/T-PDG

Primary LanguageJupyter NotebookGNU General Public License v3.0GPL-3.0

T-PDG

Transformer-based Powered Descent Guidance (T-PDG), a scalable algorithm for reducing the computational complexity of the direct optimization formulation of the spacecraft powered descent guidance problem.

T-PDG Trajectories

The lossless convexification (LCvx) algorithm, which was used for problem training and test data, was adapted from the SCP Toolbox1.

To Run:

  1. Make sure SCP Toolbox is also installed.

  2. Ensure PyCall is installed with the correct Python path and LaTeX is downloaded.

  3. In Julia run include("Tests/run_tests.jl") inside of the T-PDG folder.

To Design New Models, Train & Test, or Visualize Models with t-SNE:

  1. Open Tests/NN_Train_and_Test.ipynb and navigate to the most relevant section for your task.

Included Folders and Files

  • T-PDG
    • src - Contains required files for running the algorithm
      • Data - .pkl files including mean and standard deviations for the datasets are stored here, as well as standardized training and testing data
      • Models - Trained transformer models are stored here
      • Results - Result figures and datasets are saved here
      • Sampling - .csv files sampled from LCvx with tight constraints and optimal final times are stored here
      • definition.jl - LCvx optimization problem created, constraints are added, and the optimization problem is solved
      • parameters.jl - Constructors for setting up Rocket and Solution structures
      • T-PDG.jl - Creates a package from the src files
      • tests.jl - Tests the T-PDG algorithm and compares runtime and feasibility with LCvx
    • Tests - Contains files for running the guidance algorithm and plots

      • NN_Train_and_Test.ipynb - Preprocess data, train and test transformer neural networks, and visualize embeddings using t-SNE

      • plots.jl - Contains all plotting functions

      • run_tests.jl - Run T-PDG using

          include("Tests/run_tests.jl")

Citing

If you use T-PDG in your work, kindly cite the following associated publication.

@article{TPDGSciTech2024,
  year = {2024},
  month = jan,
  publisher = {American Institute of Aeronautics and Astronautics ({AIAA})},
  author = {Julia Briden and Trey Gurga and Breanna Johnson and Abhishek Cauligi and Richard Linares},
  title = {Improving Computational Efficiency for Powered Descent Guidance via Transformer-based Tight Constraint Prediction},
  journal = {{AIAA} SciTech},
  note = {Free preprint available at [https://arxiv.org/abs/2311.05135](https://arxiv.org/abs/2311.05135)}
}

Footnotes

  1. Danylo Malyuta, Taylor P. Reynolds, Michael Szmuk, Thomas Lew, Riccardo Bonalli, Marco Pavone, Behçet Açıkmeşe. "Convex Optimization for Trajectory Generation: A Tutorial on Generating Dynamically Feasible Trajectories Reliably and Efficiently". IEEE Control Systems, 42(5), pp. 40-113, 2022. DOI: 10.1109/mcs.2022.3187542. Free preprint available at https://arxiv.org/abs/2106.09125 ↩