One of my favorite things about your videos is that you're not scared of giving us technical details a bunch of other creators leave out. And I always find the discussion of how earlier versions failed really fascinating. A thought I had about the item usage reward, have you considered doing something along the same lines as the guy who made the trackmania AI? His AI never learned how to drift, because driving solidly without drifting is quicker than starting a drift without knowing how to capitalize on it. He then trained his AI to prioritize drifting over everything else basically and then removed the drift reward after it learned how to drift constantly, and it then managed to unlearn the constant drifting, while still retaining the knowledge of how to start a drift. So once the AI is comfortable with using items you don't need to reward it for using them anymore, other than the indirect rewards that come from getting ahead in the race through the item usage.
I look to include even more details in the future! Starting with rewards and phasing them out over time is something I should really do more often, it’s a very elegant solution
Very cool! I'm developing my own Mario kart neural network/reinforcement learning code right now too. This series is really entertaining and has helped with planning/problem solving 🙂
That's really cool, hope it goes well for you. I know Reinforcement Learning can be tough at times, but hope you can get things working! Also, check the description to join a discord server! It's pretty new, but hoping to get many interested devs on there
@@aitango i remember in a previous video some time you said you used the rainbow deep Q learning algorithm. do you still use that? how often does architectures like the algorithms you use change like should you ever restart the learning if a slightly better algorithm comes out?
@@Litschi21 Still in development, learning different neural algorithms and seeing which type best fits the best for my system and provides optimization with both disk space and processing time to execution
12:53 You are playing as Dry Bowser. The timestamp shows DB backing up and getting an item, which is very unlikely to be an action with a positive reward for such a niche play.
You are playing as Dry Bowser as it's shown in the AI coaching section. Also it seems the AI model hasn't learn to take the shortcut, probably a situation too unusual for the model to have learned it casually.
I mean, even for humans, casual players who never looked at other people play usually dont know about most shortcuts, even I was blown away when discovering on the internet certain shortcuts that I never even atempted because they need things like mushrooms and I never found them while having a mushroom as a kid
Gone back through some of your old videos and this is really impressive… is getting to the point where AI as a whole can beat the design of the original developers coded AI.. it’s scary and I hope companies find a way to implement stuff like this in the future for other games to become super adaptive Earned a subscription and a like from me and a tip for just how cool this is
Also missed a POW Dodge while the others didn't and, unless I missed something in the footage, was the only one that knew what a stand-still mini turbo was. It's probably Dry Bowser.
This makes me wanna see AI handle a co-op game, but like this case where there's two or more independent AI that train the Master AI, rather than one AI holding two virtual controllers
Yaay a new video!! I think you was playing dry bowser. And that is because dry bowser plays different and hold his drifts longer compared to the others
As someone with a background in computer science (but not using it in my day job), I really like the technical bits you put in - what rewards are, how you made a mistake rewarding turning during wheelie, stuff like that. It feels like I'm listening to a really good college professor - one that is entertaining enough to keep me interested, but dropping enough important and interesting information that I'm actually learning stuff! Edit to add: A lot of the other comments seemed to identify who you were playing as due to making choices the AIs weren't rewarded to do (driving off-road with an item, reversing for an item, holding onto items, etc.) In a scenario like this, would it be beneficial (or possibly even detrimental!) to show the AI what it's like to try those strategies? For example, once the AI is regularly completing the race, what if you changed its rewards to suddenly value using boost items to take a corner? Or to reward it for holding an item behind it, or the percentage of time it has an unused item, or boost it if an item actually affected an opponent? It would be able to take advantage of the knowledge it already has (how to win) and would potentially learn some human strategies. Or, is this something that would cause problems, and/or go against the idea of creating an AI not based on human driving habits? I ask because ... isn't this all how we learned to play Mario Kart? At first we drove into walls. Then we slowly started finishing tracks. Eventually we learned simple strategies (such as the boost pads, flips during jumps, or corner boosts). But then eventually we start learning more niche mechanics (other color corner boosts, timing item use based on other players' placements and positions, leaving items where they are likely to interfere with other players). Maybe we do it on accident and then realize how big it was (I can still remember my roommate's frustration when I learned how to time lightning when he was going off a jump), or maybe we see others do it and then slowly copy it. Maybe I guess the question is - is there truly a way to teach the AI how to win, while also leaving it flexible enough that it can try new things? Or is that exclusive to the realm of humans (for now)?
As someone who loves Mario Kart and has no idea about IT, I like the way you put the most important information in a nutshell. I appreciate the videos and I'm looking forward for some more!!
I was waiting for a new video on AI Mario Kart! I am so glad it’s here, seriously underrated channel for how interesting this content is. Also everything makes sense which is the best
Fantastic video as always, love seeing your content :) I guessed who you were! Shortcuts with item turbos + item usage + using reverse view gave me a strong suspicion, and the backing up to get a bullet bill was the confirmation ^^
For the items, maybe try giving the AI a continuous small reward for holding items behind it? That would protect it from most assaults, and if it presses and holds the item button shortly after hitting any item box, it'll still nearly immediately use items like bullet bills and stars, while holding things like green shells.
Very cool! Do you think it would be possible to teach it more advanced tactics by connecting a racer to the main AI and have it repeat tactics like dodging blue shells with mushrooms? Or practicing shortcuts?
So I think using what's called "curriculum learning" could definitely be used to do that. Effectively, I'd just get a savestate right before a blue shell when the AI has a mushroom, then the AI would be likely to learn that.
Great Vid! Did you have to account for the different visual placements of the lap counter, position, and item, as with 4 player they're on opposite sides, or did you pull this info directly from the emulator?
Also grew up with Mario kart in various forms. On the mobile app version of the game I don't even play multiplayer since I'll often get pitted against really dedicated players who can consistently lob bananas in front of me. I can only imagine the ai here doing that on an even greater scale and it terrifies me
I’m genuinely curious to see how long the AI would take to beat something like Elden Ring and what the gameplay would look like once it fully mastered everything. Also, great video and commentary. :)
Drybones is the only one who changes their strats based on items. Any time you get a mushroom you offroad the corner and actually time item usage. Also the only character that doesn't have pixel perfect drive lines lol.
I know this would take longer but I’m wondering if you could train it without the position rewards, rather just giving a huge reward at for high positions at the end i’m curious if it will pick up a bagging-type strat also (this would take even longer to train but whatever) maybe lessen the slowness punishment and and increase the checkpoint reward or something so that the ai can find shortcuts?
@@aitango Would it be possible to the AI to play online with random people? Since it seems people are using the dolphin emulator to play now. Obvious problem being the AI can't drive on all the maps.
Quite a bit of work would need to go into setting that up, but it is possible. I'd likely need to get in touch with someone who runs something like Wiimmfi and get special permissions
If there was an AI in the future to learn any game (or Dolphin games only), I'd definitely try it on Shrek Smash n Crash Racing, since I'd love to see what it would be like seeing them melee and such lol!.
I could tell that you were playing as dry bowser because you were adjusting your angle or rapidly moving your joystick while using standstill miniturbos
if you now give each of the four ai drivers full access to all of the screen then coul it probably get to be even more devious and maybe even ambush more
All your videos are really interesting! Do you plan on training a model for time trial runs that could beat world records? I race a very mean ghost valley 2 and id be curious if and how fast it could beat my PB! Its 53.918 no shortcuts/glitch
I would probably avoid giving it a reward for just straight using an item. There is so much to item play that there are moments where it should even be penalized for using an item. For example a player in first place should avoid using an item most of the time to use their current item for protection. Instead, give it rewards for hitting other players with items, blocking incoming items, taking shortcuts with items, etc. It might be a bit complex to teach it the first 2, but for taking shortcuts with items that could probably be done by teaching it that using items in offroad doesn't result in a penalty. Maybe even reward it for using boost items offroad. Of course this could result in weird cases where it's using an item off-road where it's clearly better on-road, so I would make sure it doesn't prioritize it too much.
Would be cool if you gave it reward for going through offroad without a speed detriment, like with a mushroom or whilst hopping (as long as it actually saves time)
You were Dry Bowser. How can i tell? The AI was releasing their mini turbo frame perfect, whereas we could actually see the blue flame on Dry Bowser as you drove.
@@aitango The movement. Drifting on general at the curves and mostly item usage. The Ai's would instantly use their items, however DK would strategisely use them to cut the track and to his advantatge. (However I saw some other people say Dry Bowser, hope that I'm not wrong)
I think instead of rewarding for just using an item, they should be rewarded for successfully using an item. Like if they hit another racer with a shell, or boost through offroad with a mushroom
it seems that for optimal play the AI should focus more on finishing first, not just being first during the race. maybe after it reached a sufficient skill (not sure how to measure that) there should be an "finish position reward" that is slowly scaling up. first i thought it should maybe replace the position reward completely at some point, but i don't think it makes sense to to deprive it of such a useful source of reward altogether.
Cool AI. It plays almost like a pro but I wonder does it can run on different map with the same knoledge. Cause I mean right now you have the basic of what is a mario kart race on a simple map, so theoreticaly it can generalize to other map. Maybe with sub optimal line but you know... I would love to see your AI able to play on different map.
I really want to do something like this, but I have no patience to search through memory values for useful data, nor an idea on how to access them (memory values) in real time in python.
There are so many tells that you're playing as dry bowser, I'd be shocked if anyone didn't get it. Forget the backing up for an item at 12:55, the way the AI drives can only be described as 'indecision'. It still struggles to complete drifts, it loves to hop out of them instead of holding it for the entire corner. It does stupid wheelie-hop-wheelies that no human player would do unless they were looking behind them trying to dodge a shell or something. Also, current MKWii players are much much better players than yourself, as these are the people who have stuck with the game and downloaded things like CTGP and played online with Wiimfi. You'd be below average in online servers today (
The only problem is the AI doesn’t understand concepts like bagging and so it can only front run. If it was set to play the team mode that is popular for competitive, I wonder if it would figure it out by itself.
I wonder if it would preform better or worse if it was given only specific memory values rather than images. I also wonder if it would compute any quicker.
It would certainly compute faster, however the challenge would be collecting enough memory values that actually give the AI enough information. In theory though, it could work very well
idea i notice that if you connect mario kart trough wiimfi the in game cpu's get a little more difficult then normal if you could apply this to the game and train the AI's again may be the then get better by learning from te cpu's
Would it be possible to make a reward for hitting an opponent with an item to boos the chance of that heppening? I believe the game keeps track of it in one way or another because if you hit someone with an item, your character will react, so I feel like it should be.
I bet you can't make an AI that can even beat bots in Speedrunners. Yes, it is like a 2D version of mario kart (sort of) but it will never be able to figure out how to swing or wall jump.
I think they're just what the game defines. However it must also interpolate them to show your position, so, using that data would give effectively unlimited checkpoints.
Yeah so I can technically use as many as I want, but the reason I don't use too many is because they aren't evenly spread out when you start looking really closely. If I use fewer, they tend to be a bit more consistent
One of my favorite things about your videos is that you're not scared of giving us technical details a bunch of other creators leave out. And I always find the discussion of how earlier versions failed really fascinating.
A thought I had about the item usage reward, have you considered doing something along the same lines as the guy who made the trackmania AI? His AI never learned how to drift, because driving solidly without drifting is quicker than starting a drift without knowing how to capitalize on it. He then trained his AI to prioritize drifting over everything else basically and then removed the drift reward after it learned how to drift constantly, and it then managed to unlearn the constant drifting, while still retaining the knowledge of how to start a drift. So once the AI is comfortable with using items you don't need to reward it for using them anymore, other than the indirect rewards that come from getting ahead in the race through the item usage.
I look to include even more details in the future! Starting with rewards and phasing them out over time is something I should really do more often, it’s a very elegant solution
ai tango is a comment bot 💀💀💀💀💀💀🥶🥶🥶🥶🥶
?!?
@@RENZOHENGELMOLEN ?!?
Damn the bots are trying to frame people now lmao
Very cool! I'm developing my own Mario kart neural network/reinforcement learning code right now too. This series is really entertaining and has helped with planning/problem solving 🙂
That's really cool, hope it goes well for you. I know Reinforcement Learning can be tough at times, but hope you can get things working! Also, check the description to join a discord server! It's pretty new, but hoping to get many interested devs on there
@@aitango i remember in a previous video some time you said you used the rainbow deep Q learning algorithm. do you still use that? how often does architectures like the algorithms you use change like should you ever restart the learning if a slightly better algorithm comes out?
How's it going rn (How good is the AI)?
@@Litschi21 Still in development, learning different neural algorithms and seeing which type best fits the best for my system and provides optimization with both disk space and processing time to execution
@@donskelz7771Cool! Let me know if you have any updates (if possible of course)!
12:53 You are playing as Dry Bowser. The timestamp shows DB backing up and getting an item, which is very unlikely to be an action with a positive reward for such a niche play.
That's not the only reason, it was obvious from the way he was playing in general, saving mushrooms for shortcuts for example
Reverse gear also isn't listed in the reward predictions.
Plus DB drove off course purposely when with star power and held items behind him as a shield
You are playing as Dry Bowser as it's shown in the AI coaching section.
Also it seems the AI model hasn't learn to take the shortcut, probably a situation too unusual for the model to have learned it casually.
Yeah, it would probably need special training to learn how to take shortcuts
I mean, even for humans, casual players who never looked at other people play usually dont know about most shortcuts, even I was blown away when discovering on the internet certain shortcuts that I never even atempted because they need things like mushrooms and I never found them while having a mushroom as a kid
he game them a small punishment for driving offroad but shouldve also added a reward for driving on offroad while using mushrooms etc
Gone back through some of your old videos and this is really impressive… is getting to the point where AI as a whole can beat the design of the original developers coded AI.. it’s scary and I hope companies find a way to implement stuff like this in the future for other games to become super adaptive
Earned a subscription and a like from me and a tip for just how cool this is
Thank you ever so much, means the world to hear such kind words
I’m excited to see the agents tackle a full Grand Prix
Imagine if he made 12 agents play online.
The backing up into item boxes at 12:55 did not look like AI behavior, so I guess Dry Bowser.
Also missed a POW Dodge while the others didn't and, unless I missed something in the footage, was the only one that knew what a stand-still mini turbo was. It's probably Dry Bowser.
It's also very obvious if you look at 10:40
Thats really cool how the ai's were able to share training data and not just be 4 different ai's.
Yeah I was unsure if it would work well, but seemed to do amazingly to my happy surprise
Great video! Loved the addition of the AI coaching 👍
Thank you! 👍
Dry Bowser was able to resist the instant and extreme gratification of using the bullet for several seconds, so that was probably the human player!
This makes me wanna see AI handle a co-op game, but like this case where there's two or more independent AI that train the Master AI, rather than one AI holding two virtual controllers
I'm guessing Dry Bowser based on the inhuman speed the other players are releasing their miniturbos
Yeah and when looking at the AI feedback, when the graph spikes who is screwing up? That's right, Dry Bowser
and the little wiggle when he gets hit and does a stmt (standstill mini turbo), its very human, and you can see the AIs do a very awkward stmt
You make some of my favorite videos keep up the great work can’t wait to see what you do next!!
Glad you like them, thanks!
A time trial video against famous historical ghosts would be such an amazing video to watch
Yaay a new video!! I think you was playing dry bowser. And that is because dry bowser plays different and hold his drifts longer compared to the others
As someone with a background in computer science (but not using it in my day job), I really like the technical bits you put in - what rewards are, how you made a mistake rewarding turning during wheelie, stuff like that. It feels like I'm listening to a really good college professor - one that is entertaining enough to keep me interested, but dropping enough important and interesting information that I'm actually learning stuff!
Edit to add: A lot of the other comments seemed to identify who you were playing as due to making choices the AIs weren't rewarded to do (driving off-road with an item, reversing for an item, holding onto items, etc.) In a scenario like this, would it be beneficial (or possibly even detrimental!) to show the AI what it's like to try those strategies? For example, once the AI is regularly completing the race, what if you changed its rewards to suddenly value using boost items to take a corner? Or to reward it for holding an item behind it, or the percentage of time it has an unused item, or boost it if an item actually affected an opponent? It would be able to take advantage of the knowledge it already has (how to win) and would potentially learn some human strategies. Or, is this something that would cause problems, and/or go against the idea of creating an AI not based on human driving habits?
I ask because ... isn't this all how we learned to play Mario Kart? At first we drove into walls. Then we slowly started finishing tracks. Eventually we learned simple strategies (such as the boost pads, flips during jumps, or corner boosts). But then eventually we start learning more niche mechanics (other color corner boosts, timing item use based on other players' placements and positions, leaving items where they are likely to interfere with other players). Maybe we do it on accident and then realize how big it was (I can still remember my roommate's frustration when I learned how to time lightning when he was going off a jump), or maybe we see others do it and then slowly copy it.
Maybe I guess the question is - is there truly a way to teach the AI how to win, while also leaving it flexible enough that it can try new things? Or is that exclusive to the realm of humans (for now)?
As someone who loves Mario Kart and has no idea about IT, I like the way you put the most important information in a nutshell. I appreciate the videos and I'm looking forward for some more!!
Glad you enjoyed it, thank you!
I was waiting for a new video on AI Mario Kart! I am so glad it’s here, seriously underrated channel for how interesting this content is.
Also everything makes sense which is the best
Glad you enjoy it, thanks for the kind words
Fantastic video as always, love seeing your content :)
I guessed who you were! Shortcuts with item turbos + item usage + using reverse view gave me a strong suspicion, and the backing up to get a bullet bill was the confirmation ^^
I am unbelievably happy that someone made this, love MKW and ai is really cool, best of both worlds
Really glad you like it! I think its a pretty cool and unique combo :)
For the items, maybe try giving the AI a continuous small reward for holding items behind it? That would protect it from most assaults, and if it presses and holds the item button shortly after hitting any item box, it'll still nearly immediately use items like bullet bills and stars, while holding things like green shells.
Good video. I have a strong feeling you are playing as dry bowser
Hey! I know Mario games are the most known ones, but there’s a game that I would like to see AI try, it’s Spelunky. Love your videos!
Great video! You can tell that your dry bowser from the frame 1 mini turbos that the AI gets.
The only thing left is to put the AI into the actual game itself... The coaching thing sounds cool too. I imagine it would help speedruners practice
Such a fucking cool channel
Keep up the phenomenal work 💪👍
Thank you, will keep the videos coming!
+1 editing rep
:)
now we need netplay to play against them haha. Im also waiting for the next pro to play against your ai
I want TWD98!
That's really impressive and entertaining
Great Video! I was wondering how you are merging together the different Q learning AIs?
Pretty sure you are bone bowser
You hold your items for protection and as soon as you get a star you get offroad to cut more corners^^
Thank you for another great video, friend. c:
My pleasure!
Very cool! Do you think it would be possible to teach it more advanced tactics by connecting a racer to the main AI and have it repeat tactics like dodging blue shells with mushrooms? Or practicing shortcuts?
So I think using what's called "curriculum learning" could definitely be used to do that. Effectively, I'd just get a savestate right before a blue shell when the AI has a mushroom, then the AI would be likely to learn that.
After watching your video, I want to play this game right away!
You should, Mario Kart Wii is a brilliant game!
Great Vid! Did you have to account for the different visual placements of the lap counter, position, and item, as with 4 player they're on opposite sides, or did you pull this info directly from the emulator?
Great vid tango! And I have a suggestion, Super meat boy, that one would be absolutely insane :)
Is there a version available on Dolphin Emulator? If so I'm happy to give it a go at some point
@@aitango far as I can tell super meat boy was never released for the wii :/
@@aitango oh I didn’t know it had to be dolphin thanks though
Also grew up with Mario kart in various forms. On the mobile app version of the game I don't even play multiplayer since I'll often get pitted against really dedicated players who can consistently lob bananas in front of me. I can only imagine the ai here doing that on an even greater scale and it terrifies me
You're Dry Bowser
He was the only one who would consistently hold items behind him, the AI pretty much never does that
I’m genuinely curious to see how long the AI would take to beat something like Elden Ring and what the gameplay would look like once it fully mastered everything. Also, great video and commentary. :)
Thank you! I haven't played elden ring before, but from what I know I imagine the AI would find very strange ways of beating it!
Drybones is the only one who changes their strats based on items. Any time you get a mushroom you offroad the corner and actually time item usage. Also the only character that doesn't have pixel perfect drive lines lol.
I just got home how did you know to upload right now
It was just for you of course!
He is in your walls...
You have come very far!
Thanks! Good to see some long-term fans of the channel!
I know this would take longer but I’m wondering if you could train it without the position rewards, rather just giving a huge reward at for high positions at the end
i’m curious if it will pick up a bagging-type strat
also (this would take even longer to train but whatever) maybe lessen the slowness punishment and and increase the checkpoint reward or something so that the ai can find shortcuts?
maybe also split the punishment for slowness into a sudden deceleration punishment and a slowness punishment
my toxic trait thinking that I would easily beat them even tho I haven't reached 9999vr 😂😂
Anyways Great vid!
Everybody gangsta until they lose haha
another certified banger 🗣🔥
Thanks, really glad you enjoyed it!
Can’t wait till you try to make a 4-player AI speedrun for NSMB on the Wii for
Do you know how modern reinforcement algorithms perform on FPS games? Would be interesting to see if they can outperform the bots in those games.
True, but depending on the game he could get banned, so i hope that won't be a problem.
lotsa fps games have singleplayers, private matches, and local splitscreen matches, i dont think itd be a problem :) esp like xbox 360 era shooters
I wonder if you are going to be able to train the AI off of staff ghost lines, that would be fun to watch, and then have it compete against CPUS.
I can’t believe my fear of AI comes from someone trying to teach it how to play Mario Kart
Strange how things work haha
The next videos "AI defeating the best players in the world"
Let's hope, if someone like TWD98 would be interested I'd love to make an AI for them to face
@@aitango Would it be possible to the AI to play online with random people? Since it seems people are using the dolphin emulator to play now. Obvious problem being the AI can't drive on all the maps.
Quite a bit of work would need to go into setting that up, but it is possible. I'd likely need to get in touch with someone who runs something like Wiimmfi and get special permissions
It would be cool to see an ai for kirby bosses, or kirby in general cause i wanna see a speedrunner ai beat a boss up lol
cool video
Thanks!
I wonder if it’s possible to train AI to race like a team in a 6 v 6 scenario
Setting that up would be quite tough as playing locally only allows 4! If possible though, it would be very interesting what it could come up with
I need to learn how to make AIs like this.
If there was an AI in the future to learn any game (or Dolphin games only), I'd definitely try it on Shrek Smash n Crash Racing, since I'd love to see what it would be like seeing them melee and such lol!.
T'as video est géniale franchement continue
We need to get Troy in a race with the AI
dry bowser is the only one who doesnt hop randomly
I could tell that you were playing as dry bowser because you were adjusting your angle or rapidly moving your joystick while using standstill miniturbos
The only thing left to do is to use all characters on all circuits against itself
Have you considered giving a reward for hitting other plays with items? Also, maybe a punishment for getting hit?
if you now give each of the four ai drivers full access to all of the screen then coul it probably get to be even more devious and maybe even ambush more
I considered doing that, but thought it might confuse the AI a bit! Would be funny to see the AI start screen-peaking though
All your videos are really interesting! Do you plan on training a model for time trial runs that could beat world records? I race a very mean ghost valley 2 and id be curious if and how fast it could beat my PB! Its 53.918 no shortcuts/glitch
Time trials are something I’ve wanted to do for a really long time! Maybe I’m even going it right now :)
@@aitango Sick! Cant wait to see the results
Imagine developing an Ai to defeat the time trial pro ai's lol.
I would probably avoid giving it a reward for just straight using an item. There is so much to item play that there are moments where it should even be penalized for using an item. For example a player in first place should avoid using an item most of the time to use their current item for protection.
Instead, give it rewards for hitting other players with items, blocking incoming items, taking shortcuts with items, etc. It might be a bit complex to teach it the first 2, but for taking shortcuts with items that could probably be done by teaching it that using items in offroad doesn't result in a penalty. Maybe even reward it for using boost items offroad. Of course this could result in weird cases where it's using an item off-road where it's clearly better on-road, so I would make sure it doesn't prioritize it too much.
Would be cool if you gave it reward for going through offroad without a speed detriment, like with a mushroom or whilst hopping (as long as it actually saves time)
Plus reward for hitting people.
So it's definitely Dry Bowser right
Can you please make AI play Taikyoku Shogi?🙏🙏🙏
I believe the AI company Deepmind already made an AI to play Shogi, using their famous algorithm AlphaZero
i can see by the way you drive, dry bowser haha. also i would love to drive against your best ai
You should play around with the goldeneye game on Wii, which has a 4 player split screen versus mode.
You were Dry Bowser.
How can i tell? The AI was releasing their mini turbo frame perfect, whereas we could actually see the blue flame on Dry Bowser as you drove.
Were you playing as DK?
What makes you think that?
@@aitango The movement. Drifting on general at the curves and mostly item usage. The Ai's would instantly use their items, however DK would strategisely use them to cut the track and to his advantatge.
(However I saw some other people say Dry Bowser, hope that I'm not wrong)
na he is dry bowser he drove back to take an item the bot can't do that
I think instead of rewarding for just using an item, they should be rewarded for successfully using an item. Like if they hit another racer with a shell, or boost through offroad with a mushroom
it seems that for optimal play the AI should focus more on finishing first, not just being first during the race. maybe after it reached a sufficient skill (not sure how to measure that) there should be an "finish position reward" that is slowly scaling up. first i thought it should maybe replace the position reward completely at some point, but i don't think it makes sense to to deprive it of such a useful source of reward altogether.
Cool AI. It plays almost like a pro but I wonder does it can run on different map with the same knoledge. Cause I mean right now you have the basic of what is a mario kart race on a simple map, so theoreticaly it can generalize to other map. Maybe with sub optimal line but you know... I would love to see your AI able to play on different map.
I do have a video on AI generalising to other maps! Check my my video, "Can AI play a track its never seen before?"
itd still get crushed competitively
My idea technically:D
I really want to do something like this, but I have no patience to search through memory values for useful data, nor an idea on how to access them (memory values) in real time in python.
That part is pretty tedious, however there are some good tools out there which make it a lot faster
You were playing as skelobowser right? He holds the miniturbo way longer than the other players… you also reversed for some juicy items
Got ya?
yes
you should make it so that the ai gets a reward for hitting it's item and not just using it
There are so many tells that you're playing as dry bowser, I'd be shocked if anyone didn't get it. Forget the backing up for an item at 12:55, the way the AI drives can only be described as 'indecision'. It still struggles to complete drifts, it loves to hop out of them instead of holding it for the entire corner. It does stupid wheelie-hop-wheelies that no human player would do unless they were looking behind them trying to dodge a shell or something.
Also, current MKWii players are much much better players than yourself, as these are the people who have stuck with the game and downloaded things like CTGP and played online with Wiimfi. You'd be below average in online servers today (
The only problem is the AI doesn’t understand concepts like bagging and so it can only front run. If it was set to play the team mode that is popular for competitive, I wonder if it would figure it out by itself.
is there anyway to teach it shortcuts?
I wonder if it would preform better or worse if it was given only specific memory values rather than images. I also wonder if it would compute any quicker.
It would certainly compute faster, however the challenge would be collecting enough memory values that actually give the AI enough information. In theory though, it could work very well
idea i notice that if you connect mario kart trough wiimfi the in game cpu's get a little more difficult then normal if you could apply this to the game and train the AI's again may be the then get better by learning from te cpu's
Would it be possible to make a reward for hitting an opponent with an item to boos the chance of that heppening? I believe the game keeps track of it in one way or another because if you hit someone with an item, your character will react, so I feel like it should be.
That's neat and all, but when can we play against it? 😏
100k subs :)
this falls for the classic problem of over sampling
How would someone go about getting into making their own ai bots like this?
Dry bowser? because the other 3 would immediately release miniturbos and you cant even see the blue, you held it slightly longer
What learning algorithm are you using? The way you describe it makes me think its not an actor / critic model
The algorithm is a q-learning based approach. It’s built off of rainbow dqn, with additional improvements. I’m hoping to publish it soon
@@aitango interesting, thanks! Great work on this
dry bowser took a shortcut so I assume you were him
were you playing dry bowser?
You got me
I bet you can't make an AI that can even beat bots in Speedrunners.
Yes, it is like a 2D version of mario kart (sort of) but it will never be able to figure out how to swing or wall jump.
is there any chance you release a mod or something to play against the ai? I would love to play against it.
If I knew how to do that I'd love to. If anyone knows how to, please get in touch!
You should make the ai play chess
Is it possible to make more than 20 checkpoints if it needs more?
I think they're just what the game defines. However it must also interpolate them to show your position, so, using that data would give effectively unlimited checkpoints.
Yeah so I can technically use as many as I want, but the reason I don't use too many is because they aren't evenly spread out when you start looking really closely. If I use fewer, they tend to be a bit more consistent
Mario Kart wii online with ai next?
I'd love to! I really hope to get into contact with someone who runs Wiimmfi and set that up
plz realase sourse code so we can use it to
If I hit 100k subs, I will!
🙄
@@aitangoPeople will use it to cheat.
yo if you could can you teach Ai how to play cod? or you could do any type of Shooting game idrc😅
Uhhh... DK or Rosalina maybe?