This repository contains two different MPC controllers: a state feedback and an output feedback based one that both use a linearized kinematic bicycle model. Besides there is also an LQR controller implementation. The car is assumed to be a point and the dynamics are simulated with a nonlinear model. The controller takes state and input constraints into account and has a terminal constraint to ensure stability. A terminal set can be automatically constructed by finding a control invariant set around the goal using the unconstrained LQR control law. More information can be found in the paper, which also contains the proof that ensures asymptotic stability.
The repository contains several environments to run a simulation in with a controller. These can be found in lib/environments.py:
RoadEnv |
RoadOneCarEnv |
RoadMultipleCarsEnv |
|---|---|---|
 |
 |
 |
examples/run_MPCStateFB.py contains an example which runs the state feedback based MPC.
examples/run_MPCOutputFB.py contains an example which runs the output feedback based MPC.
examples/run_LQR.py contains an example which runs the LQR controller.
examples/find_terminal_set.py computes the terminal set for a specific environment with a specific goal and saves it to terminal_sets/.
The MPC controller looks for a terminal set, when it needs to run in a specific environment with a specific goal.
When there is no terminal set for that specific combination, it uses the goal state for the terminal constraint.
Since this is a much stricter constraint, the region of attraction reduces, which might make the MPC problem infeasible.
RoadOneCarEnv |
RoadMultipleCarsEnv |
|---|---|
 |
 |