01
Localization
Localization estimates a robot's state relative to a known map or reference frame. The state estimate incorporates uncertain motion and sensor measurements.
P(x_t | z_{1:t}, u_{1:t})02
Bayesian Filtering
- Kalman filters provide optimal linear-Gaussian state estimation.
- Extended Kalman filters linearize nonlinear models locally.
- Particle filters represent complex probability distributions with weighted samples.
- Factor-graph approaches represent relationships among states and measurements.
03
Simultaneous Localization and Mapping
SLAM addresses the coupled problem of estimating robot pose while simultaneously constructing a map from observations.
The difficulty arises because localization depends on the map while mapping depends on knowing where the robot was when each measurement was taken.