How do we estimate internal, hidden state variables through noisy measurement data?
The solution to state estimation is related to the specific Motion Equation and Observation Equation of a robot and the noise probability distribution. Based on the motion/observation equations, and whether the noise is Gaussian, state estimation is divided into linear/non-linear and Gaussian/non-Gaussian systems.
Linear Gaussian System (LG)
- Simple
- Unbiased optimal estimation can be given by Kalman Filter
Non-linear non-Gaussian (NLNG)
- Complex
- use Extended Kalman Filter and Non-linear Optimization