PARTICLE FILTER

Sequential Monte Carlo robot localization: belief distribution, resampling, motion & measurement update
SIR filter: (1) Propagate particles through motion model. (2) Weight by likelihood. (3) Resample. Effective sample size = 1/Σw².
Blue particles = belief distribution. Orange dot = true robot. Green = landmarks. Size = weight.
Effective Sample Size (ESS) over time. Low ESS triggers resampling.