Tackling the Premature Convergence Problem in Monte-Carlo Localization
An example of premature convergence
in Monte Carlo localization. The
particle population is prematurely
converged to the wrong solution.
Within the field of genetic algorithms, a number of techniques exists that form a solution to the premature convergence problem, the so called niching methods. We implemented a few of these methods on a particle filter for Monte Carlo localization (i.e., crowding, sharing and local selection).
The experiments show a significant improvement in the diversity maintaining performance of the particle filter. The results support our claim that much can be gained from cooperation between the fields of particle filters and genetic algorithms.
DemoYou can experiment with the presented niching methods in this Java applet. Contact me if you are interested in the code.
- Kootstra, G. & de Boer, B. (2009) Tackling the Premature Convergence Problem in Monte-Carlo Localization. Robotics and Autonomous Systems 57(11): 1107-1118. Excepted manuscript: doi:10.1016/j.robot.2009.07.003. not available