Popular Courses
Beginner – Linear Regression
Introduction
Beginner – Decision Trees
Introduction
Beginner – Neural Networks
Introduction

Sign in to Curvingo

or

New Account     Reset Password     

Plans

Details
Reinforcement Learning
Introduction
Simulation
Line World
Grid World
Where Do We Go From Here?
Code with Python
Summary
Curvingo · Made in India

Share

Send us Feedback

Please let us know if you have any feedback and suggestions, or if you find any errors and bugs in our content.

Sorry, your message couldn’t be submitted. Please try again!

Thanks for your feedback!

Reset Progress

This will delete your progress and chat data for all chapters in this course, and cannot be undone!

Glossary

Select one of the keywords on the left…

Reinforcement LearningGrid World

Reading time: ~5 min

The lane had 5 states and 2 actions. Real problems are bigger. In Grid World, Sina lives on a 5 × 5 board. She starts in the top-left corner, the sweet shop (+10) is in the bottom-right corner, and there is a puddle (−10) in the middle. She can step up, down, left or right; stepping into a wall or off the board leaves her where she is. That makes 25 states with 4 actions each — a Q-table of 100 numbers, filled in by exactly the same update.

You design the world. Complete the three missions.

Each training episode drops Sina into a random empty cell, so she learns good arrows everywhere, not only along one path. She never sees a map. She bumps into walls, collects zeros, and now and then reaches the puddle or the sweet shop — the arrows are everything she learned from that experience.

When the World Changes

After training, you put a new wall right on Sina's path and send her off without training again. What happens?

She sees the wall and walks round it
She follows her old arrows and bumps into the wall
Her arrows update themselves at once

A policy is only as good as the world it was learned in — try it in the grid above. When the environment changes, the agent has to keep exploring and learning, which is why real RL systems often keep a small ε switched on for good.

Sina