Key Engineering Takeaways
- •AMCL represents robot pose uncertainty as a cloud of weighted particles [x, y, θ].
- •KLD-sampling reduces particle count when the robot is confident in its position, slashing CPU usage.
- •AMCL broadcasts the essential `map -> odom` transform to correct for cumulative wheel odometry drift.
Prerequisites
- • Probability basics
- • Occupancy grid maps
Probabilistic Particle Filter Theory
Monte Carlo Localization (MCL) uses a recursive Bayes filter where each particle represents a hypothesis of the robot pose. As LiDAR measurements arrive, particles that align with the known map walls receive higher weights, while conflicting particles are eliminated during resampling.
Tags:#AMCL#Particle Filter#Localization#Nav2#Likelihood Field#KLD-Sampling