AI Learns to Play Dodgeball

AI vs AI Playing Dodgeball!
If you want to learn more about AI and deep reinforcement learning (how Albert is trained), there are amazing courses teaching those exact concepts on Brilliant! You can use my link to get a free 30 day trial with 20% off! I've personally gone through the course "Introduction to Neural Networks", and it's one of the best courses on Neural Networks I've ever seen. They're paying us to promote them, but they're genuinely a great service, I've had a Brilliant account for over 5 years and can't recommend it enough :)

In this video two AI Warehouse agents named Albert and Kai learn to play Dodgeball against each other. The AI were trained using Deep Reinforcement Learning, a method of Machine Learning which involves rewarding the agent for doing something correctly, and punishing it for doing anything incorrectly. Albert and Kai's actions are controlled by Neural Networks that are updated after each attempt in order to try to give Albert and Kai more rewards and less punishments over time

Thank you for watching :D

Current Subscribers: 545,807

[aiwarehouse]

To learn more about the AI in the video, keep reading!

These agents we're quite tricky to train. Despite the task being simple, we actually had to do a good amount of reward shaping in order to get interesting behavior. The agents are punished for dying, being in the air (to discourage spam jumping, which was the bane of my existence for a while), the ball being on the agents side, the ball not being held and being close to the ball while the ball is lethal. The agents are also rewarded for hitting the opponent with a throw, picking up the ball for the first time (otherwise they would just push the ball back and forth), throwing the ball while in the air (to get the speed boost) and being close to the ball while it's ready to be picked up.

It was a massive challenge training the agents in the final room, largely due to the fact that it seemed as though no matter what we did, the Kai's would just spam jump when there were more than a couple balls on their side, finally Andrew had the great realization the reason for that is to try to avoid the friendly fire that happens when you have 10 agents on one side lol, so we disabled friendly fire on the final battle and tried resuming the training considering we already had 50m steps trained, the behavior wasn't perfect but it looked pretty cool and we've already kept you guys waiting for a video for too long we decided to use it! I hope it meets your guys standards!

Tyler built the demo you can download on our discord server and it's absolutely awesome, you get to play as Albert and try to beat Kai, who has the brain that was trained in the first room and is REALLY hard to beat ahaha, make sure to check it out! We're hoping to release a demo like this for each video we upload so the wait between videos doesn't feel as long

These videos are really challenging to make, especially quickly, which is why there are so few videos on this channel, but thanks to our team and support from our community discord server (seriously you should join it) we're really on our way to uploading monthly, I'm 99% sure we'll get another video out to you guys within a month!


Thank you so much for watching:D

[marcorocchi]

7:43 "you suck, Albert."
*Albert dodging three balls at once, catching one and still back flip dodging:*

[Pumble]

Where is the part of the video when you teach Albert to say GG EZ??? It's an amazing video, very fun. I wonder how it would look like a 3v3 match. In my job, I was creating a machine learning model(I know it's something different), but I noticed that when you change the environment, the models need to learn almost everything again, as you showed in your video. I know you spent a lot of days working on this; you did it great. Thanks for the content; it's amazing.

[chickenyouate]

For 2 years straight, I’ve been here within a minute of every single video upload. I don’t know why I’m so enamored with this cube named Albert but I can’t get enough.

[mrboost]

I waited about 6 months for this and it was more than worth the wait. Epic video man.

[josephhenderson]

I love that Kai, with fewer brain cells, decided the best course of action was "Throw rock harder."

*And it worked*

[ChaoticCats]

7:46 “but there’s nothing that being trapped and punished can’t fix!”wow, so motivating! 😊

[Lavollager]

I adore those bandanas that Albert and Kai are wearing, it adds so much damn personality to the orange and blue rivals!

[rdtyphon]

1:04 He's just like my ranked teammates!

[donutlover]

Why does Kai get a trophy while Albert has been denied his reward in every other video 😭

[Reptayl]

4:38 Man not only kai learned how to throw and catch, he learned to emote aswell

[plastercyn]

0:06 “they will learn to play dodgeball! 😁”
*brutally mutilates them*

[infoplayer-kuz]

Albert was getting absolutely bullied...

[nathan]

I appreciate the way you present sponsors in a way that won’t disrupt the flow of the video. It’s a nice change of pace.

I’m tired of seeing content creators awkwardly squeeze in sponsor messages via awkward and cheesy fill ins in the middle or end of video.

[RageLaugh]

2年前にこの動画を見つけてから2人が大好きになりました！
これからも応援しています！

[nicolasn]

"Albert: Jump, Jump, jump
Kai: Jump, Jump, Jump
Both for no reason: Jump, Jump, Jump"
Would this be the equivalent of spamming shift in Minecraft? I just realised they do that everytime over the course of the videos.

[RobbieDeath]

7:43 “You suck, Albert” Bro Albert winning 36% of the games even though it’s a 3v1 it’s obvious Kai’s the favorite

[Tylorean]

Machine learning is just like school.
You can’t go home
if you do something wrong, you get punished
If you start to even remotely understand something, everything changes and gets harder

[xyborg]

Music:
0:00 Both Shoes Gone Walkin' - Moss Harman
1:00 & 9:41 Late For Work - Helmut Schenker
3:40 Blame It on Me - Radio Night
5:52 Jim & the Umbrella - Mike Franklyn
7:23 Taste My Groove - Mike Franklyn
8:35 The Out of Bounce Dance - Stationary Sign

[notPento]

How does this man keep coming up with the most banger things to learn to 2 simple colored cubes and why is it so god damn entertaining to see them fail?!?