Opportunistic Safety Outside the Maximal RCIS

Authors: Zexiang Liu, Hao Chen, Yulong Gao, Necmiye Ozay

Dependency

MATLAB

• MPT3
• PCIS
• GUROBI

Python (only if you want to run drone experiments)

• Numpy
• Scipy
• Matplotlib
• CVXPY
• cflib

Reproduce the experimental results in the paper

Environment Setup (required before running any examples)

Open MATLAB. Run the following prompts in the workspace of MATLAB:

# Set up GUROBI
cd 'directory_of_gurobi/linux64/matlab'
gurobi_setup

# Set up PCIS
cd 'directory_of_pcis'
addpath(genpath('pcis/lib/'));

# Initialize MPT3
cd 'directory_of_tbxmanager' # tbxmanager is installed with MPT3
startup
clc;
mpt_init

# Set up our code
cd 'directory_of_this_repo'
initial

Car Following Example

Run the following prompts in the workspace of MATLAB:

cd 'directory_of_this_repo/experiments/car_following'
open run_experiments.m

Then, follow the instructions in run_experiments.m to reproduce the results.

For additional experiments with disturbances in truncated Gaussian distribution, run the following prompts in the workspace of MATLAB:

cd 'directory_of_this_repo/experiments/car_following'
open run_experiments_gaussian.m

Then, follow the instructions in run_experiments_gaussian.m to reproduce the results.

Lane Keeping Example

Run the following prompts in the workspace of MATLAB:

cd 'directory_of_this_repo/experiments/lk'
open run_experiments.m

Then, follow the instructions in run_experiments.m to reproduce the results.

Safe Tracking for Aerial Vehicles

We use Crazyflie 2.1 for the experiments.

Preparation

To implement the safety supervisors, we need to identify a model of the drone from data, and then compute the maximal RCIS.

For system identification, run the MATLAB script drone_sys_id.m under the directory OppSafe/experiments/drone. The result is saved to OppSafe/experiments/data/drone_dyn_3d.mat.

For RCIS computation, run the MATLAB script compute_cis_drone.m under the directory OppSafe/experiments/drone. The result is saved to OppSafe/experiments/data/alpha_filter_py.mat.

We included our precomputed results in the folder OppSafe/experiments/drone/data.

Run the experiment

Make sure you install all the Python dependencies and connect your laptop to the drone. Ensure that there are no obstacles within a 2-meter radius of the drone.

In a terminal (we only test the code in Ubuntu), navigate to the folder OppSafe/experiments/drone and run the Python script drone_landmark_vel.py. If everything works well, the drone should take off and start tracking an "M"-shape reference trajectory. Press Ctrl-C in the terminal for emergency stop.

Visualization

In a terminal, navigate to the folder OppSafe/experiments/drone and run the Python script generate_plot.py. If you run your own experiments, make sure you modify the data file name in the script to yours.