Particle Filters
Nonparametric filtering for non-Gaussian beliefs.
Why it matters in robotics
Particle filters are the go-to answer for state estimation when beliefs are multi-modal or non-Gaussian and the EKF/UKF break down, so interviewers use them to probe whether you truly understand Bayesian filtering beyond Kalman. Expect to be asked to walk through Monte Carlo Localization (the global "kidnapped robot" problem), explain the predict-update-resample loop, and reason about practical failure modes like particle deprivation and sample impoverishment. Strong candidates can also discuss resampling strategies, effective sample size, and the compute-vs-accuracy tradeoff of particle count.
Application focus
The same topic, tailored to the robot you're building. Your choice is remembered across the roadmap and every topic.
At a glance
The particle filter recursion: each cycle predicts particles through the motion model, weights them by the measurement likelihood, then resamples to focus particles on high-probability regions.
What to study
- ✓Representing belief as weighted samples vs. a parametric (Gaussian) distribution; sampling from the motion model and weighting by the measurement likelihood
- ✓The predict to update to resample loop and how it implements the recursive Bayes filter
- ✓Resampling: low-variance/systematic resampling, effective sample size (Neff), and avoiding sample impoverishment and particle deprivation
- ✓Monte Carlo Localization (MCL) and Adaptive MCL (KLD-sampling), including global localization and the kidnapped-robot problem
Study by time budget
Pick the path that fits the time you have before your interview.