Communication Dataset from Distributed Model Predictive Control

This repository contains the dataset created and used for the work Reducing the Communication of Distributed Model Predictive Control: Autoencoders and Formation Control and consists of `dataset.csv` - recorded data from inter-agent communication `metadata.txt` - information regarding the training tasks `plotdataset.py` - a Python script to visualize the data The dataset consists of post-processed sequences of candidate inputs along the prediction horizon of inter-agents communication in 2000 simulated scenarios using distributed model predictive control. The dataset is stored in the csv format and the velocities alternate between linear velocity $v$ and angular velocity $\omega$ in the columns, i.e., they are stored following the prediction horizon. All scenarios used two robots in a DMPC setting. The linear velocites are denoted as `v_1` to `v_20` and the angular velocities similarly with `omega_1` to `omega_20`. The higher the number, the further away from the current state the prediction of the input is. The metadata file `metadata.txt` contains setup information for each scenario including, for example, the number of robots, the simulation duration `tend`, and the initial state `x0` as well as the desired output. The initial state is given in absolute coordinates in the inertial frame of reference in the format $\begin{bmatrix}x_1 & y_1 & \theta_1 & x_2 & y_2 & \theta_2 \end{bmatrix}^\top$, where $x_i$ and $y_i$ are the position, and $\theta_i$ the orientation of robot $i$. The desired output consist of the absolute values for the position and orientation of the geometric center of the formation, the positions of the robots relative to the geometric center and the desired orientation of the formation. A metadata entry has the following structure: Number of robots: 2 x0: [-5.024750007781551, 6.088088095621365, 2.604305537257288, -6.0247588076953615, 6.088107912584617, 2.6041282948414115] Desired state: [0.0, 0.0, 0.0, 0.5, 0.0, 0.0, -0.5, 6.123233995736766e-17, 0.0] Prediction horizon: 20 Time: step size = 0.25, tend = 40 Weights: [1.0, 5.0, 0.1, 0.125, 0.0125] Message timeout = 200 Optimizer: alpha = 1, beta = 0.1, sigma = 0.5 Number of Iterations per time step: 3 Plot the Candidate Inputs The repository contains a Python script, which allows to easily visualize the candidate input sequences for each sample. Requirements matplotlib pandas To install the required packages run, for example, pip install pandas matplotlib in a terminal. Run To run the script, use the following command, in Linux, e.g., in bash: python plotdataset.py <path_to_csv_file> where `<path_to_csv_file>` is the path to the dataset file. Running the script plots a figure with slider allowing to slide through the samples in the dataset.