02. Overview

Markov Localization and the Kidnapped Vehicle Project

The localization module culminates in the Kidnapped Vehicle Project. In that project our vehicle has been kidnapped and placed in an unknown location. We must leverage our knowledge of localization to determine where our vehicle is. The Kidnapped Vehicle Project relies heavily on the particle filter approach to localization, particularly "Implementation of a Particle Filter," an upcoming lesson. This leaves the question; How does Markov Localization relate to the Kidnapped Vehicle project?

Markov Localization or Bayes Filter for Localization is a generalized filter for localization and all other localization approaches are realizations of this approach, as we'll discuss later on. By learning how to derive and implement (coding exercises) this filter we develop intuition and methods that will help us solve any vehicle localization task, including implementation of a particle filter. We don't know exactly where our vehicle is at any given time, but can approximate it's location. As such, we generally think of our vehicle location as a probability distribution, each time we move, our distribution becomes more diffuse (wider). We pass our variables (map data, observation data, and control data) into the filter to concentrate (narrow) this distribution, at each time step. Each state prior to applying the filter represents our prior and the narrowed distribution represents our Bayes' posterior.

Bayes' Rule

If you'd like a reminder about how Bayes' rule works, make sure to go back and watch Sebastian's Bayes' rule video from the Localization Overview lesson!