ISOSC

Official repo to store program code, trained models, and data of simulation and experiment results for the paper, Deep Imitation Learning for Optimal Trajectory Planning and Initial Condition Optimization for an Unstable Dynamic System

Requirements

Programs in this project were mainly run on Python. Figures in paper were plot by MATLAB.

• python==3.7

• Tensorflow==1.14.0

• Keras==2.2.4

• CUDA v9.2

• CuDNN 7.4.1

• MATLAB 2018

Folder and file structure

• ini_cond/: folder stores initial conditions (ICs) to run dynamic system simulation in Figure 4.

• ML_data/: folder stores data related to neural network (NN) training.

  • model/: folder stores trained model parameters for $x_{mB}$-Net and $\theta_{B}$-Net by ICs generated from different random seeds (annotated numbers in parentheses), and training and validation loss progress during training phase. Files with "exp" are for real robot experiments, while files with "extreme" are for simulation.

  • simulation_result/: each subfolder with the corresponding figure name contains the simulation result data (.csv), the MATLAB code to plot the figure (.m), and the figure (.fig).

  • other CSV files: feature data (ICs of the dynamic system) and label data (input/output variable trajectories of the dynamic system) to train the NN models. ICs were generated by different random seeds (annotated numbers in parentheses).

Program code structure

• class_libraries.py: class library file to put the class of iterative linear quadratic regulator (iLQR) solver

• Libraries.py: function library file to put all utility functions

• optml_traj_cmprss_test.py: the main program to run all simulations and to generate desired trajectories for real robot experiments

• train_optml_traj.py: program to train the $x_{mB}$-Net and $\theta_{B}$-Net NN models for simulations and experiments

• TrajectoryPlanning.py: functions to generate trajectory points for different criteria given via points

*Note: some data and folder paths in the codes might be changed due to repo reorganized, but data names should be unchanged.

Copy rights

The current license is a temporary license before paper published. All rights reserved.

Citation

@article{chen2024deep,
    author = {Chen, Bo-Hsun and Lin, Pei-Chun},
    title = {Deep Imitation Learning for Optimal Trajectory Planning and Initial Condition Optimization for an Unstable Dynamic System},
    journal = {Advanced Intelligent Systems},
    volume = {6},
    number = {1},
    pages = {2300379},
    doi = {https://doi.org/10.1002/aisy.202300379},
    year = {2024}
}