This repository contains trajectory optimization code developed for the Global Trajectory Optimization Competition 13 (GTOC13) by the team Yume Space.
This work does not reproduce the best-known solution achieved by the School of Aerospace Engineering, Tsinghua University, Laboratory of Astrodynamics, which reached a score of 345.216.
However, this repository explores 3 optimization strategies (for the moment only Method 1 is available) for the GTOC13 problem and demonstrates how to reach a solution scoring around 75 points with the so called 'Method 1' shown hereunder.
The repository compares several methods and is intended as an experimentation toolbox for the GTOC13 problem, it will hopefully be complemented by a detailed guide explaining the implemented techniques and modeling choices.
The problem statement and associated data can be found on the following website.
The solution outputed from this script can be validated against the checker available on the website (an account is needed).
This project covers several approaches divided in 'methods', including:
- Method 1 : Classical Minimization techniques --> Nelder-Mead / L-BFGS-B (heavily relies on scipy.minimize lib)
- Method 2 :Evolutionary algorithms
- Method 3 :Reinforcement learning techniques
As of today only Method 1 is available on this repo ! Work in progress :) !
For the solar sail arcs solving :
- Direct single-shooting methods
- Indirect optimal-control shooting ! Work in progress :) !
GTOC13-Trajectory-Optimization/
│
├── data/ # Problem setup files (GTOC13 configurations)
├── doc/ # Methods description (Markdown format)
├── results/ # Example outputs, plots, logs
├── src/ # Core source code
└── README.md
Under results/ folder are listed some obtained solution. The submission.txt file is the file format requested by the GTOC13 checker and the file has been validated against it. The html file contains a 3D animation of the solutions (do not hesitate to zoom and pan while running the sim but avoid rotational button as they usually makes things messy to watch).
The 75 points solution html file is visible to watch here and can be found in the results folder.
Here is the gif version :
Clone the repository:
git clone https://github.com/floriancal/GTOC13-Trajectory-Optimization.git'
python main.pyDo not hesitate to open main.py to play with the available options (method selection will be placed here in the future).
This project depends on :
Mandatory :
-
The PyKEP library, released under the GPL license, which can be installed from the following PyKep Documentation
Dario Izzo. (2019). esa/pykep: Bug fixes and more support on Equinoctial Elements (v2.3). Zenodo. https://doi.org/10.5281/zenodo.2575462
-
Poliastro
for orbital mechanics
-
Orbital for orbital mechanics
-
Numba to speed up computations.
Additional :
- Plotly : for visualization, Plotly Technologies Inc. Collaborative data science. Montréal, QC, 2015.
- Matplotlib for visualization.
It is better to install pykep via conda as mentionned in the documentation, for the other dependencies pip requirements files are available at the project root requirements.txt
Contributions, suggestions, and improvements are welcome. Feel free to open an issue or submit a pull request.