Implementing Machine Learning the Unity way
summary
Shows how Unity’s ML‑Agents uses TensorFlow Python for external training and exports a .bytes model that is reimported as an internal brain. Demonstrates concrete examples: a balancing platform with +0.1 per step and -1 on failure, flying cubes, and a roguelike with four small arenas and mixed heuristic and trainable agents feeding one brain. Uses PPO‑style hyperparameters, trains at 100x time scale, handles discrete vs continuous inputs, and the imported models work but require many iterations.
AI-written summary · about
chapters (9)
- 0:00Introduction to speaker
- 0:43What is machine learning
- 3:08Machine Learning Agents in Unity
- 9:33Simple ML agent examples
- 15:14Roguelike game with ML
- 21:17Setting up training environment
- 26:57Running the training process
- 32:33Testing the trained model
- 34:01Live demo of the game
related talks
32mMachine Learning for videogames in Unity3D
30mBuilding, designing, teaching and training simulation environments for Machine Learning
10mMachine Learning & Unity ML-Agents update
48mMachine Takeover - Improving Your Game With Machine Learning & AI
59mExtended Machine Learning
32mThe Road to General AI
35mCrowd dynamics, RL and Unity: A Journey
32m..Stories About Industrializing In-Game Machine Learning Solutions