Practical Extensions to Vision- Based Monte Carlo Localization Methods for Robot Soccer Domain Kemal Kaplan, Buluç Çelik, Tekin Meriçli, Çetin Meriçli ve H. Levent Akın Boğaziçi University, 2005
Basic Monte Carlo Localization 1.Quantize Environment 2.Initialize beliefs 3.Update beliefs 4.Resample 5.Mutate particles
Vision Based MCL •Use information from vision to update particles
Robocup Field Setup
Considering Number of Percepts Seen •One landmark with 0.75 confidence vs 4 landmarks with 0.9 confidence 0.9 x 0.9 x 0.9 x 0.9 = < 0.75 !!! •Proposed correction
Using Inter-Percept Distance
Using Inter-Percept Distance (II)
Using Inter-Percept Distance (III)
Variable-Size Number of Particles
Dynamic Window Size
Results
Reults (II)
Questions?