Split-Grid RL

Batch Size 20
Exploration 0.2

Worker Progress

EPISODES 0
TOTAL STEPS 0

Guided Walkthrough

  • 01. Task Decomposition: Notice the maze is split into three grids. Grid 1 focuses on finding Point 1, Grid 2 on Point 2, and Grid 3 on the Goal.
  • 02. Parallel Training: The green flashes represent "Batch Processing." All three sub-brains learn their respective paths simultaneously.
  • 03. Policy Transition: When the white agent reaches a checkpoint in one grid, it "hands off" control to the next brain, moving to the next grid.

Technical Summary

HIERARCHICAL REINFORCEMENT LEARNING

By partitioning the mission into sub-goals, the state space complexity is reduced from exponential ($S \times 2^n$) to linear ($n \times S$), enabling near-instant solving.

BATCH REPLAY EFFECT

Setting the "Batch Size" high allows the engine to run hundreds of hypothetical simulations per visual update, accelerating convergence for "Big Data" state spaces.

This simulation demonstrates how complex AI problems are broken down into isolated, manageable "mini-environments" to find optimal solutions at high speeds.