dyne

dyne is a Python package containing a collection of algorithms for state estimation in dynamic systems. Specifically it deals with discrete-time stochastic dynamic systems of the form:

$$\begin{gather} X_{k + 1} = f_k(X_k, W_k) \\\ Z_{k} = h_k(X_k) + V_k \end{gather}$$

With

• $k$ - integer epoch index
• $X_k$ - state vector
• $W_k$ - process noise vector
• $Z_k$ - measurement vector
• $V_k$ - measurement noise vector
• $f_k$ - process function
• $h_k$ - measurement function

The system can be depicted as a chain graph:

The task is to estimate $X_k$ given the known process dynamics and measurements.

Package content

At the moment the following estimation algorithms are implemented in dyne:

Algorithm	Function	Causal?	Requires Jacobian?
Linear Kalman Smoother	run_kalman_smoother	No	Not applicable
Extended Kalman Filter	run_ekf	Yes	Yes
Unscented Kalman Filter	run_ukf	Yes	No
Moving Horizon Filter	run_mhf	Yes	Yes
Batch Nonlinear Optimizer	run_optimization	No	Yes

The file examples.py contains example problems which can be setup and run by all available algorithms.

Installation

The package is pure Python and can be easily installed on any platform. Runtime dependencies include (versions in parentheses were used for the latest development):

• numpy (1.25.2)
• scipy (1.11.3)

To perform a regular install, execute in the cloned repository directory:

pip install .

To perform an editable (inplace) install:

pip install -e .