Date of Completion

4-5-2019

Embargo Period

4-4-2020

Keywords

autonomous vehicles, motion planning, coverage, multi-robot system, resilience, curvature-constrained vehicles, potential games

Major Advisor

Prof. Shalabh Gupta

Associate Advisor

Prof. Peter B. Luh

Associate Advisor

Prof. Shengli Zhou

Associate Advisor

Dr. Thomas A. Wettergren

Field of Study

Electrical Engineering

Degree

Doctor of Philosophy

Open Access

Open Access

Abstract

Autonomous vehicles are rapidly advancing with a variety of applications, such as area surveillance, environment mapping, and intelligent transportation. These applications require coverage and/or time-optimal motion planning, where the major challenges include uncertainties in the environment, motion constraints of vehicles, limited energy resources and potential failures. While dealing with these challenges in various capacities, this dissertation addresses three fundamental motion planning problems: (1) single-robot complete coverage in unknown environment, (2) multi-robot resilient and efficient coverage in unknown environment, and (3) time-optimal risk-aware motion planning for curvature-constrained vehicles.

First, the ε* algorithm is developed for online coverage path planning in unknown environment using a single autonomous vehicle. It is computationally efficient, and can generate the desired back-and-forth path with less turns and overlappings. ε* prevents the local extrema problem, thus can guarantee complete coverage. Second, the CARE algorithm is developed which extends ε* for multi-robot resilient and efficient coverage in unknown environment. In case of failures, CARE guarantees complete coverage via dynamic task reallocations of other vehicles, hence provides resilience. Moreover, it reallocates idling vehicles to support others in their tasks, hence improves efficiency. Finally, the T* algorithm is developed to find the time-optimal risk-aware path for curvature-constrained vehicles. We present a novel risk function based on the concept of collision time, and integrate it with the time cost for optimization.

The above-mentioned algorithms have been validated via simulations in complex scenarios and/or real experiments, and the results have shown clear advantages over existing popular approaches.

COinS