Abstract:
|
Model Predictive Control (MPC) has had an increasing role in autonomous driving
applications over the last decade, enabled by the continuous rising of the
computational power in microcontrollers.
In this thesis a collision avoidance trajectory generation algorithm based in MPC
formulation is developed. The operating environment consists in a one-way highway
with two lanes. The overall system is equipped with a low-level controller capable
of tracking the trajectory generated by the MPC planner. In the path towards this
goal, a MPC based lane changing application in an obstacle-free highway
environment has been developed. A point-mass kinematic vehicle model is used as
the MPC plant model for its simplicity and enabled by the usage of a low-level
controller.
This thesis studies several obstacle representation approaches and then, explains
in detail the development process of the collision avoidance trajectory generation
application, defining and discussing simulation results for each intermediate
approach obtained.
Both applications have been implemented in a BeagleBone Black online board
situated in small-scale trucks (1:12) for testing purpose. The experimental results
have been studied and discussed to prove the algorithms functionalities, as well as
to check the board capabilities to run online MPC applications in comparison with
polynomials based approaches. |