Event
M.S. Thesis Defense: Sheng Cheng
Wednesday, July 25, 2018
2:00 p.m.
AVW 2224
Maria Hoo
301 405 3681
mch@umd.edu
ANNOUNCEMENT: M.S. Thesis Defense
Name: Sheng Cheng
Committee:
Professor Nuno C. Martins, Chair/Advisor
Professor Richard J. La
Professor Nikhil Chopra
Date & Time: Wednesday, July 25th, 2018 at 2:00 pm
Place: AVW 2224
Title: Two-stage Optimal Control Methods for Target Reaching Inside a Restricted Area: Theory and Experiment
Abstract:
In this thesis, a new class of problem is studied where a mobile agent is controlled to reach a target. The target is enclosed within a special area in which restrictions apply. The presence of the restricted area requires a controller to have two stages: the outer stage steers the mobile agent to enter such area while the inner stage steers the mobile agent towards the target.
We consider two types of restricted area: a time-costly area and a denied area. For the time-costly area, we formulate a two-stage optimal control problem where time is explicitly specified in the cost function. We solve the problem by solving its subproblems. The key subproblem is a nonconvex quadratic programming with two quadratic constraints (QC2QP). We study the QC2QP independently and prove the necessary and sufficient conditions for strong duality in a general QC2QP. Such conditions enable efficient solution methods for a QC2QP using its dual and semidefinite relaxation. For the denied area, we formulate another two-stage optimal control problem where perturbation is considered. To deal with the perturbation, we propose a robust controller using the variable horizon model predictive control. The performance of the two-stage controller for each type of the restricted area is demonstrated in simulations.
We construct and implement a two-stage controller that can steer a quadrotor to reach a target enclosed within a denied area. Such controller utilizes the formulation and solution methods in the theoretical study. We show experimental results where the controller can run in real-time using off-the-shelf fast optimization solvers. We also conduct a bat experiment to learn bat's strategy for target reaching inside a denied area.