Can AI play a track it's never seen before? | Mario Kart Wii

  Рет қаралды 176,218

AI Tango

AI Tango

Күн бұрын

Пікірлер: 586
@sammyclassicstarfoxfan9827
@sammyclassicstarfoxfan9827 Жыл бұрын
I think putting the AI on a track it already learned, but with a texture hack could be interesting, or putting the AI on a different track but with the same textures.
@aitango
@aitango Жыл бұрын
That would be a cool thing to see, you'd expect it to be able to adapt much faster
@ThePapaja1996
@ThePapaja1996 Жыл бұрын
Or drive it backwards
@nashh600
@nashh600 Жыл бұрын
​@@aitangoor a different vehicle
@phabrick8r21
@phabrick8r21 Жыл бұрын
Bikes vs Karts @@nashh600?
@tribble_omg
@tribble_omg Жыл бұрын
The track in mirror mode
@DoMorrMusic
@DoMorrMusic Жыл бұрын
Now I want to see how well the AI does on every track, and then have it go against a human.
@vincenzofranchelli2201
@vincenzofranchelli2201 Жыл бұрын
Human would still crush it
@captainphoton1693
@captainphoton1693 Жыл бұрын
​@@vincenzofranchelli2201hen his little brother
@zxqu2622
@zxqu2622 Жыл бұрын
@@vincenzofranchelli2201 give the ai 10000 hours of non stop training and then do it
@vincenzofranchelli2201
@vincenzofranchelli2201 Жыл бұрын
@@zxqu2622 i wouldnt be surprised. U got 10000 hours?
@zxqu2622
@zxqu2622 Жыл бұрын
@@vincenzofranchelli2201 no i don’t even have 100, but although i would love to see it happen against someone with that many hours
@ghastlyGrenadine
@ghastlyGrenadine Жыл бұрын
I think I get why the AI struggles so hard the moment it makes a single mistake outside of training mode... If it doesn't fall off a cliff, it has no idea how to recover! I think if you could figure out how to train this AI to recover from mistakes? It would be a top player.
@glitchy9613
@glitchy9613 Жыл бұрын
I think just add a reverse option
@RushyoTakha
@RushyoTakha Жыл бұрын
@@glitchy9613 same issue. it'd never choose to use the reverse option after a while because it would never need to use it, so it wouldn't realise when to put it in to effect.
@budgetcoinhunter
@budgetcoinhunter Жыл бұрын
@@RushyoTakha I'm not sure. I'd think that it would quickly learn that if the rewards start to decrease rapidly, it should probably try reversing and then going towards the next checkpoint.
@ghastlyGrenadine
@ghastlyGrenadine Жыл бұрын
@@budgetcoinhunter the problem is that he has it set to auto kill it after a crash... he'd need to train it specifically to recover from a crash in a specific training ruleset
@ghastlyGrenadine
@ghastlyGrenadine Жыл бұрын
@@MustacheMerlin my idea was more of to pre-pick a few definite crashes around the course and give it a huge reward for raising it's checkpoint flag by a decent amount, but I suppose that's also a good way to train it for crashes and normal play all at once...
@OfficialBiggBenn
@OfficialBiggBenn Жыл бұрын
I really like that you keep pointing out the problems with AI approaches (not conceptually, I mean practical problems, like how you need a convolution, and how improvement quickly stagnates). Too many people out there who only point to the cool things possible with technologies from the AI field and extrapolate from there.
@jonahschreiner6531
@jonahschreiner6531 Жыл бұрын
video idea: ai plays against itself (as other racers) and can use items. reward is based off its distance ahead of the others (i.e winning the race)
@alvaro97vlc
@alvaro97vlc 11 ай бұрын
Amazing idea! It could take so much time, but seeing creative ideas like being last in the last lap to get the bullet and crush the game or the IA learning how to aim to other player with a shell... Truly interesting and at the same time scary
@sushi_rolld
@sushi_rolld 21 күн бұрын
bump
@ChuckSploder
@ChuckSploder Жыл бұрын
There's so much stock footage in the intro I thought it was sarcasm
@fernbear3950
@fernbear3950 Жыл бұрын
ML engineer/researcher here. Good stuff. Try training but with occasional input perturbations based upon the average score achieved by the network (will self-normalize as it continues to get better). This should make it more robust towards running into walls, etc, trying to approach a learned invariance. You'll likely need a power distribution of multi-button inputs to simulate things that would cause you to thonk into the wall like hop cancels and etc. This will also provide the requisite noise for your network to explore new strategies, ala genetic evolution. Hope this helps, and this should greatly ameliorate and/or fix your issues here in terms of/w.r.t. the retraining, esp as the 'learning space' of the network becomes so much larger. Gating it based on performance as well ensures that the regularization/extra data-learning/learning to be robust happens naturally as a curriculum, and affords a consistent level of information ingest to the network over the course of training.
@connorkapooh2002
@connorkapooh2002 6 ай бұрын
have you ever read his code or listen to his explanations of convolutional layers
@fernbear3950
@fernbear3950 6 ай бұрын
@@connorkapooh2002 ?
@y..l..8412
@y..l..8412 Жыл бұрын
I think you could get better consistency if you added some random filter to the different images. Making it less relying on the specific looks of levels
@sjoerdgeraeds6757
@sjoerdgeraeds6757 Жыл бұрын
Noise filters
@OuroBoros_Games
@OuroBoros_Games Жыл бұрын
But this does make it lose potential?
@TVPInterpolation
@TVPInterpolation Жыл бұрын
​@@OuroBoros_Gamesusually it isnt the case, computer vision can adapt in such ways rather gracefully.
@user-qw1rx1dq6n
@user-qw1rx1dq6n Жыл бұрын
@@OuroBoros_Gamesweirdly enough usually it’s quite the opposite neural networks tend to have too much capacity that starts causing something called over fitting where the network just becomes really good at memorizing the data you’ve shown it that hurts generalization but if you add noise to the input the network can’t memorize all the data and has to rely on heuristics more becoming more resilient
@thejunkman
@thejunkman Жыл бұрын
I think that is the main thing that is the problem with any of these simple AI's. They don't know and never will know they are playing a game, much less Mario Kart.
@g_fenocchi
@g_fenocchi Жыл бұрын
It would be cool to compare how the AI's average reward compares to that of a human.
@luis-sophus-8227
@luis-sophus-8227 Жыл бұрын
well, humans mostly struggle because of their general skill, which doesn't apply to ais ais are perfect skill but no perception we are perfect perception but no skill
@bananajamin8284
@bananajamin8284 6 ай бұрын
Nah he’s saying if a human driver drove how much reward would the AI detect
@UhDots
@UhDots Жыл бұрын
I’d love to see the AI learn some custom tracks from CTGP as well as any AI’s you’ve had to kill because of failure. Great video as always though!
@mjcox242
@mjcox242 Жыл бұрын
CTGP (Wii) isn't possible with this (easily) You would need a tas bot hooked to a real Wii, then deal with the added anolouge noise and in-ability to save state.
@theunpopularcuber9554
@theunpopularcuber9554 Жыл бұрын
​@@mjcox242The custom tracks (as far as I know) should still be possible though.
@nixienooo
@nixienooo Жыл бұрын
I’d love to see an AI attempt mushroom peaks
@mystora7159
@mystora7159 9 ай бұрын
​@@mjcox242you don't need ctgp to play custom tracks
@mystora7159
@mystora7159 9 ай бұрын
I think it'd be more interesting to see other tracks not in ctgp, specifically the crazy ones like nightmare, electric shredder (lol) and anti gravity area (i think that's what it's called)
@trunkit8749
@trunkit8749 Жыл бұрын
I’d love to see this ai race on a track where every texture is replaced with a color for its corresponding material (like off-road, road, wall)
@gavin_bigred
@gavin_bigred Жыл бұрын
It would be cool to train it on every map, and then give it custom maps and see what it can do with that knowledge. If it’s possible maybe even other Mario kart games
@hansjmo
@hansjmo Жыл бұрын
when approaching a limit like a lap time, progress in terms of time will go down, but the skill is improving. I would not say that Twice as fast is Twice as good, it's way more than that :) if someone is 1% faster than the other close to the theoretical limit, I would say that he is way more than twice as good because it gets harder and harder approaching the theoretical limit :)
@joep2999
@joep2999 Жыл бұрын
That's a great point. Maybe the fitness function could be influenced by the difference between the final time and the human world record. I think an exponential of some kind would capture the skill difference you're talking about.
@samuelthecamel
@samuelthecamel Жыл бұрын
Now for the ultimate challenge: AI vs. Rainbow Road
@starwars4427
@starwars4427 Жыл бұрын
My biggest wish is to see how it fares vs a human player after like the 120 hour mark. Cause just the limitations you put on it with the controls would probably make it pretty easy to beat
@chicklucas6682
@chicklucas6682 Жыл бұрын
150 hour AI vs 10,000 hour player, battle of the ages
@starwars4427
@starwars4427 Жыл бұрын
@@chicklucas6682 Let it run for a while then! I just think itd be interesting to see cause we always talk about how good and objective the AI are so seeing how it fares vs a human would be cool.
@potatojuiceultra
@potatojuiceultra Жыл бұрын
@@theAstarrr It looks like it would actually be pretty easy to beat, since it doesn't go for as many mini turbos and does a lot of random unnecessary hops.
@zchilly7775
@zchilly7775 Жыл бұрын
@@theAstarrr I'm not a grandmaster at Mario Kart Wii, but even my slightly decent lines could stomp the AI. They aren't chaining mini-turbos into wheelies, and they constantly lose their wheelie speed by hopping. I would be 11% faster and gain time on every straightaway, making it a wash
@duckdudette
@duckdudette Жыл бұрын
"Less savvy side-kick" *shows picture of a robin*. I see you, editor! 😂
@leafly7917
@leafly7917 Жыл бұрын
I'm like 90% sure it's an AI that edits those gifs lol
@phabrick8r21
@phabrick8r21 Жыл бұрын
@@leafly7917 Who's editing the film, Batman?
@leafly7917
@leafly7917 Жыл бұрын
@@phabrick8r21 bro what
@realElo.2
@realElo.2 Жыл бұрын
sorry if this isnt true, but i mean this in the kindest way possible: why does the script feel ai written?
@Colewis1
@Colewis1 Ай бұрын
lol
@looki4561
@looki4561 Жыл бұрын
Loved the commentary on this one! Defintiely feels like an improvement to the last video. Overall the video felt made a bit better, but I can not put my finger on what it was.
@aitango
@aitango Жыл бұрын
Really glad you liked it! I'm trying my best to improve each video as much as I can
@austinhurst467
@austinhurst467 Жыл бұрын
I hope winter comes soon for you to keep your room cool. These videos are awesome!!!
@aitango
@aitango Жыл бұрын
Thank you, I've never been so excited for it to get cold haha
@AntoLne
@AntoLne Жыл бұрын
Your videos are genuinely such good quality and keep getting better! I really enjoy watching them and I'm sure you'll soon get the recognition you deserve!
@Delkaryon
@Delkaryon Жыл бұрын
I wish you the same chose
@NZIA
@NZIA Жыл бұрын
will an ai be playing in esi4
@aetherspill
@aetherspill Жыл бұрын
Hi! I love your videos and they're super interesting to me as I go through my software engineering degree. I'd just like to say something about a thing you said in your video. 5:27 "Here's where our AI brings an objective perspective to the table. It doesn't possess these personal biases or preference; it simply just learns and adapts." I think it's important to make sure viewers know that reinforcement learning models are not necessarily always free of bias. It's definitely true for this example and your other Mario Kart AI videos as there's a definitive metric for measuring success in a goal's completion, but reinforcement learning is not a perfect solution to erasing biases from a problem. Whatever biases the creator imparts into the AI (i.e. what is considered rewardable/punishable behaviour), unknowingly or otherwise, will influence how the AI makes its decisions. It's up to us to determine these biases and be aware that we have the ability to make them intrinsic to the AI. Obviously you never claimed that reinforcement learning is free of bias, so I'm not trying to make a callout. I just think this is info that more people should be aware of :>
@theftking
@theftking 11 ай бұрын
After training it for a short time on the new courses, is there a possibility that the AI would actually perform _worse_ on the initial courses it had spent most of its time training on? Basically like, could the AI have become more intelligent overall for a diverse set of courses, yet less specialized for the initial four it had been tailored made for?
@Mezuzah87
@Mezuzah87 Жыл бұрын
What happens when you put the AI back on the original tracks? Does the new track's learning hinder it? Does generalizing it to additional tracks lost its specialization to the original? Maybe this is a silly question lol.
@timberlin4075
@timberlin4075 Жыл бұрын
i think that's called 'catastrophic forgetting' and it's definitely a legit concern
@JordianLoL
@JordianLoL Жыл бұрын
I assumed he trained it on all previous tracks as well just added the tracks into the pool. Or maybe he has weights so he weighs the new maps really high, but the ai still gets some old maps so it doesn't completely forget.
@That_Awesome_Guy1
@That_Awesome_Guy1 Жыл бұрын
@@timberlin4075 I don't know much about this stuff, but what if as they did the transfer learning back and forth on new and old tracks they slowly increased the size of the network? I think that would help it generalize while still allowing capacity for remembering things it has learned.
@Wato-1876
@Wato-1876 Жыл бұрын
i heard jort storm in the background lmao, also amazing video
@branbri8178
@branbri8178 Жыл бұрын
I wasnt a big fan of the random images and gifs but the racing visuals were entertaining :)
@donskelz7771
@donskelz7771 Жыл бұрын
Just curious but does every video use the same code or is it revitalized each time?
@nicolasn12
@nicolasn12 Жыл бұрын
"its going to just rely on the screen" finaly someone did what i always wanted.
@aitango
@aitango Жыл бұрын
I really like that it only uses the screen, makes it feel like it has a level playing field
@PinkyNardo
@PinkyNardo Жыл бұрын
Again another amazing video on how AI can adapt and learn then carry its previous learned attributes to the table in the next stages. It is an awesome way of showing how it can retain its reputation and keep up. All it has to do is tweak a little for the next track and its all good, and all of that is part of the previous tracks processes. This is the AI I want to see impact the future of gaming. Let us Pray 🙏
@illford
@illford Жыл бұрын
Ngl tho if people took this shit online I'd be pissed
@Luweg01
@Luweg01 Жыл бұрын
Really cool to not only see that the ai is doing decent on the unknown tracks but also manages those after very little training. Why dont you do this more often and always start from ground zero ? I would love to see you give an ai a really good control over the game, then start with one track until it can beat it consistently and then go to the next track. It would be interesting to see how long it takes for each new track and how its driving on an unseen track might change depending on the tracks it has seen. You could then see how it performs on the original stage after it has seen so many others. You could then make changes again like enabling npcs or items to see who that changes it or how long it needs to adapt. The important parts are to show enough footage of the new stages and performance that we can see the improvement. You could also turn it into a series, a challenge of sorts to master the entire game.
@autumn4442
@autumn4442 Жыл бұрын
I think the purpose of starting fresh each time is because they're trying to test something new. If you start fresh, you can see how the change affects how the A.I learns differently and compare it to a control. It's really about what your goal is: to see how they learn, or how they adapt when thrown into a new set of circumstances.
@Bleuthatup
@Bleuthatup Жыл бұрын
Ah, yes, the type of literal video editing where if the voice says “square root”, the video shows a square shape, and, next, the root of a tree. Premium stuff
@8BitRip
@8BitRip Жыл бұрын
Just show us what your doing, not random stock footage. It’s so distracting from the actually interesting content
@rafaelleone678
@rafaelleone678 Жыл бұрын
5:39 I laughed because he was doing so good and the speech sounded so motivating
@dougfoster445
@dougfoster445 Жыл бұрын
Thanks for putting the music u use in ur summary and notes! Love the tracks and was grateful you gave the artist credit!
@blockmath_2048
@blockmath_2048 Жыл бұрын
Finally! An AI video where the AI actually learns to drive, and not spit out an evolved series of inputs!
@badischertraum1806
@badischertraum1806 Жыл бұрын
My dream would be a 1v1 between AIs to see how different or similar they learn and if one of them winning the most races
@Atlas_Redux
@Atlas_Redux Жыл бұрын
Could be interesting indeed, having parallel competitive model training.
@icanlevitateoff
@icanlevitateoff Жыл бұрын
these videos are super interesting, kinda nuts that after you trained the initial 4 tracks an hour of training was enough for more tracks
@aitango
@aitango Жыл бұрын
Yeah I honestly didn't expect it to adapt quite that fast
@NeXaSLvL
@NeXaSLvL Жыл бұрын
its almost like it has an infancy stage for the first day
@l-bear_0528
@l-bear_0528 Жыл бұрын
I’d be interested to see a pro MKWii player go against this bot! Keep working hard!
@eduardozanotti7046
@eduardozanotti7046 Жыл бұрын
Awesome video bro!! Great AI you created there. Would love to see more
@Mobtrapper-hq8is
@Mobtrapper-hq8is Жыл бұрын
Now try changing from Funky Flame to a Kart after it’s mastered an entire GP
@kalebheavyshadow5838
@kalebheavyshadow5838 4 ай бұрын
Is it only me but the first 55 seconds kinda went crazy
@WhyHelloThere779
@WhyHelloThere779 Жыл бұрын
I know this would probably take way longer, but I’d love to see the ai with all normal controller inputs (minus 360* turning) and only being rewarded based on what place in the race it came. Wondering what weird strategies is would come up with to win, maybe find shortcuts.
@alansmithee419
@alansmithee419 Жыл бұрын
It would never learn to complete a race. You would have to take a trained AI, and then switch to this reward structure. If you never reward it unless it finishes a race it will never discover that it needs to finish the race in order to get a reward.
@RikkunFops
@RikkunFops Жыл бұрын
If we were to read memory well enough, we could be able to learn when the player character is offroad and punish the AI simply for going offroad. This may speed up and teach the AI to better map out the safe areas of the circuit.
@BobzBlue
@BobzBlue Жыл бұрын
Now put one on ctgp (Maniacal laugh)
@aitango
@aitango Жыл бұрын
(AI cries in mushroom peaks)
@phabrick8r21
@phabrick8r21 Жыл бұрын
@@aitango *Clip montage intensifies*
@bricksalad_
@bricksalad_ 11 ай бұрын
I would really like to see a video of how an AI (similar to the one in your training with items video) could handle HARD CPUs, unless it is already doing that (which would be amazing!).
@aitango
@aitango 11 ай бұрын
It's currently not against hard AI, but that would be really cool and something I'll look to do in the future!
@DougSalad
@DougSalad Жыл бұрын
Stil camt seem to stop them bots from hopping around like bunnies, but was REALLY cool to see how relatively quickly it could adapt to the new courses
@Vernafveik
@Vernafveik Жыл бұрын
I love how the footage used to represent the AI being taken is someone robbing a car
@Kaitri
@Kaitri Жыл бұрын
i like how the thumbnail is basically "ai never seen before" sounding like this is some crazy god ai :D
@shal6178
@shal6178 Жыл бұрын
It really makes sense the AI progress slows down after reaching a certain point, cause like in any speedrun there is less ways of saving time so you get just some micro optimizations of the run
@JRS06
@JRS06 Жыл бұрын
So this is why there's just a million people who always play as Funky Kong and all play fiercely competitive...
@Ascension721
@Ascension721 Жыл бұрын
I get the tiktok text intros are good for the algorithm and thats good and all, but seeing the text come up word by word with the text at the bottom too is kinda funny to me
@dinamiteurdinamiteur2324
@dinamiteurdinamiteur2324 Жыл бұрын
Now, take this AI trained on 4 courses and put it on 4 other ones and see how it adapt. Do it 3 or 4 times and you should have an AI who can plan every tracks or almost
@devilblackdeath
@devilblackdeath Жыл бұрын
I'd love to see every track added in little by little ! See how much that reduces the training time every time, maybe even to the point that it would actually be able to complete a race on its first try ?
@DeloofTegOrb
@DeloofTegOrb 11 ай бұрын
I love all the stock footage used in this video for absolutely no reason lol
@aitango
@aitango 10 ай бұрын
Glad you appreciate it haha
@timhorton8085
@timhorton8085 Жыл бұрын
Subtitles for framed text which is being read allowed [closed captioning available]
@apr1l-813
@apr1l-813 Жыл бұрын
no fucking way bro is using JORT STORM for the bgm
@mackowacko5166
@mackowacko5166 Жыл бұрын
It’d be cool if you let the ai train on one track only for a whole year straight, and then have it go against a human on that track.
@hephaestus8266
@hephaestus8266 Жыл бұрын
Jortstorm, Jortstorm’s coming tonight.
@bugsisland3061
@bugsisland3061 Жыл бұрын
The final evolution of this AI is playing other kart racers
@Woopah194
@Woopah194 Жыл бұрын
This video gave me a interesting idea what if you trained a ai on a track but then once it learnt the track you put it in mirror mode
@memcgee7832
@memcgee7832 Жыл бұрын
this is what we've been waiting for
@mrbanana7110
@mrbanana7110 Жыл бұрын
Now i have my answer ty Next the ultimate test would be to learn the item again and then all of the tracks just so it know the color and then make it compete against mariokart pro but ik this could be impossible as how difficult it looks like
@chadd990
@chadd990 Жыл бұрын
great, now you can train it to navigate the main menu, start up a new game, and complete the Lighting Cup
@dungeontortoise2432
@dungeontortoise2432 Жыл бұрын
Be careful when claiming that AI gives an “objective perspective”. Often, we unintentionally insert our biases when we decide what outcomes should be rewarded, or what tools machines can learn with.
@Atlas_Redux
@Atlas_Redux Жыл бұрын
This, and AIs even gets specific preferences randomly between runs on some training dataset.
@durpy4442
@durpy4442 Жыл бұрын
been waiting for this stuff thanks, instead of just training one for each map
@tristanderoo
@tristanderoo Жыл бұрын
That Yoshi Falls race was impressive!
@professordragon
@professordragon Жыл бұрын
Really impressive that it was able to adapt to the new tracks so quickly!
@HugRunner
@HugRunner Жыл бұрын
I think it would have been interesting if you later let the retrained model play the original tracks again. I'm pretty sure it would get worse on those, and that's the problem with AIs like this. They find a somewhat general solution that is pretty much an average/optimization that can solve several tasks, but only just. It's not becoming an expert on every track individually, or remembering that track 2 has a shortcut or whatever, in case it stumbled upon that in it's initial training and got a huge reward that time. I think that unless you introduce some kind of memory-system for the AI, it will eventually fail (or lose performance) on pretty much all tracks, given it has been trained on enough different tracks, because the "average solution" doesn't apply well enough for all tracks any more.
@Atlas_Redux
@Atlas_Redux Жыл бұрын
This, always try and go BACK to see if the retrained model still manages. That will give an indicator if it's trained to be both adaptable AND retain skillset, or the skillset keeps getting manipulated specifically for the current dataset.
@jolkyb2039
@jolkyb2039 Жыл бұрын
Okay, video was fantastic. Great job. But I have to know... Are you using AI to choose the b roll footage? It's like... Related to what you're saying but only by a thread.
@mikehulsebus
@mikehulsebus 11 ай бұрын
When I talk with my friends we are like "man that track was so hard, I couldn't detect the edges". I am a human.
@aitango
@aitango 11 ай бұрын
How to fail a captcha 101
@frex102
@frex102 Жыл бұрын
instead of learning tracks i wonder how good he could be on the other gamemode A.I is something completely new to me (in a sense that im actually curious about their potential) i’ve been gaming the past 15+ years of my life and being myself not much of a pvp player i often find A.I in different game intricate I know they aren’t neural network they work based on pre determined pattern and I often wonder how far can neural networks go if we give him a ridiculous amount of hour would he be the best at what he do ? would it be better than human ? this is kind of my dream A.I tbh normal A.I can already do crazy things human will never do but even with that they aren’t better. there is a game called rain world i would love train the A.I to (plus screen wise the A.I won’t have much trouble at all recognising things) but I am unable to get a PC myself and I will patiently wait until I get one.
@illford
@illford Жыл бұрын
In this instance with the current conditions it probably wouldn't be better than a human, which is only because the AI doesn't use items, items are like half the life of this game in FFA so not using them puts you at a big disadvantage, the other issue is you need to give it a lot more control but that just increases the amount of time required to an unviable point for any normal person tbh
@notchase
@notchase Жыл бұрын
please stop using stock videos as visual representations of every other word. its really annoying to see all this unrelated stuff happening.
@omarrodriguez1929
@omarrodriguez1929 Жыл бұрын
That's the part you complain about?
@hellium6613
@hellium6613 Жыл бұрын
You should try making a similar ai for a game called track mania, the scenery is so much less diverse and the driving mechanics are simple but elegant. I also feel like only allowing the ai to see the screen isn’t doing us humans justice, we memorize tracks, we don’t just look at the screen and decide what to do based on that. Either way, great video 👍
@nrxpaa8e6uml38
@nrxpaa8e6uml38 Жыл бұрын
It would be interesting if it was possible to make an AI that learns faster, i.e. get the same fitness in much fewer hours. I've had this idea (but no time to implement it myself) that you could train a transformer that just predicts pairs of frames (in the embedding that is produced by the vision CNN) and actions. You see, the transformer learns from literally every frame, even in epochs that fail to produce positive reward, so it could be much more efficient in building a good internal representation of the game. The pre-trained transformer can then be put through reenforcement learning to predict pairs of actions and frames that actually maximize the reward. This is inspired by how ChatGPT was made: They first pre-trained a transformer that just gobbles up all the text there is, and then they use REL to get it to follow instructions. If they didn't have the pre-trained transformer, then it would have been completely hopeless to try to get an instruction-following AI from human feedback alone.
@someperson8249
@someperson8249 Жыл бұрын
I like all the stock images that have minimal relation to the words said
@DeejayNiXoT
@DeejayNiXoT Жыл бұрын
Would AI ever stop driving Zick Zack?
@GroveOfGraves
@GroveOfGraves Жыл бұрын
Have you considered livestreaming the training process of one of these to KZbin? That way you get to record all of it not on your device, plus people can watch if they wanna and check in on its progress. The only downsides I see is if you're data capped or bandwitdth limited
@aitango
@aitango Жыл бұрын
I actually did livestream one of my previous AIs! I might do it again, but the stream is pretty boring since the improvements are so slow. Also I wouldn't want to livestream it unless I know it'll work, which isn't usually the case
@RM-mv5yz
@RM-mv5yz Жыл бұрын
Would love to see the AI go up against a human.
@zephielexe4527
@zephielexe4527 Жыл бұрын
Man, do I wish I knew how to do this. I would love to take an AI and train it to play fighting games. And I’m also curious what the best time your AI gets on those tracks are.
@EXFrost
@EXFrost Жыл бұрын
Fighting games would be hard i think
@phabrick8r21
@phabrick8r21 Жыл бұрын
Smash Ultimate Amiibo without an Amiibo? Nintendo would be furious 😅
@EXFrost
@EXFrost Жыл бұрын
@@phabrick8r21 comment has nothing to do with amiibos. Why smash ultimate amiibos??? You know they were in wii u too, why specify? Also AI is way way more advanced than than the amiibo "learning" system nintendo has in place in smash. Itd be easier to make an ai for traditional fighting games rather than platform fighters as you dont have to worry so much about movement and stage stuff which would be the most difficult thing id think. Itd be very cool to see either tho tbh. Someone might have already trained an AI in melee tho iirc. Haha you read this far and didnt realise im being an asshole and overanalaysing your comment on purpose hahahha
@Strobenz
@Strobenz 8 ай бұрын
This is amazing. Would love to see this AI attempt one of my tracks!
@misterperson3469
@misterperson3469 Жыл бұрын
This was great, exactly what the previous videos made me curious about! Though a note on the "objective difficulty" of tracks: wouldnt a reinforcement learning AI be equally susceptible to gaining a bias due to the track misaligning with how its learned to drive up to that point as a human? Not to mention issues with the reward function, say it can drive off road losing all its speed but hitting the next checkpoint faster. It needs to drive well through that part of the track for it to stop running off the road for that sweet reward bump.
@fattyMcGee97
@fattyMcGee97 Жыл бұрын
I look forward to racing games in the future using this sort of tech, as well as basically every other game type. It’s going to be a lot more engaging fighting an AI that actually knows how to play the game as opposed to the more simple AI’s that we currently see. Especially if the neural network ones learn as you play.
@illford
@illford Жыл бұрын
Tbf idk how good it can be at Mario Kart considering it doesn't really seem to play with Items or any kind of semi-advanced maneuvers. Item wildly scale how good a player is. I also just really wanna see it aim green shells
@fattyMcGee97
@fattyMcGee97 Жыл бұрын
Tango already did a video on this and showed it could be done. His AI wasn’t perfect, but hardware and time limitations is what prevents home from being able to push AI farther. Now, if you hand this to a team of devs with access to much better hardware, I am sure we’ll better results. Also Tango’s results were very promising with his limitations so I’m still looking forward to seeing where neural networks can take us with regards to gaming.
@keygremlin
@keygremlin Жыл бұрын
This is like a newborn getting better than me at Mario Kart within 5 days of birth. We as humanity are so screwed.
@TheSlimHim
@TheSlimHim Жыл бұрын
Idea 1: Train it on all default tracks then have it try against custom tracks. Idea 2: Train it on a track then have it try the mirror version as that would be very visually similar but differnt
@FranXiT
@FranXiT Жыл бұрын
This is your best work yet! The improvement since your previous videos is incredible. Keep it up man!
@KARLOSPCgame
@KARLOSPCgame Жыл бұрын
AI idea: Pick this Mario kart wii trained AI to compete with a Fresh AI in another Kart game with similar mechanics, Crash racing. Etc.
@jahlove40
@jahlove40 11 ай бұрын
Without a doubt: best videos ever, cheesiest b-rolls ever.
@aitango
@aitango 11 ай бұрын
I'll take that as a complement haha
@AzaIndustries
@AzaIndustries 11 ай бұрын
I kind of hope future dev will employ ai learning for AI apponents, it would be so much more chaotic and dynamic.
@aitango
@aitango 10 ай бұрын
I would love to see that, infact I think being that type of developer would be my dream job haha
@aurastrike
@aurastrike Жыл бұрын
I want to see the inputs expanded - perhaps with more specific input types we can see the AI begin to drive smoothly and charge mini-turbos. I would love to see an AI stomp online races because it's just that smooth, precise and.. well, good
@FormulaHavocOrland
@FormulaHavocOrland 8 ай бұрын
I think this is the best way to train ai because its not just memorizing the track. Plus it gets a chance to prove what it’s truly capable of, rather than just continuing to race the same track. Because how do we know whether or not its memorising.
@jackatk
@jackatk Жыл бұрын
0:28 What am I even looking at
@CoreyDavid22
@CoreyDavid22 Жыл бұрын
Nice Vid man!
@eulefranz944
@eulefranz944 Жыл бұрын
You could argue that it already learned how to maximize the reward and in the 1 hour it overfitted with ease to the layout of the new track. To counter this argument train it on a track and then test it on a similar looking track, you could even do a mirror version of the first track first
@Den-gw3hs
@Den-gw3hs Жыл бұрын
Heres an Idea: can AI beat champions road. (i think this would be near impossible maybe do this in the future)
@slouch186
@slouch186 Жыл бұрын
I'd like to see an AI that does have access to information not immediately visible on screen. Real world AI applications usually have these factors and optimizing around them can be interesting
@dmace14
@dmace14 Жыл бұрын
I just noticed the background music in some of the video is the same stock music Charlie slimecicle used to make jortstorm and I’m now hearing it everywhere lol
@CodeF53
@CodeF53 Жыл бұрын
I feel like a better analogy for the retraining is playing on mirrored tracks.
@Yngdady
@Yngdady Жыл бұрын
Ai doesn't have too "see", it just requires basic pathing scripts
@whattheheckman3004
@whattheheckman3004 Жыл бұрын
It’s not in Mario Kart Wii sadly, but I would love to see what an AI would do in Yoshi Valley with its multiple paths.
@alpha4935
@alpha4935 Жыл бұрын
Oh boy, do I have good news for you! Yoshi Valley IS in MKW, just not in the base game. Currently the most popular mod for the game is called Custom Track Grand Prix (CTGP). I'm sure you can already tell what it does from the name alone, but it adds a selection of over 200 fan made tracks, including basically every track from every previous Mario Kart games, some from other racing games and the rest completely original tracks.
@timberlin4075
@timberlin4075 Жыл бұрын
it might experiment a bit with the different paths but end up settling on a somewhat arbitrary path. after a certain amount of time, it'll probably just keep racing the one path it knows the best and wont have the foresight to analyze the map, backtrack, and reevaluate which path is fastest
@thorgs4871
@thorgs4871 Жыл бұрын
I clicked on this expecting it to be the other way around, so to see if the in-game ai could navigate custom maps lol
@Atlas_Redux
@Atlas_Redux Жыл бұрын
No, because they're not AIs. They're highly scripted bots that only works on set levels and does not work if you change it.
@Irohnn
@Irohnn Жыл бұрын
I've been watiting for something like this!
AI Learns Table Tennis
11:20
b2studios
Рет қаралды 1 МЛН
He said this AI is Unbeatable. I took it personally.
16:05
WirtualTV
Рет қаралды 450 М.
when you have plan B 😂
00:11
Andrey Grechka
Рет қаралды 62 МЛН
Touching Act of Kindness Brings Hope to the Homeless #shorts
00:18
Fabiosa Best Lifehacks
Рет қаралды 19 МЛН
HAH Chaos in the Bathroom 🚽✨ Smart Tools for the Throne 😜
00:49
123 GO! Kevin
Рет қаралды 12 МЛН
CPUs troll AI with Items | Mario Kart Wii
10:17
AI Tango
Рет қаралды 176 М.
The History Of Mario Kart 7’s Most Broken TAS
29:34
MKWilliam
Рет қаралды 638 М.
(TAS) Hman's Deathrun 2 with a kart  - 1:42.815
2:10
Pieplup
Рет қаралды 11 М.
The Improved Trackmania AI is Terrifying...
17:01
WirtualTV
Рет қаралды 337 М.
I Made a 32-bit Computer Inside Terraria
15:26
From Scratch
Рет қаралды 3,7 МЛН
AI Learns to DESTROY pensioner AIs (Mario Kart Wii)
10:57
AI Tango
Рет қаралды 66 М.
Ranking Mario Kart Wii tracks on how safe they are
30:33
beanzard64
Рет қаралды 205 М.
Training AI to Play Pokemon with Reinforcement Learning
33:53
Peter Whidden
Рет қаралды 7 МЛН
AI Learns to Play MORTAL KOMBAT
16:50
Will Kwan
Рет қаралды 146 М.
Neural Network learns to beat other AIs (Mario Kart Wii)
15:12
when you have plan B 😂
00:11
Andrey Grechka
Рет қаралды 62 МЛН