NANO-filter 0.1.0

Nonlinear Bayesian Filtering with Natural Gradient Gaussian Approximation

Natural Gradient Gaussian Approximation (NANO) Filter

This is the official python code for the paper "Nonlinear Bayesian Filtering with Natural Gradient Gaussian Approximation". You can find the preprint of the paper in Link.

Please contact the corresponding author of the code at [e-mail] or [e-mail].

Index

1. Installation
2. Example: Wiener Velocity Model
3. Example: Air-Traffic Control Model
4. Example: Unmanned Ground Vehicle Localization

1. Installation

1. Clone the repository.

git clone https://github.com/TianyiMehdi/NANO-filter.git

2. Install the dependencies.

pip install -r requirements.txt

2. Example: Wiener Velocity Model

Model:

Test:

cd experiments
# 1. For system without outlier
python wiener/wiener_NANO.py \
    --measurement_outlier_flag False \
    --n_iterations 1 \
# 2. For system with outlier, and you can try to change the loss_type and loss hyperparameters to see the difference 
python wiener/wiener_NANO.py \
    --measurement_outlier_flag True \
    --loss_type beta_likelihood_loss \
    --beta 9e-4 \
    --n_iterations 1 \

Figure:

3. Example: Air-Traffic Control Model

Model:

Test:

cd experiments
# 1. For system without outlier
python air_traffic/air_traffic_NANO.py \
    --measurement_outlier_flag False \
    --n_iterations 1 \
# 2. For system with outlier, and you can try to change the loss_type and loss hyperparameters to see the difference 
python wiener/wiener_NANO.py \
    --measurement_outlier_flag True \
    --loss_type beta_likelihood_loss \
    --beta 2e-2 \
    --n_iterations 1 \

Figure:

4. Example: Unmanned Ground Vehicle Localization

Model:

Test:

cd experiments
python ugv/ugv_NANO.py \
    --loss_type beta_likelihood_loss \
    --beta 2e-2 \
    --n_iterations 1 \

Figure: