AI FIGHT WITH CARS

Neural Network controlled Cars are fighting. This game was programmed using Unity3D with Reinforcement Learning (Machine Learning) using the ML-Agents Framework. The Algorithm used to train this A.I. is called PPO and it was developed by OpenAI. My name is Sebastian Schuchmann from Germany and I hope you enjoyed this video.
If you want to support me! :)
Patreon: www.patreon.com/user?u=25285137
Twitter: / sebastianschuc7
If you want to get in contact with me, go to my Webform :)
www.sebastian-schuchmann.com/...

[@opokusm]

Nice video, you could probably avoid cars going in circles by randomizing their positions at the start, otherwise each car learns to constantly compensate for opponent pushing them at one direction.

[@Timotheeee1]

when I made an ai that can park a truck it also had a tendency to spin in circles

[@joshuamanasye]

YESS!! I WANNA SEE MORE!
But keep all the funny ai fails part too while learning

[@akulvohra5344]

This is surprisingly high quality production, keep it up!

[@SebastianSchuchmannAI]

Thanks! Cool to hear, I try my best

[@eL3ctric]

would youd mind sharing some od the behind the scenes?

[@SebastianSchuchmannAI]

Yeah for sure. I try to keep the technical stuff to a minimum in the videos. This whole thing runs in a 3D game engine called Unity3D. It is really powerful. The Unity guys have a Machine Learning Framework called ML-Agents, which I use for most of my stuff. It implements algorithms like PPO (this is the same stuff OpenAI uses). I am currently in the process of moving more of my training over in the cloud to speed up the production time.. but of course this comes at a cost. Hope that answers some of your questions! :)

[@aigen-journey]

@@SebastianSchuchmannAI That wasn't really too specific :) What's the reward structure and how did you set-up those asymmetric agents?

[@mohammedthousifshaik3628]

just wow!

[@michelcusteau3184]

Hey, very cool videos. What do you use to perform all these simulations?

[@SebastianSchuchmannAI]

Unity3D :)

[@Adama_Now]

very cool! I wish I knew how to do something like this.

[@SebastianSchuchmannAI]

You can learn it, no problemo 😄

[@tylerk3130]

Congratulations, your AI invented the PIT maneuver without knowing wtf that is. Very cool, easy choice to sub. I love the simplistic art style of your worlds. Are you open sourcing the environments? I'm not much of an RL guy myself, but I would love to see someone come up with a demolition derby type multi agent game where each player has different motivations.

[@SebastianSchuchmannAI]

In General I am a big fan of open sourcing my environments but some of the assets like the cars are bought so it is not possible in this case

[@tylerk3130]

@@SebastianSchuchmannAI Ah, that's a real shame, but understandable. Looking forward to more great RL content :)

[@robosergTV]

Did you use just two brains? I found from my tests that without a league like what Deepmind did for Alphastar if you train just two agents in a competitive setting they will perpetually invent new strategies and forget the old ones.

[@SebastianSchuchmannAI]

Yes, two brains. I really don't think my results are great and they would probably forget the old ones, too. How did you manage to create League setting? Did you just create your own?

[@philhiett5027]

Great AI videos..... but I wish you wouldn't drop a lone F-Bomb in each one... it makes it hard to show these to my students!

[@SebastianSchuchmannAI]

Good to know. I'll try to remember for the next one :)