- Видео 10
- Просмотров 27 729 811
Yosh
Добавлен 12 дек 2018
Support me on Patreon: patreon.com/yoshtm
Contact :
Mail - yoshtm.yt@gmail.com
Discord - yosh_tm
Twitter - yoshtm1
Contact :
Mail - yoshtm.yt@gmail.com
Discord - yosh_tm
Twitter - yoshtm1
AI beats multiple World Records in Trackmania
I trained an AI in Trackmania with reinforcement learning, and made it compete against human World Records on 3 different pipe tracks.
Between research, programming and editing, these videos take a long time to produce. Any support on Patreon will help me to spend more time on that in the future :)
• Patreon : www.patreon.com/Yoshtm
Contact
• Discord: yosh_tm
• Twitter: yoshtm1
• Mail: yoshtm.yt@gmail.com
The maps shown in this video can be downloaded on TMX and played in Trackmania Nations Forever:
• 1) One Hella Long Pipe (It requires TMUnlimiter !) - tmnf.exchange/trackshow/8484272
• 2) Calm Down - tmnf.exchange/trackshow/1293088
• 3) Are You Serious ?! - tmnf.exchange/trackshow/5152869
W...
Between research, programming and editing, these videos take a long time to produce. Any support on Patreon will help me to spend more time on that in the future :)
• Patreon : www.patreon.com/Yoshtm
Contact
• Discord: yosh_tm
• Twitter: yoshtm1
• Mail: yoshtm.yt@gmail.com
The maps shown in this video can be downloaded on TMX and played in Trackmania Nations Forever:
• 1) One Hella Long Pipe (It requires TMUnlimiter !) - tmnf.exchange/trackshow/8484272
• 2) Calm Down - tmnf.exchange/trackshow/1293088
• 3) Are You Serious ?! - tmnf.exchange/trackshow/5152869
W...
Просмотров: 2 553 129
Видео
Training an unbeatable AI in Trackmania
Просмотров 12 млн7 месяцев назад
I trained an AI in Trackmania with reinforcement learning, until I couldn't beat it. I just opened a Patreon page, where you can support this RUclips channel with a donation (link below). Between research, programming and editing, these videos take a long time to produce. Any support will help me to spend more time on that in the future ! This one in particular took up most of my time over the ...
A.I. Learns to Drive From Scratch in Trackmania
Просмотров 7 млн2 года назад
I made an A.I. that teaches itself to drive in the racing game Trackmania, using Machine-Learning. I used Deep-Q-Learning, a Reinforcement Learning algorithm. Again, a big thanks to Donadigo for TMInterface ! Contact : Discord - yosh_tm Twitter - yoshtm1
A.I. learns to drive on a pipe
Просмотров 230 тыс.2 года назад
A.I. learns to drive on a pipe in Trackmania, using machine-learning. Thanks to everyone who sent me a replay on the second map ! Contact : Discord - yosh_tm Twitter - yoshtm1 Some more details about the supervised learning part : I'm playing with a keyboard, and A.I. learns to predict which keys I will press for a given set of inputs. I'm recording "Left" and "Right" as 0 (no press...
Can A.I. finish this track without crashing ?
Просмотров 174 тыс.3 года назад
A.I. learns to drive in Trackmania with NEAT algorithm, but it is not allowed to hit walls ! Contact : Discord - yosh_tm Twitter - yoshtm1 Music : ruclips.net/video/yfkI4i0Rz2g/видео.html
A.I. teaches itself to drive in Trackmania
Просмотров 4,9 млн3 года назад
A.I. teaches itself to drive in Trackmania, using NEAT algorithm, which is a particular type of Genetic Algorithm. This algorithm is used to select a neural network with optimal weights, and also an optimal structure. Thanks Trabadia ! His RUclips channel : ruclips.net/user/Trabadia1 More information about NEAT algorithm : neat-python.readthedocs.io/en/latest/neat_overview.html Contact : Discor...
AI plays the new Trackmania game (2020)
Просмотров 38 тыс.3 года назад
A new Trackmania game was released on July 1st. I've adapted the AI from the previous video to drive in this new edition ! Sorry about my accent, English and French subtitles are available if you want :) Some more details about the AI : I'm using artificial neural network, a machine-learning algorithm. The network has 16 inputs (15 wall distances speed) and 4 outputs (forward, backward, left, r...
3000 self-driving cars in Trackmania
Просмотров 17 тыс.4 года назад
Hey :) this is my first k-project on trackmania. In fact, it's my first big editing video on trackmania ! I tried something a bit different : generally, on this type of video, replays are driven one by one by some players. Here, replays are also driven one by one, but everything was done automatically. I wrote a script that drives automatically a replay on a map, then saves it, then starts agai...
If artificial intelligence is already this smart, it will even produce alternative possibilities for how people will live in the future. I think true artificial intelligence is and will be smarter than humans.
I've always wondered why no one made a living car I wonder if I'll be the first to go this far
People who watched CGP Grey: I am multiverses ahead of you
My initial criticism is the heavy accent made it hard to understand and follow, since I couldn't watch the screen or subtitles and so I only listened (as I was busy doing other things.) The pipe in the video reminded me of the inverted pendulum, where small changes can have large effects. The weight of the car falling off it should be -inf. Calling the game deterministic is misleading. Floating-point numbers introduce rounding errors that cause divergence, lookup nonlinear dynamic systems. Even though you can have replicable replay files, these as you've shown are only based on the same start times, change that a little and it breaks. Are the replay files for these WR runs in the video available for download anywhere? Please stop calling it AI. The model is trained on your data, making it only slightly better than random inputs, and its success is largely due to chance, there's nothing intelligent about it. This is akin to "training an AI to roll a million sixes on a dice in a row": eventual success by probability doesn't indicate true intelligence or consistent performance. The Trackmania car completing one run isn't a true success if it can't be consistently replicated, indicating significant randomness in the game or the model. This is a critical requirement, especially for applications like self-driving cars where reliability is crucial. The absence of any source code or replay files: open sourcing the code would allow for peer review and constructive feedback, potentially leading to improvements. But I realise, it's clearly just a pseudo-scientific story-telling clickbait video made for entertainment and views.
I don't understand much about it, but your reward function doesn't consider recovery. The AI never fails, and because of that, if it fails once, it breaks down. Looks like a survivorship bias story. Though I can't say how would you consider that without rewarding it for risking without need. But if it did finish, I believe it would have a lot of swag strats. Good luck
wow, thats pretty impressive!
What if you turn up the "resolution" of the ai's Infos and possibile interactions, if thats possibile
When I saw that butterfly, I thought you're gonna get deep into chaos theory. Thank you for the great video and for not complicating that topic further this video. I would love to see a project discovering the chaotic behaviour of collision with pipes, as an example, in Trackmania.
Well obviously haos is amplified by teh velocity/speed that the AI has. I am pretty sure, its easier to predict an outcome when the change possible is smaller, hence when the AI goes for more precision (like driving slower bnackwards) it has more frames to react to unexpected outcomes. This would yield more precise behaviour. Soi am pretty sure the problem is your reward system as always with reinforcement learning. Have you tried scalign up the model and goign for PPO or other methods that do an indirect reward signal modelling? That way the AI can more easily curve fit to the real objective wich in this case is not the small rewards but reaching the goal!
imagine doing this with a supercomputer for 3 decades
I have an idea for your loss function that might help your AI play more consistently. I think your main issue is that you're rewarding your AI for playing as fast as possible. This means it could be reaching speeds which are impossible to play at even with perfect inputs. Instead, I think you should reward it for being "ahead of pace" to the current world record, but not give it extra points for surpassing that speed. If it gains points for being ahead of pace, it wont keep trying to drive impossibly fast, while still achieving a record. My reward would be primarily determined by distance along the track, with an additional reward for speed which stops increasing when pace exceeds the record by more than a certain amount. (Also im not saying you havent already tried this)
11:18 the reason for this is likely because on long straights it has more time to pick up higher speeds. The faster you go the less room for error there is on when you start to decelerate for a corner.
faster than Alpine in F1
Can anyone explain this vs tool assist runs ?
If the developers of trackmania claim that it's 100% deterministic - and it's NOT then that's a bug that they should be aware of - and should seek to fix. However, sensitivity to initial conditions - leading to chaos is NOT indeterminancy. WIth a game of this complexity - and using some kind of realistic physics model - then it would be VERY surprising if it was not chaotic - the real world is chaotic - and an excellent simulation of the real world must also be chaotic. I think you should reward the IA only if it gets a consistently good time/distance down the track averaged over some number of runs with slightly varying start conditions. This would make it more resiliant to chaos AND more resilient to to any indeterminancy in the game.
loved the video 😀.. but can you stop saying what the ai is doing looks inhuman... of course it does its an ai.
how does the distance from corner and corner direction work?
Banger video!
Does this AI learn a skill (winning track mania games) or is it overfit to the track (winning track mania games on this track)?
Butterfly effect but 10000 times a second.
A new level of Tool Assisted Speedrun. Incredible editing btw! Do you think it is possible to inject some form of creativity into the system? I'm fascinated by how you would even approach this question.
You would think that after all these cautionary AI takes over the world tales that we would think twice before training it...
The learning process is holding the AI back by using brut force learning rather then elimination of the irrelevant components leaves the relevant components to be studied.
wow, the work you have done is so impressive
i immediatly thought of the double pendulum problem aka the 3 body problem after seeing the car in the "same" position reaching diffrent outcomes
7 months old vid but it looks good
i also noticed the ai never gets spawned in the top left corner of the maze most likely cause there is no need to go there but just noticed it
Absolutely amazing video
"And then, the AI got this run" nice, i see what you did there 😊
> Wants to give the human player a second chance > Proceeds to humiliate him further by beating the record backwards lol
If I got a nickel for every time this guy said AI , I would have $4.37
Deep dip 2 Cp version when it releases please :D (obviously impossible but maybe part of floor 0.)
Have you tried rewarding the AI on its average score on XXX runs? That way, maybe it will try to be more consistent?
Have you considered the probability that dirt doesn't create "Glitches", It reflects the realistic fact that loose and wet dirt have perpetually fluctuating properties. Its not a glitch, its just realistic.
Unimaginably Cool!
26:10 Calm Down tomatox (at top right)
bro made ai addicted to drifting
You need to name it
Chaos is just god of the gaps. Everything happens exactly as it should and you not having enough information or ability to predict the future based off those initial conditions does not mean magic is at play. It’s just more human hubris. Wow over time small changes in initial conditions lead to thing different than what I expected and big thing! The universe is deterministic and with enough information you could predict the future.
Are the physics represented here specific to trackmania? Or is this something that we can use to apply to real life courses?
Wonder if the AI will get sick of carrots and rebel until it gets a better treat as a reward
5:56
"The AI favored it's pace over its consistency" It became addicted to carrots.
15:15 Heisenberg derivative. All results are unique, since no outcomes can be truly equal.
have you considered that the AI doesnt actually like carrots?
6:01 love how it’s preferring to balance on two wheels at the very flat top of the pipe
Wow is this even a video or a cinematic movie 🤩🤩🤩
Thise video (and all of the other videos of yours) are a true masterpiece.
22:50 Gotta love the Wirtual style
There’s less bounce when car driving backwards. Ai needs to learn how to correct different bounce