DD-DR-CVaR Motion Planning

This package contains code for solving a Data-Driven Distributionally-Robust CVaR-based Motion Planning problem. The package uses CVXPy to model and solve the problem.

Installation

• Download/Clone the package
• (Optional) Create a conda environment by running the following in terminal:

  conda create --name dr_cvar_safety_filtering python=3.10 pip -y
  conda activate dr_cvar_safety_filtering
  

• Install dependencies by running the following in terminal:

  pip install numpy==1.23.5 
  pip install scipy==1.9.3 
  pip install matplotlib==3.6.1
  pip install cvxpy==1.2.1
  pip install polytope==0.2.3
  pip install colorama==0.4.6
  

Architecture

We highlight the following files in backend:

• dynamic_vehicles.py: defines classes for dynamic vehicles; i.e. vehicles with both geometry and a dynamics models
• dynamics.py: defines classes for the vehicle dynamics
• geometry_vehicles.py: defines classes for geometric vehicles
• ref_traj_generation.py: defines classes to generate reference trajectories for the ego vehicle
• safe_halfspaces.py: defines classes to obtain safe halfspaces
• safety_filters.py: defines classes for MPC-based safety filters
• simulation_functions.py: mostly includes plotting functions (and some to keep drone_simulations.py cleaner)

Running the Scripts

The following scripts are useful for running the experiments:

• drone_simulations.py: runs the simulation
• experiment_setup.py: setup for experiment scenarios (intersection, head-on, multi-obstacles, ...)
• multi_exp_sim.py: runs Monte Carlo simulations and plots the results
• paper_figures.py: generates the halfspace comparison plots and the halfspace compute time plots

Running simulations and generating experiment data:

• To plot paper supporting figures (non-experiment figures), run paper_figures.py.
• To set up a new type of experiment, edit the experiment_setup.py file
• To run a single simulation, edit the setting of the reach_avoid function in drone_simulations.py and run the script
• To run Monte carlo simulations, edit the toggles and ranges at the top of multi_exp_sim.py and run the script
• To plot experiment data
  • For single experiments, set the plotting variables in drone_simulations.py to True
  • For multiple experiments, set the plotting variable in multi_exp_sim.py to True