Date of Award

Spring 1-1-2018

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

First Advisor

Christoffer Heckman

Second Advisor

John Hauser

Third Advisor

Sriram Sankaranarayanan

Fourth Advisor

Nikolaus Correll

Fifth Advisor

Bradley Hayes

Abstract

Autonomous Ground Vehicles (AGV) are mobile robotic platforms used in variety of applications to execute tasks which could be dangerous for humans to operate. Recently, autonomous cars are discussed in carrying passengers from point to point without human interaction. Sophisticated controllers are required to operate autonomous vehicles while responding to both normal and hazardous driving conditions. Dangerous conditions which might be easily perceivable by sensors in the system require controllers that can readily benefit from the new sensory information. In this thesis, we address this problem by asserting that the design of controllers and corresponding calibration and local planning methods are required to quickly adapt to changes in both the dynamic model of vehicle as well as changes in environment.

A full pipeline of calibration, local planner and model predictive controller has been developed and tested in simulation and on physical platform. Properties of a high fidelity model and a simpler model has been studied and their pros and cons has been discussed. Also, a calibration algorithm has been developed to calibrate parameters of dynamic models based on informativeness of robot's motion. Next, A local planning algorithms has been developed to plan vehicle's reference path between consecutive waypoints and finally a model predictive controller has been designed to stabilizes the vehicle to the reference path. A theoretical proof for stability of proposed controller is given. One of the goals behind this work has been design of an adaptive method in a sense that system can quickly adapt to changes in robot's model or environment.

Share

COinS