• Library for simulating, testing and tuning a Model Predictive Controller (MPC) for autonomous vehicles.
• Currently the code uses either a unicycle or bicycle model (with ackermann steering) to model the dyanmics of the vehicle.
• The objective of the controller is to track a given reference trajectory in the presence of added process noise.
• The non-linear dynamics of the system are linearized about the reference trajectory.
• The MPC problem for the linearized system is then cast as a constrained Quadratic Programming (QP) problem.
• The constrained QP problem is solved using the OSQP optimization library and its C++/Eigen wrapper

• boost v1.58
• osqp
• osqp-eigen
• Eigen3
• cmake v3.0+
• googletest (only for testing).

• spdlog - For logging/debugging
• json - For reading in params
• matplotlib-cpp - For plotting

git clone https://github.com/vTechSavyy/vehicle-mpc-controller.git
cd vehicle-mpc-controller
mkdir build && cd build
cmake ../
make

./mpc-bicycle-double-lane-change

The params of the simualtion and mpc-controller are set through a json file in the params folder

• OSQP paper
• A brief introduction to MPC

• Interface this library with a high-level planner that provides the reference
• Interface with an open source simualtor (eg: Gazbeo or Webots) for better testing
• Parmeterize the linearized error dynamics in terms of lateral and longitudinal errors w.r.t the reference trajectory.
• Compare mpc with other control methods (eg: Pure pursuit, PID)
• Any further suggestions/feedback is always welcome.