Smart AI vs Simple AI
- junlachaktophat
- Jul 15
- 1 min read
Updated: Jul 16
Smart AI vs Simple AI: A Unity Machine Learning Comparison
In this project, I explored the capabilities of Machine Learning (ML) in game development by comparing a traditional rule-based AI (NavMesh) to a learning-based AI agent trained using Unity ML-Agents.
The goal was simple: let the Smart AI (ML Agent) survive and adapt, while the Simple AI (NavMesh Agent) pursued the player using fixed logic. Through multiple training phases and environment iterations, I evaluated how AI can evolve, adapt, and outperform scripted behaviors.
Key Highlights:
NavMesh AI used static pathfinding and tagging logic.
ML Agent used RayPerceptionSensorComponent3D to detect enemies and collectibles.
Reward-based training led to intelligent behaviors: dodging enemies, collecting items, and targeting threats dynamically.
Three Phases of Training:
Phase 1: Basic movement and attack training using random spawn logic.
Phase 2: Added collectible logic to encourage multitasking and reward collection.
Phase 3: Implemented ray perception, camera lock-on, and refined training to stabilize reward and loss metrics.
Tools Used:
Unity 2023.2+
ML-Agents Toolkit (v0.28+)
TensorBoard for monitoring training progress.
Anaconda for running training scripts.
Results:
The ML-trained agent learned to rotate toward enemies smoothly, manage multiple tasks (combat + item collection), and react dynamically to changing environments.
Training graphs showed consistent improvement, proving ML's flexibility and adaptability in gameplay AI.
Future Vision:
Introducing procedural environments to test AI generalization.
Simulating emotion-like behaviors (e.g., retreating when low on health).
Enabling self-play training to evolve two ML agents against each other, inspired by OpenAI's Dota 2 model.
Full Project Documentation (Google Docs)
Comments