Скачать или смотреть I Trained an AI to Dodge Falling Objects (It Took 4 Tries)

I built a simple game where an AI has to dodge falling blocks... and it turned into a much longer project than I expected.

Over 4 iterations, I tweaked the environment, rewrote the reward system (a lot), and slowly taught the agent to go from wall-hugger to actual dodger.

This is a silly but honest look at how training an AI actually goes. Bugs, weird behavior, and the occasional success. Hope you enjoy :)

🧠 Project uses: Python, pygame, Stable Baselines3 (PPO)
🎮 RL environment: Custom 1D dodge simulator
🧪 Training: 1M+ steps, reward tuning, model saves via callbacks

📌 Related:
GitHub (coming soon?)
Discord Server:   / discord  
Music Used:
  / juno-1  
  / we-need-mirrors  

#AI #MachineLearning #ReinforcementLearning #IndieDev