| Testing | Coverage | Documentation |
|---|---|---|
 |
 |
 |
GaussianFilters implements methods to define and run Kalman, Extended Kalman, Unscented Kalman, and Gaussian-Mixture Probability Hypothesis Density Filters on simulated data. It also implements simulation functions for the Kalman-class filters.
The documentation for the package can be found here: https://sisl.github.io/GaussianFilters.jl/latest
GaussianFilters can be installed by running:
using Pkg
Pkg.add("GaussianFilters")Basic usage follows along defining appropriate models, constructing an appropriate filter, and running the filter with known actions on some measurement data.
using GaussianFilters, LinearAlgebra
# dynamics model
A = [1 0.1; 0 1]
B = [0; 1]
W = [0.5 0; 0 0.5]
dmodel = LinearDynamicsModel(A, B, W)
# measurement model
measure(x, u) = LinearAlgebra.norm(x, 2)
V = [0.01]
omodel = NonlinearObservationModel(measure, V)
# filtering given some action and measurement
ukf = UnscentedKalmanFilter(dmodel, omodel)
b0 = GaussianBelief([0, 0], [1 0; 0 1])
b1 = update(ukf, b0, action, measurement)Covariances may be passed as any AbstractMatrix, or in identity form
(I(n), s*I(n)). GaussianBelief also accepts a bare UniformScaling,
inferring the size from the mean vector:
b0 = GaussianBelief([0.0, 0.0], 2.0*I) # Σ becomes 2*I(2)See the examples/ directory for full demonstrations.
Each example script lives in examples/ with its own Project.toml.
Run any of them with:
julia --project=examples examples/kf_2d_motion.jl| Script | Filter | Description |
|---|---|---|
kf_2d_motion.jl |
KF | 2D point-mass with linear dynamics |
ekf_spinning_satellite.jl |
EKF | Rigid-body rotation with saturated observations |
ukf_nonholonomic_robot.jl |
UKF | Differential-drive robot with range observation |
gmphd_surveillance.jl |
GM-PHD | Multi-target tracking |
gmphd_aircraft_carrier.jl |
GM-PHD | Aircraft carrier scenario |
gmphd_tests.jl |
GM-PHD | Visualization helpers |