Unity Engine

3D environment simulation, physics, rendering, objects.

Agent perceives environment through sensors.

Observations (images, rays)

ML-Agents Integration

Converts Unity observations to numerical tensors for RL.

Sends agent actions back to Unity.

Actions (movement commands)

Python Reinforcement Learning

Algorithms (PPO/SAC) update the policy neural network

based on rewards and state transitions.

Updated Policy Parameters

Agent Behaviour

Learned policy determines agent actions in Unity.