MechE PhD Prospectus Defense: Zachary Serlin

Starts:
10:00 am on Monday, September 23, 2019
Ends:
12:00 pm on Monday, September 23, 2019
Location:
110 Cummington Mall, Room 245
TITLE: DISTRIBUTED FORMAL METHODS AND SENSING FOR ROBOTIC SYSTEMS.

ABSTRACT: Modern multi-robot systems quickly become too complex for humans to control. For humans to control a team of robots from only high level mission specifications, flexible algorithms for planning, coordination, control, and perception have become a necessity. Automated control of a robotic team requires a shared understanding of the mission and goals, planning of individual robot motions, and coordinated perception along the planned routes. These problems arise in applications such as persistent surveillance, robotic agriculture, post-disaster search and rescue, and autonomous driving.

Unlike other work in planning, control, and perception for distributed systems, this work considers both homogeneous and heterogeneous robot teams, tasks the team from generally applicable specification languages, and employs more efficient data association techniques for perception. Overall, this work aims to address issues that arise at every level of specification based planning, coordination, control, and perception; including multi-agent task planning from temporal logic specifications, coordinated lower level motion and path planning, and task level sensing, perception, and object tracking, all in the context of a distributed team of robots.

This prospectus begins with completed work and then moves to current and future work. It begins by introducing a solution for coordinating a team of homogeneous mo- bile robots, each with a noisy sensor array, subject to linear temporal logic constraints. The next section explores task level multi-robot, multi-image, object matching and tracking via the QuickMatch algorithm. The final section summarizes current and future research, which includes tasking of a heterogeneous multi-robot team from formal specifications, algorithms for reactive high level task planning with safety guarantees for a single robot, and a distributed extension of the QuickMatch algorithm for object tracking called NetMatch.

COMMITTEE: ADVISOR Professor Calin Belta, ME/SE/ECE; Professor Roberto Tron, ME/SE; Professor Sean Andersson, ME/SE; Professor Rebecca Khurshid, ME/SE