Resumen de Advances in planning and control for autonomous vehicles
This thesis presents some advances to the state of the art of state estimation, automatic control and trajectory planning fields applied to autonomous vehicles.
Such contributions have a common aspect throughout the thesis, all of them are model-based techniques.
The Linear Parameter Varying (LPV) Takagi-Sugeno (TS) theory are used to generate control-oriented models by using the non-linear embedding approach. Several vehicle models are proposed depending on lhe application and estimation -control -planning technique. First, non-linear vehicle formulations are presented. Later, the same models are represented in the LPV form. In the area of control and estimation, the thesis shows different approaches for diferent applications: normal and racing driving modes.
First, for normal driving, gain scheduling (GS) LPV state feedback techniques are developed.
In the first instance, an LPV-Linear Quadratic Regulator (LQA) design via Linear Matrix lnequality (LMI) formulation is stated for control at low velocities. Later, a cascade scheme including kinematic and dynamic control layers is presented to improve the last design. Here, both controller designs are set up using the LPV-LQR design via LMI formulation and a LPV-Unknown Input Observer (UIO) is presented for estimating vehicle states and exogenous friction force. Second, for racing driving, optimal techniques are explored leading to introduce the Model Predictive Control (MPC) technique as a basis for racing behaviours. In the first instance, the cascade scheme is maintained where the outer control layer is governed by a TS -MPC controller.
At this point, an advanced estimation technique is presented, the TS-Moving Horizon Estimator-UIO (TS-MHE-UIO). lt is shown that by using the TS formulation both optimal-based controller and estimator reduce greatly the computational effort in comparison to their non-linear formulation. Then, the idea of designing a unique controller is explored through the LPV-MPC technique. In this case, it is shown the potential of this strategy being able to be executed in real time in small embedded platforms for controlling the vehicle in racing situations.
Finally, an online robust MPC is considered that aims at improving the computational load using zonotope theory while preservin high levels of robustness and performance in racing scenarios.
In the area or planning, the thesis focuses on trajectory planning approaches from the optimal point of view.
First, the non-linear MPC is formulated as a planner (NL-MPP) in space domain where the goal is the minimization of the total lap time.Later, an innovative real time solution is explored leading to a LPV-MPP.
The method follows the structure of the model predictive optimal strategy where the main objective is to maximize the velocity while fulfilling varying constraints. In particular, the aim is on reformulating the non-linear original problem into a pseudo-linear problem by convexifying the objective function and making use of the LPV vehicle formulation.
