AI Plays Sonic the Hedgehog

I trained an AI to play Sonic the Hedgehog (Sega Genesis) using only the pixels on the screen, with no cheats and no level maps. This is version 6, attempting all 19 acts of the game. It clears Green Hill Zone Act 1 in 39 seconds, beats Dr. Robotnik in Green Hill Act 3 (8 hits, no mercy), and blasts through Spring Yard Zone Act 2. Then Marble Zone humbles it. 😅 🧠 How it works • Reinforcement learning (PPO) – the AI starts out pressing random buttons and is rewarded for moving right, collecting power-ups, hitting bosses and finishing acts • It sees what you see: a tiny 84×84 black-and-white version of the screen, 4 frames at a time • Trained on my home lab server, then deployed to a Raspberry Pi 5 with a Hailo AI accelerator, where it plays live on a monitor 📊 v6 results ✅ Green Hill 1 – cleared in 39 s ✅ Green Hill 3 – cleared, boss defeated ✅ Spring Yard 2 – cleared in 86 s ❌ The rest… let's call them "learning opportunities" ⏱️ Chapters 0:00​ Green Hill Zone Act 1 0:41​ Green Hill Zone Act 2 1:14​ Green Hill Zone Act 3 (Boss!) 2:44​ Marble Zone Act 1 4:03​ Marble Zone Act 2 5:15​ Marble Zone Act 3 5:45​ Spring Yard Zone Act 1 7:02​ Spring Yard Zone Act 2 8:31​ Spring Yard Zone Act 3 9:54​ Labyrinth Zone Act 1 10:48​ Labyrinth Zone Act 2 11:55​ Labyrinth Zone Act 3 13:05​ Star Light Zone Act 1 13:27​ Star Light Zone Act 2 15:02​ Star Light Zone Act 3 17:12​ Scrap Brain Zone Act 1 18:21​ Scrap Brain Zone Act 2 18:52​ Scrap Brain Zone Act 3 19:19​ Final Zone #Sonic​ #AI​ #MachineLearning​ #SonicTheHedgehog​ #ReinforcementLearning​ #RaspberryPi​ #SegaGenesis​ #RetroGaming​ #ArtificialIntelligence​ #AIPlaysGames​ #Python​ #MegaDrive​ ======