Reinforcement Learning

The demonstration shows how a computer masters the game Flappy Bird using reinforcement learning, a machine learning paradigm. This is an approach in which an algorithm learns to perform or avoid certain actions through rewards (or punishments).

What is reinforcement learning?

Reinforcement learning (RL) is an area of machine learning that deals with how machines can learn complex tasks without receiving explicit instructions. For relevant actions, the algorithm receives feedback in the form of a reward or punishment. Based on this feedback, it adjusts its strategy. The goal is to develop a strategy through trial and error that maximizes the accumulated reward over time.

Where is reinforcement learning used?

Reinforcement learning has applications in science and engineering. In robotics, it enables machines to autonomously learn and perform complex tasks, from simple grasping of objects to navigation in unknown environments. In the context of computer games, it enables the integration of behavioral creatures that can adapt to new environments.