A Nonlinear Model Predictive Control based Evasive Manoeuvre Assist Function

A Nonlinear Model Predictive Control based Evasive Manoeuvre Assist Function

van Lookeren Campagne, Gijs (TU Delft Mechanical, Maritime and Materials Engineering)

Mazo, M. (mentor)

Delft University of Technology

Mechanical Engineering | Systems and Control

2019-07-02

In recent years, research in the field of autonomous driving has been subject to a significant increase in interest. Along with the advances in automation and electrification, cars are slowly transitioning to computers on wheels. The ever-increasing processing power in vehicles gives rise to the implementation of advanced control strategies. The increase in computational power not only allows for novel solutions to automated driving but can also be used for advanced control systems aimed at collision avoidance. At the same time, significant advances have been made towards efficient nonlinear optimisation methods, giving rise to new possibilities for real-time applications. This thesis focuses on an Evasive Manoeuvre Assistance Function (EMA) to avoid obstacles at the limits of handling using a Model Predictive Control (MPC) approach. Vehicles at the limits of handling typically operate in the nonlinear region of the tyre. For this reason, a Nonlinear Model Predictive Control (NMPC) approach is employed. A baseline scenario with a lateral displacement of 2 metres is considered, representing a common near rear-end collision situation. An optimal trajectory and steering sequence is computed by the MPC over a defined horizon. As MPC considers a model to calculate an optimal control input, a nonlinear two-track model with a simplified Magic Formula is used. The main drawback of real-time NMPC has always been its computational burden. For the baseline scenario, the limits of real-time NMPC are explored, using the state-of-the-art non- linear optimisation methods: Sequential Quadratic Programming (SQP) and Interior-Point Method (IPM). Both methods are compared in simulation and an SQP approach is tested on an embedded system within a test vehicle, subject to the disturbances and uncertainties that come with physical driving. It was found that the proposed NMPC strategies make for a robust yet aggressive manoeuvre with high lateral accelerations, utilising the nonlinear operating region of the tyre. Furthermore, computation times on an embedded system in a test vehicle were found to be sufficiently low for real-time application.

Nonlinear
Model
Predictive
Control
Evasive
Manoeuvre
Assist
Function
Sequential
Quadratic
Programming
Interior
Point
Method
Optimisation
Real-time
Embedded

http://resolver.tudelft.nl/uuid:21edac7c-93b5-4140-bc28-663879f63108

57.728228, 11.859602

Student theses

master thesis

© 2019 Gijs van Lookeren Campagne