AI Learns To Park Vs 2 Humans

utm_id=integration Try making your own Parking AI: Competitor Version (Free): https://discord.gg/pRC88zP3W4 Supporter (has additional features): https://www.patreon.com/posts/game-car-parking-154650872 If you're interested in being a part of future projects like this: https://forms.gle/xgQTp9pubqQsNnoY9 This project uses a custom visual scripting system that works at runtime. It was built using Unity but functions just like a normal game. The game developers used this system to program behavioral ai for their cars (GOFAI or good old-fashioned artificial intelligence). The AI uses Unity's MLAgent package for machine learning. The package uses PPO or proximal policy optimization to train the neural network. (fun fact... PPO was developed by OpenAI and is what ChatGPT uses to train) The agents are rewarded when they do well and punished when they do something they shouldn't. Over time the neural network learns how to optimize for rewards. The machine learning also requires observations about the world. For this AI... Observations... -car's relative position to the parking stall -current velocity of the car -hit information from 8 spherecasts (sensors) placed around the car -car's distance to the parking stall (pathfinding algorithm was used to measure the distance but not for driving) Rewarded for... -progress towards the parking stall -all four tires being inside the parking stall -large reward after being fully in the stall for 3 seconds -the quickness it's able to successfully park Penalized for... -timeout (taking too long) -running into parked cars or other obstructions Similar Videos: https://youtu.be/tfg_Ay1FSe4 https://youtu.be/m2CDStGTXr8 https://youtu.be/jWJLcI7DRDw https://youtu.be/5a5O8yQpQu8 https://youtu.be/5ymw0v7Wz3A https://youtu.be/VMp6pq6_QjI https://youtu.be/r428O_CMcpI https://youtu.be/SX08NT55YhA https://youtu.be/nZ80P6ba9i4 https://youtu.be/Aut32pR5PQA