RPE: Piyush Khandelwal, GDC 3.516

Contact Name: 
Lydia Griffith
Date: 
May 21, 2014 11:00am - 1:00pm

RPE:  Piyush Khandelwal

Date: Wednesday, May 21 2014
Time: 11AM
Place: GDC 3.516
Committee Members: Peter Stone (chair), Isil Dillig, Vladimir Lifchitz

Title: Leading the Way: An Efficient Multi-robot Guidance System

Short Abstract:
Traditionally, human guidance has been studied using only a single
robot, where the robot leads a human from start to finish. However,
most robots do not have sufficient mobility yet to match a human's
speed or ease of navigation in indoor environments. When it is
necessary to guide the human to his goal quickly and efficiently,
multiple coordinated robots may provide a superior solution. This
paper formulates a multi-robot treatment of the human guidance problem
as a Markov Decision Process (MDP). Solving the MDP produces a policy
to efficiently guide a human, but the size of the state space makes it
unfeasible to optimally solve it. We first show that allowing the
robot to instantaneously move in the environment allows us to
optimally solve this problem, but at the same time prevents the
solution from being implemented in the real world. Next, we drop this
assumption and use the Upper Confidence Bound for Trees (UCT) planner
to obtain an approximate solution on the complete problem description.
Empirical comparisons demonstrate the superiority of this solution to
a heuristic approach and to using a single guide robot.