The Borg Lab

Monte Carlo Localization


To navigate reliably in indoor environments, a mobile robot must know where it is. Thus, reliable position estimation is a key problem in mobile robotics. Probabilistic approaches provide a comprehensive and real-time solution to the robot localization problem. However, the type of representation used to represent probability densities over the robot’s state space is crucially important. In 1999 we introduced the Monte Carlo Localization algorithm, where we represent the probability density involved by maintaining a set of samples that are randomly drawn from it, i.e., a particle filter. By using a sampling-based representation we obtain a localization method that can represent arbitrary distributions. The resulting method is able to efficiently localize a mobile robot without knowledge of its starting location, and it is faster, more accurate and less memory-intensive than earlier grid-based methods.


Related Papers

Thrun S, Fox D, Burgard W, Dellaert F
Artificial Intelligence
Robust Monte Carlo Localization for Mobile Robots, Thrun, Sebastian, Fox Dieter, Burgard Wolfram, and Dellaert Frank , Artificial Intelligence, Volume 128, Number 1-2, p.99–141, (2001)
Dellaert F, Fox D, Burgard W, Thrun S
Monte {C}arlo {L}ocalization for Mobile Robots, Dellaert, Frank, Fox Dieter, Burgard Wolfram, and Thrun Sebastian , ICRA, (1999)
Fox D, Burgard W, Dellaert F, Thrun S
Monte {C}arlo {L}ocalization – {E}fficient position estimation for mobile robots, Fox, Dieter, Burgard Wolfram, Dellaert Frank, and Thrun Sebastian , AAAI, (1999)