Localisation estimates a robot’s pose from motion and sensor information. A particle filter represents possible poses with weighted samples, allowing several competing location hypotheses to coexist. Motion prediction, measurement weighting and resampling update that representation. Convergence depends on the model, observations and coverage of the hypotheses; it is not guaranteed merely by repeating the updates.
Testing and Real-World Use
This could be tested by replaying a fixed sensor stream with documented noise settings and random seeds, checking whether the particle distribution retains plausible locations through ambiguous observations. Particle filtering is useful when localisation uncertainty cannot be represented well by a single point estimate, while exact reproduction requires the original map and implementation.