Distributionally Robust Risk Map for Learning-Based Motion Planning and Control: A Semidefinite Programming Approach

This repository includes an official python implementation of DR-RRT* and DR-MPC presented in Distributionally Robust Risk Map for Learning-Based Motion Planning and Control: A Semidefinite Programming Approach

1. Requirements

• All our experiments are performed in MATLAB
• The DR-risk map is constructed by solving the SDP problem using MOSEK
• The MPC is solved using FORCESPRO 6.0.0

2. Quick Start

2.1. DR-RRT*

To test the performance of our motion planning algorithm, we have implemented to environments:

• Highway Scenario
• Intersection Scenario

To run the experiments, first navigate either to DR-Risk-Map/DR-RRTStar/Highway or to DR-Risk-Map/DR-RRTStar/Intersection then simply run the script DR_RRTStar.m.

Example:

cd DR-Risk-Map/DR-RRTStar/Highway
DR_RRTStar

2.2. DR-MPC

For DR-MPC Neural Networks are trained to construct the risk maps for several Wasserstein radii. The corresponding weights are saved in DR-Risk-Map/DR-MPC/NN_Weights/.

To run the experiment for the service robot, navigate to DR_Risk-Map/DR-MPC and run the script ServiceRobot.m. It will first generate a FORCESPRO solver scripted in ServiceRobot_Forces_Gen.m then apply the controller to the robot.

Example:

cd DR-Risk-Map/DR-MPC
ServiceRobot