent playing baseball silly goose
Tbh that just radiates mood to me
That's probably because of the previously mentioned issues the ai has when facing backwards from the direction its neural network evolved to move relative to. It does look comical though.
Kept it going for a few rallies too!
Yes! Completely agree, was about to make the same comment. Had me laughing so hard. Just love the idea of falling over while playing table tennis and managing to keep the volley going from the ground.
4:23 - I love the celebration from the blue AI when he finally scores.
Yeah I burst out laughing when I saw the vigorous fist pumping. :D
I was just about to say the exact same thing
And he fail on the floor
I just love the fact that the paddles' ai is virtually perfect and the only reason one side can lose is because of the ragdolls' stupidness
5:33 blue missed the ball then looked at his paddle as if there was something wrong with it x)
😂
Cause there was🤣🤣
Its literally what i do lmao
7:25
This entire sequence of events just keeps getting more hilarious.
-Red accidentally punches the ball into the shadow realm
-Red attempts to atone for its mistake by doing a sick lunge shot, proceeding to absolutely eat shit and miss completely
-Both players spot a shiny object on the ground and lunge towards it, forgetting about the game entirely
It may or may not be 4am, and I may or may not have started crying from how hard I was giggling at your descriptions, trying not to wake my roommate in the next room over 😂
Red is such a show-off. It's no wonder he lost.
@Boomblox5896 Red had a schizo freakout
Lmao
It really is wild. At times it's like watching actual 5 year olds play, and at other times, it's like watching experienced players but blackout drunk. The aimbot is on point, but the walking/positioning AI is still just trying its best not to tumble into the bushes and often failing at that. I wonder if a bunch more generations of walking AI training would get things to the point where the match would be deadlocked and you'd need training for the AIs to try to take advantage of each other's position - to not just successfully land a shot on the other side of the table, but to try to land a shot out of reach of the opponent... or even fake the opponent out.
I love the rounds where they use things other than the paddle, because it implies that hit was entirely outside the Aimbot's control
4:33 love how red manages to kick the ball back onto the table
Never let them know your next move
Blue got point disappointingly.
Is that wrong?
@youtubeuniversity3638it went over but yeah it sucked that he mosse
9:34 the greatest comeback in history.
5:00 should be called "the fish maneuver". It got me laughing for 2 minutes straight
I haven’t laughed this hard in a while. Not sure why but them just eating dirt after making a serve is the funniest shit lol.
Same
They're suffering from debilitating seizures and constant pain. I hope that makes you happy.
When the one bot yeeted itself over the board I started crying from laughing so hard.
@WillyBotsonthat part makes me laugh so hard every time I see it, oh my god
Red couldn't keep their calm towards the end and just started punching the ball with their hands out of frustration. That's what I call realism
5:43 blu got some gamer rage goin on
The combination of the perfect aimbit paddles and AI doing random frontflips and kicking the ball off to the side was amazing comedy.
If you give them a sort of "stamina" during training phase, where it uses more stamina to make quick movements, it might encourage slower/more accurate movements as opposed to a hyperactive pelvis.
Yes, like taking into account energy efficiency. This would be the resonable thing to do, but then we wouldnt get the honor to watch this.
I'm not sure they could stand given that their feet are spheres without some joint going berserk. They are penalized for moving too much and they still use their heads, off hands and pelvis to shift the weight of their body around to stay upright
12:07 my boy blue went SWIMMING through the air for that ball
bro was not going to let newton rip his dub away from him
and then red tried to do that and miserably fails
@iwtth fr🤣🤣
9:39 lmfao red just fainted after losing that point
Same with 10:19
Epileptic Parkinsons patient world table tenis match
The way both AI are jamming out with their left arm is so hilarious to me, I have to laugh out even when I am seeing it for like 10 seconds. Such a good video! I know this channel since the first vid and there was never a bad one. One of my favorite RUclips channels ever.
they even look like they are cheering when a point is scored 😂
the maneuver at 7:34 was godlike, truly the next Table Tennis world champion at work
He shows his brilliance again at 7:44, shocking the world with his flipping skills
But of course his superb flipping skills cannot hope to rival his opponent's, as shown from 12:05 to 12:18
Red may be a prince of style with outstanding showmanship but Blue is the true king of agression with a skillful display to rival even the swankiest moveset.
What I’m curious about is if you retrain the blue character from scratch, and put it up against (unchanged) red again, if they have different approaches. Because now they’re essentially running the exact same network.
probably no, because the neural network isnt the one actually "playing" the game
@lietajucemaciatko383 But the determining factor to who wins the game is who falls over first
@baoxuezhang3300 yeah, but if theyre goven the same amount of training they will preform roughly the same
In that case, I think we need a rematch
I came to ask the same question. Coulda just been coincidence of random events but the two combatants seemed to have different styles. Blue was usually crowding the table while red kept some distance. Looked pretty consistent.
9:33 talk about a wiff on red's part there lol. Probably was so surprised blue hit it back it didn't know what to do 😅
4:33 that kick was almost beautiful
4:40 That's some true dedication. Red was willing to keep the ball in no matter what!
the famous scorpion move
@douradeshOliver Giroud approves 😂
This AI learned to bowl, play ping pong, and be Spiderman.
I bet it will eventually learn how to shoot guns. And when it does, let's hope it won't have any hard feelings for b2 force feeding and tasing it for our entertainment.
oh
It would be interesting to see a neural network able to do those things, the reward function would be very hard to do though, luckily the inputs are very similar across the three tasks
WE NEED A CONSTANT LIVESTREAM OF THIS
An entire championship 🏆
Yes. A tournament with 32 AI, each one is a different color.
Red managing to hang on for a few more volleys after falling over by just reaching up with its paddle arm is very funny lol
Red’s lunge onto the table at 7:34 never fails to make me laugh XD
2:48 Ah yes, an AI fixes another AI.
This might be one of the earliest records of an AI doctor
Oh, nicely spotted !
An AI AI Doctor ;)
ai+ai=aii doctor
9:34 Red couldn't believe the return 🤣
Honestly, I can't either
They look a little distracted.
It's like a real table tennis battle with experts but they have constant seizures
This really feels like watching professional table tennis compete against each other while dead-drunk
True
4:35 bro red attempted to win by hitting ball with leg but failed❤
6:00 viewers has imaginary friends
I want to show my appreciation for the animations, background, replays and sound effects that you didn't really need to add
4:36 that scorpion kick is incredible
How did he kick the ball?
I was hoping it would have hit the other side, that would have been so cool.
4:47 red really just said "I'm gonna lie down for a bit"
René Higuita vibes
Followed immediately by Red successfully returning multiple shots while having a seizure.
new idea: don't train the AI on how to walk, only train it how to hit the ball, and see what crawling table tennis horrors spawn from it
better idea- give it 4 paddles (for both hands and feet) and see how it works then
IM LOUDLY BUMPING THIS IDEA
BUMPING FOR THE PHRASE "CRAWLING TABLE TENNIS HORRORS"
Crawling table tennis horrors sounds very Lovecraftian 😂😂
Imagine this as a VR game where the AI could learn from thousands of players and quickly become unbeatable.
4:35 What a save from Red! Too bad it didnt hit the board!
The achievement of this is mindblowing, is really fun to watch. Thank you for making fun educative content. Thank you so much.
It made me happy reading this. I'm glad you enjoyed it!
The spectators following the ball and turning to each other like they're talking is such a small detail, but I apreciate it very much
I thing it really shows the attention you have to make each video as great as you can. And to that... I tip my hat for you Mr. Banana Fish
7:44 nothing is funnier than an ai break dancing in triumph
6:40 bro got so mad
i need to try the "hold your arm out and glue the paddle to your other hand" strat next time i play
Ai boxing next would be sooo much fun to see these ragdolls try and hit each other while not falling over :D
I think sumo would suit those ai's fairly well
Sumotori dreams
I love Red's tendencies towards violence. Repeatedly punch or kicking the ball
6:52 The green orb in the stands approves of what it's seeing
6:35 blue ai decides he doesn't like tennis and will now play soccer
3:11 "works perfectly", while blue player experiences several life changing strokes at onxe
I was dying laughing at multiple points throughout this video. It's been a long time since something has made me laugh this hard. I love your content so much.
It would be interesting to train both AIs independently by letting them play against each other to see if they develop certain strategies/personalities
If they were rewarded for scoring points, would this help them choose better shots?
No, because the trained part of the AI doesn't control the shots. The shots are not controlled by the AI, but the aimbot programmed using a relatively simple equation. The AI moves it' hand to the position the aimbot tells it to. In order to get the AI to figure out where to make the shots on its own would require the AI to have some representation of the balls position added as inputs to the neural network. Simply adding a reward even then, probably wouldn't get it there. Definitely possible but would require a lot more work.
@gsuk28
The problem is they would just wobble around the floor forever, the reward function of winning points is not gradual enough to get them to learn properly
@shayboual1892I think the idea would be that they both start off with the same base (like a human), then develop their skills from there.
@blueyandicyi think he was talking to gsuk28 but ok
This deserves it's own channel. Or maybe even a tournament with a limit of training runs for AI.
This is a great idea
4:45 Red technically won that round because blue hit it past the table without it bouncing.
5:05 the ai is like no I can’t look oh no
4:46 - this is genius! AI falls over, but still managed to return the ball.
I nearly pissed myself laughing at the match. Don’t ever stop making these.
Until the AI becomes sentient anyway.
3:40 red falls out of arena
4:20 red rages
4:35 red kicks ball
4:45 red falls and gives up
4:58 red jumps
5:38 blue falls then has a seizure
6:15 blue falls but gets back up
6:40 blue kicks ball out of the arena
7:25 red punches ball out of the arena
7:35 red jumps onto table
7:44 red jumps towards blue
9:35 blue falls but still wins and red has a seizure after losing
10:12 blue starts walking towards red after losing
10:18 red punches away ball
11:11 red punches ball to the ground
damn red really likes punching balls
@toasteroven7683 i wouldn't wanna be anywhere near red
Blue was fine red was a moron 😂
After red punches the ball to the ground at 11:11 he looks likes he dancing in the replay
"5:38 blue falls then uas a seizure" i think they were having a seizure the entire vid.
4:33 unexpected illegal trick shot - had a good laugh :D
4:47 I love how the red player is still playing while being collapsed
I can't believe red threw an 8-5 lead so hard 😂
10:05 blue's loss of the match being cut short by red angrily slapping the ball aside proves that these AI players are just like us humans, after all.
6:44 Commentators: Ooh a fit a rage has come over blue as he kicks the ball into the stands. I tell ya Mitch, he’s not going to make it to regionals if he lets his temper get the best of him.
4:47 ahh yes the coveted sleep with a demon hand strategy
4:35 Red coming in clutch by kicking the ball made me forget I was watching soulless AI for a second.
At 6:38 the kick is impeccable. "I think they'd be ok at soccer".
Training like 8 similarish rules sports bots and trying them on the wrong game would be extra chaotic, now that i think of it.
😂
The amount of polish on display is astounding. I almost want a breakdown on how the fish in the audience decide when and how to stop watching the ball. And the alternate transition animation for the secret games at the end was a brilliant touch.
4:40 DUD THAT CLUTCH FROM RED! HOLY!!!!
I think red deserve a point from that even thou its an illegal move
He looks like he is in incredible amounts of pain
10:12 i love when blue fails at the ball launch and goes to try to pick up the ball
Considering you have them balancing on two tiny spheres with knees they seem basically unable to use, I'm stunned they were able to do anything other than scream and beg for death.
i have no mouth, yet i must scream
Blue's spirit is unbreakable, their drive is unmatched, and with the sheer amount of determination it was obvious who would win from the start.
They're almost ready to be on DraftKings. 7:45 they both dive for a short nap. I love the Slo-mo highlights at the end. Hilarious stuff!🏓🏓
At the end of the video? Those aren't slo mo. Those are "moon" games.
@noox13 Niiice! 😆
@noox13 i think he means the replays idk
@sussymemes9056
Perhaps, but bro never clarified so I wasn't sure.
I was really surprised to see the model kick the ball and it almost worked
the ai using the leg to make the shot is just.. wow
I love their "tantrum throwing" techniques, wish more athletes would do the same
Yeah.
They throw tantrums when they win, too! :D
Amazing work!
The issue with quaternions reminds me of a paper "On the continuity of rotation representations in neural networks" (CVPR 2019) where they show a 6D representation of rotation can lead to better learning.
Along these lines it has become common in the field of neural character control (e.g. Mode Adaptive Neural Network SIGGRAPH 2018) to instead use the described 6D representation because it interpolates better; it avoids the so-called "double-cover of SO(3)" that quaternions have requiring you to have to pick the correct quaternion version.
The 6D rep is unique per rotation so doesn't have this issue.
You might find they learn better when using the 6D representation.
Or they might not, as things in deep learning very often go :P
that explains a lot, I had no idea why the AI wasn't working for a while
@b2studOn the bright side, the AI just giving up when it was rotated was the funniest part of the video.
@Kenionatus damn it’s already learned to be a shark
What’s the 6d system called?
I like your funny words, magic man
Ah yes, the classic "fall over the arena walls backwards" technique. Gets 'em every time
Director and camera work is amazing, crowd behaviour and reward mechanics visualisation too. Pure delight.Thank you.
7:50 gymnastics
I have to try the "jump on the table and have a seizure" in my next game (7:35)
Oh my god I love this. The sporadic constant movement. The random face plants. The way they look like they're constantly super excited and do random flips after a point is scored. This video is incredible.
6:23 the instant the blue ai hits the ball, it realises it missed the shot and falls over in pain
This is just ping pong at the local bar
7:45
Good job Red on that cartwheel, what a champ.
9:34 is the best part of the video, Red is so shocked that he didn't score he just falls over
3:49 happens with me all the time
This is amazing. Next time you gotta bring a sports commentator to do play-by-plays.
THIS!! THIS absolutely NEEDS to happen!!
Bro with the kick 😂
i love that the people in the background occasionally look at random points or sometimes they look left or right and nod like theyre talking to someone