f16-flight-dynamics

F16 Simulation Rendered in FlightGear

Example Air Collision Avoidance Scenarios Simulated Using F16 Dynamics Model

Overview

This repo contains an efficient implementation of the aircraft model described in Stevens and Lewis "Aircraft Control and Simulation",3rd Ed., written in C++. The project produces a shared library f16_flight_dynamics as well as the python module f16dynamics that accesses it via python bindings. The python module provides drop-in replacement objects / functions for the same models implemented in AeroBenchVVPython and F16 CSAF.

Citation

This implementation is based on and tested against the model available in AeroBenchVVPython. The model was designed to test autopilot and analysis methods. No claim is made about its accuracy; the F-16 model is based on a common aircraft model with additional controllers placed on top of it.

Heidlauf, P., Collins, A., Bolender, M., & Bak, S. (2018, September). Verification Challenges in F-16 Ground Collision Avoidance and Other Automated Maneuvers. In ARCH@ ADHS (pp. 208-217).

Stevens, B. L., Lewis, F. L., & Johnson, E. N. (2015). Aircraft control and simulation: dynamics, controls design, and autonomous systems. John Wiley & Sons.

Benchmarks

Benchmarks were run on the following scenarios

GCAS Scenario - A GCAS autopilot was installed on a single f16 and placed at low altitude. The simulation is run for 2 minutes.
ACAS Scenario - An ACAS autopilot, being a NN compressed early prototype of ACAS Xu, was installed on a single f16 and placed near an intruder f16 heading towards it (2 f16 plants in total). The simulation is run for 2 minutes.

See this notebook for more.

	GCAS Scenario	ACAS Scenario
F16Dynamics (ours)	0.618 s	1.431 s
CSAF F16 - numba enabled	3.990 s	4.68 s
CSAF F16	4.933 s	10.548 s

Installation

Docker (easiest)

Build the docker at the repo root via

docker build . -t f16dynamics

Note that this build may take a few minutes to get the build dependencies together.

Running without any arguments will put you in a python REPL with csaf and the f16dynamics library

docker run -it f16dynamics

Run a script with

docker run -v $PWD:/mydir f16dynamics /mydir/myscript.py

`f16dynamics` Python Module

pip

The C++ library uses Boost and will not build unless it's installed. A simple way to install boost, using common python tools, is via conda

conda install -c conda-forge boost==1.76

Now clone the repository and install the python module

git clone https://github.com/EthanJamesLew/f16-flight-dynamics.git
cd  f16-flight-dynamics
pip install .

Check that you can import it

python -c "import f16dynamics"

C++ Libraries

Install Boost and Boost NumPy. Use CMake to build

mkdir build
cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make 
sudo make install

Tests

Proper testing is TBD, with the coverage

f16_flight_dynamics C++ unit tests
f16dynamics python unit tests
Testing the C++ dynamics against the VVaerobench implementation and CSAF
Integration testing with VVaerobench and CSAF

Jupyter Notebooks

Navigate to notebooks and launch jupyter via

jupyter notebook

EthanJamesLew/f16-flight-dynamics