Ph.D. Defense

Dongliang Zheng

(Advisor: Prof. Panagiotis Tsiotras)

"Informed Sampling-based Kinodynamic Motion Planning for Deterministic Systems and Stochastic Systems"

Monday, December 4

12:30 p.m.-1:30 p.m.

Montgomery Knight Building 317

Virtual Link: 

https://gatech.zoom.us/j/6852641782

Abstract
Motion planning, as a fundamental component of robot autonomy, has been studied extensively in the last three decades to increase its efficiency and capability. Efficiency means faster convergence to the same solution or finding a better solution given the same amount of time. Efficient planning algorithms are crucial for robots with limited computation power and for replanning in changing environments. Capability means dealing with more complicated planning problems characterized by high dimensional state space, cluttered environments with irregular obstacles, differential dynamics constraints, and motion and measurement uncertainty. In this thesis, our goal is to develop efficient algorithms for those complicated problems.

Five planning algorithms are developed and evaluated in this thesis for the kinodynamic motion planning problems. Specifically, Kino-RRT*-PFF and Kino-FMT*-PFF are developed for deterministic systems, and CS-BRM, IBBT, and particle IBBT are developed for stochastic systems. A dimensionality reduction heuristic based on the partial-final-state-free (PFF) optimal controller is proposed. The Kino-RRT*-PFF and Kino-FMT*-PFF are developed to take advantage of the dimensionality reduction heuristic and accelerate the speed of finding better solutions. The analytical solution of the PFF optimal control for linear systems is derived. For systems with nonlinear dynamics, Kino-FMT*-PFF using a neural network controller as the steering function is also introduced. CS-BRM is an extension of the popular PRM algorithm to belief space planning. It uses covariance steering to achieve belief node reachability and construct a belief roadmap for multi-query motion planning. For online, anytime, incremental single-query belief space motion planning, IBBT is developed. It divides the stochastic optimal control problem into several simplified problems and uses an informed cost-to-go heuristic for efficient graph search. Particle IBBT inherits all benefits of IBBT and removes the Gaussian assumption in IBBT by representing beliefs using particles. It deals with general noise distributions and nonlinear dynamics directly and has more broad applications.  All proposed algorithms are tested in various problems and promising results are obtained.

Committee

• Prof. Panagiotis Tsiotras – School of Aerospace Engineering (advisor)
• Prof. Kyriakos G. Vamvoudakis– School of Aerospace Engineering
• Prof. Jonathan Rogers – School of Aerospace Engineering
• Prof. Ye Zhao – School of Mechanical Engineering
• Dr. Yebin Wang– Senior Principal Research Scientist, MERL