Model-based Reinforcement Learning (Glossary)

Model-based Reinforcement Learning is a term that is particularly at home in the fields of Artificial Intelligence, Automation, and Industry 4.0. It involves machines or computer programs learning to make decisions independently – based on an internal model of the environment.

Unlike other machine learning methods, model-based reinforcement learning not only observes the actual outcomes of its decisions but also builds its own „model of the world“. Using this model, it can try out actions beforehand before actually executing them. This allows the system to learn more quickly and efficiently because not every possibility needs to be tested in real life.

A vivid example: Imagine a robot in a factory that is to learn how to stack packages as efficiently as possible. With model-based reinforcement learning, the robot first builds up an idea of how packages behave when stacked. It can play through different strategies in the model and find out which one works best. Only when it has found a good method does it implement it in real life. This saves time, resources, and reduces errors.