About

Introduction.txt

Introduction

Imagine teaching a dog to fetch a ball. At first, the dog has no idea what “fetch” means. But over time, with a lot of practice and some treats as rewards, the dog learns that fetching the ball leads to a treat. This back-and-forth learning through rewards is similar to how a reinforcement learning agent learns to make decisions in complex environments.

In this section, we’ll break down the basics of reinforcement learning.

elementary_ideas.txt

Agent and the Environment

Agent: The learner or decision-maker. In the Mountain Car problem, the agent is the car you controlled.
Environment: Everything the agent interacts with. In the Mountain Car problem, it is the valley and the hills, including the car's starting position, the slope of the hills, and the goal at the top of the hill.

A reinforcement learning agent learning to play Donkey Kong. The game Donkey
Kong is the environment here. The player is the learning agent.

Actions and Rewards

Actions: Choices the agent makes. For the Mountain Car, these are movements like “forwards” or "backwards.”
Rewards: Points given based on the agent’s actions. The agent learns by trying to maximize its rewards. For instance, in Mountain Car, every step away from the goal cost the car points (a penalty of -1), so the goal was to find a quick, efficient way to reach the flag.

information.txt

Did You Know? Rewards can be positive (for good actions) or negative (for mistakes). This helps guide the agent to learn better choices over time!

Trial and Error: Learning by Exploration

In reinforcement learning, an agent doesn’t start with all the answers. It tries different actions, observes the rewards, and gradually learns which actions are more effective.
This trial and error process is called exploration, and it’s essential because it allows the agent to discover actions that help it succeed.

interactive_quiz.txt

Think back to the challenge. What did you do when you first started?

Adjusted your strategy to reach the flag faster.Watched how the car movedTried random keys

Thank you! Your submission has been received!

Oops! Something went wrong while submitting the form.

the_learning_loop.txt

Putting It All Together: The Learning Loop

Reinforcement learning follows a basic cycle:

Observe the environment (Where is the car? What and where is the goal?).
Choose an action (Forwards, backwards, or stay).
Receive feedback (How effective/ relavent was the action to my goal?).
Learn and adapt based on rewards (Choose more helpful actions over time to get closer to the flag).

video.mp4

Watch a Demo: See how an RL agent learns to solve an obstacle course.

To see an interactive animation of the agent-environment interaction click the "Interactive Animation" button

Interactive Animation

real_life_examples.txt

Real Life Applications of RL

1. Self-Driving Cars

Self-driving cars use reinforcement learning to make decisions on the road. RL algorithms help cars "learn" how to drive by giving them rewards for safe driving actions (like staying in the lane, slowing down at red lights) and penalties for risky ones (like veering off the road). Over time, they improve their ability to make safe, smart decisions.

Interactive Animation

2. Robotic Control

Robots often use reinforcement learning to train to perform tasks efficiently, getting rewards for successfully completing a task and penalised for mistakes (like coliding with obstacles or taking a long route). With training, robots learn to be faster and more accurate.

game.js

You are a warehouse robot (blue) tasked with moving a package. Reach the goal (green) in minimum number of steps without colliding with the obstacles (red).

Reward System:
-1 for each step.
-10 for bumping with obstacles.
+100 for reaching the goal.

‍