SE PhD Prospectus Defense of Arman Karimian

TITLE: RESILIENT VISUAL PERCEPTION THROUGH MULTI-AGENT CONTROL

ABSTRACT: In the bearing-based localization and formation control problem, the relative positions of all the agents can be uniquely determined or controlled up to a common scale if and only if the problem is rigid. The notionof rigidity, therefore, plays a central role. In some situations (for instance, restoring rigidity after an agent leaves a formation), it is necessary to solve the rigidity recovery problem: find a way to turn a non-rigid problem into a rigid one by adding new bearing measurements. Furthermore, given the rigidity, maintaining other agents in the line of sight using sensors with a limited field of view is a necessary challenge for eachagent in preserving rigidity and achieving the desired formation. Finally, in the Simultaneous Localization and Mapping task, two or more maps made by single or multiple agents are often needed to be joined by finding the correct connections and discarding outliers.

In this prospectus, we present solutions for all the three problems discussed above. First, we provide analgorithm for rigidity restoration problem with provably minimum number of edges. Second, we provide a path planning algorithm and a controller that avoid the field of view limitations. Third, we model the outlier detection problem as a discreet inference problem and provide a solution.

COMMITTEE: ADVISOR/CHAIR Roberto Tron, SE/ME; David Castaňón, SE/ECE; John Baillieul, SE/ME; Sean Anderssonz,SE/ME

When 1:00 pm to 3:00 pm on Thursday, January 31, 2019
Location 15 Saint Mary's Street, Rm 105