===========================
1).Define the elements in the state.
2).Path Generation
This func produce the series of state. e.g. x=[...], x_dot=[...], x_dot_dot = [...], theta=[...] .......
3).Get the raw measurement data
Generate the noisy measurement from the perfect actual state. And define the elements in the state.
4).Raw measurement data-----> Estimate data
5).Implement the dynamics of the machine(robot/drone) to predict next state.
| Known | What we want |
|---|---|
| Thrust/Angular velocity | Acceleration |
6).Controllers use different kinds of data. And simulate the states of the machine.
The controllers may be a PID controller or sth similar, but the data they use are different. So We can understand the influence of estimation.
Controller 1: PID_controller_with_measured_values
Controller 2: PID_controller_with_estimated_values
Only use measurement. Simply just the average ---or--- Recursive average (which is a kind of weighted average).
Not only use measurement, but also use the control to predict the state. Bayes Filters are introduced.
Kalman Filter. Gaussian + Matrices.
Extend Kalman Filter(EKF). Gaussian + Nonlinear-----Linearize----->Make it linear and use Kalman Filter.
Unscented Kalman Filter(UKF). Gaussian + Nonlinear-->Don't need the linearization step.
Particle Filter. Not Gaussian + Nonlinear
State means several different things: and in the course they are mentioned like this:
1).ture state / acutal state
2).measured state / measurement
3).estimate state / estimation /belif
4).predicted state / predict / belif_bar
In coding, for Kalman Filter how to represent the state?
Kalman Filter: state is a Gaussian, just use mean and covariance.
here state mentioned means: estimated state and predicted state, and in the course pseudo-code they use the symbol bel and bel_bar respectively.
For the pseudocode, how to reprsent state
Simplified procedure of Bayes Filter:
Simplified procedure of Kalman Filter:
1). From control predict next state.
2). Receive measurement from
****