GPT Codex Pac-Man Part 2 | AI Learns to Play Pac-Man with Reinforcement Learning

#chatgpt #coding #aiagent #openai #codex #reinforcementlearning
AI Learns Pac-Man with Codex, GPT & Reinforcement Learning!
In this video, we train an AI to play Pac-Man using GPT + Codex, show how we build a Python reinforcement learning training script, export the model to ONNX, and finally run it inside a browser game you made yourself.

We walk through the full pipeline — from training with Gymnasium and a DQN agent to exporting the model, loading it in JavaScript using onnxruntime-web, and watching the AI control Pac-Man. Whether you’re into game AI, Python ML tutorials, GPT/Codex workflows, or ONNX deployment, this video breaks it down step by step.

🚀 What You’ll Discover:

• How to train an AI agent for Pac-Man gameplay
• Using Reinforcement Learning (DQN) in Python
• Exporting a trained model to ONNX format
• Loading and running the ONNX model in your browser game
• How game state feeds into the model and actions drive game behavior

00:00 - Introduction and Overview
00:59 - Phase 6 Prompt (ONNX load + Human/AI toggle)
04:51 - Phase 6 Implemented (UI + status)
05:43 - Phase 7 Prompt (Gymnasium env + DQN)
09:11 - Phase 7 Agent output review
10:06 - Phase 7 Setup (virtual environment + install dependencies)
13:06 - Phase 7 First training run (no logs)
13:52 - Phase 7 Training run #2 (logs explained: reward, average reward, epsilon)
13:37 - Phase 7 Evaluating and Exporting the training model (errors)
15:10 - Phase 7 Evaluating and Exporting the training model (fixed)
16:05 - Testing the exported ONNX model (failed)
16:20 - Testing the final exported ONNX model (passed)