Reinforcement Learning Onramp Overview

Reinforcement learning is a machine learning technique for control problems-- that is, applications where you want to choose an action to achieve a particular goal. When should you file the rockets to land your probe gently? How much torque should you apply on each joint of your robotic hand to have it pick up an object?

Should you turn on the burner or open the exhaust vent to keep your hot air balloon steady? Should you steer right or left to keep your car in its lane? The goal of reinforcement learning is to build a smart controller that will choose the best action to take in a given situation.

Reinforcement learning is a machine learning technique because the choice of action is learned from experience rather than being specified directly by human designer, as it would be with traditional controls methods. It's therefore best suited to situations where you can run a lot of experiments quickly and cheaply, which often means using a simulation rather than a physical prototype and where traditional controls techniques have a hard time because you're trying to control a whole complex system with an overall goal, not just an individual component with a specific purpose.

RL is different from other common machine learning tasks such as clustering or a classification. It's unique because you are trying to achieve a measurable goal, but you can't label or score any individual action. That means you have to let things play out. Try different actions in different situations. See how well you achieve the goal and learn from these experiences to find the best action in any given situation.

In this course, you'll learn how to use MATLAB and Simulink to create an RL agent for a robot trying to maneuver between the shelves in a warehouse. To get started, all you need is a web browser. You'll interact with a web-based version of MATLAB where you'll receive step-by-step instructions and instant feedback. And you can also experiment and try things out on your own.

You don't need to know Simulink or learn a lot of theory to start doing reinforcement learning. But it will help if you know a little bit of MATLAB, just the basics. If you've never used MATLAB before, it's easy to get started, and we recommend you first take MATLAB Onramp to get you up to speed quickly.

This course should take about two hours to complete, but you can leave any time and come back later. And when you're done, you can download a shareable certificate of completion. Click the Launch button to get started today.