Hierarchical Monte-Carlo Localisation Balances Precision and Speed

Abstract

Localisation is a fundamental problem for mobile robots. In dynamic environments (robotic soccer) it is imperative that the process be very efficient. Techniques like Monte-Carlo Localisation or Markov Models have been shown to be effective in dealing with partial recognition of landmarks, errors in odometry and the kidnap problem. But they are particularly CPU intensive. However, many times decision-making does not need high accuracy, and thus, we have developed a hierarchical version that allows us to balance real-time efficiency of computation with precision in localisation.