Curvature-Aware Model Predictive Contouring Control based on the following paper:
Note
If using the drone_racing branch, switch to that branch and follow the installation instructions there.
Clone this repo
This repo is a ROS package, to use it as an MPC controller first clone this repo in a catkin workspace.
git clone https://github.com/Lorenzo-Lyons/curvature_aware_mpcc_pkg.git
You will then need to build the package and re-source the workspace.
Install vehicle dynamic model
This repo is intended to work as an MPC controller for a 1:20 scaled down car-like robot developed in the Cognitive Robotics department in Delft university of technology called DART (Delft's Autonomous-driving Robotic Testbed). This repository uses the dynamic models available in the DART git repo, please follow the instructions there on how to install the dynamic model's python package.
Build the MPC solvers
This repo requires the solvers used by the controller to be built in advance. To do so you will need a running installation of acados. We additionally provide the forcespro version of the low-level controller, if you are planning to use it follow the steps on their webpage to set up.
Then navigate to curvature_aware_mpcc_pkg/src/MPC_generate_solvers and run the files named *build_solvers_**. Note that more options for the same solver are available, e.g. the build_solvers_ACADOS_high_level.py has the option to build the CAMPCC or the MPCC version of the lane following mpc. All the options need to be pre-build. Do so by specifing the all the options inside the file.
With a roscore running launch the dart simulator with no visualization (we will use the visualization from this repo instead).
roslaunch dart_simulator_pkg dart_simulator_no_visuals.launch
now launch the visualization from this repo
roslaunch curvature_aware_mpcc_pkg rviz_visualization.launch
now launch the controller
roslaunch curvature_aware_mpcc_pkg mpc_controller.launch
Now select the deisred MPC options from the GUI, select the pygame "safety window" and press the spacebar.
The ACADOS version of the high level CA-MPCC algorithm and the ACADOS version of the low level dynamic bicycle model solvers do not work well for high speeds.