npc-ml-agents

Use this skill when the problem is learning behavior, not just running inference. ML-Agents covers environment design, observation/action spaces, reward shaping, training configuration, run monitoring, and exporting a model that Unity can later consume.

When to use this skill

• Training an NPC to navigate, fight, cooperate, or adapt
• Designing observation vectors, action spaces, and episode boundaries
• Shaping rewards without creating easy exploits
• Running PPO training and monitoring reward curves
• Exporting a trained ONNX model and deploying it via unity-sentis
• Resuming or extending an interrupted training run

Instructions

Step 1: Treat the environment as the product