On Transfer from Multiagent to Multi-Robot Systems Elizabeth Sklar 1,2 , A. Tuna Ozgelen 1 , Eric Schneider 3 , Michael Costantino 4 , J. Pablo Munoz 1 , Susan L. Epstein 1,3 , and Simon Parsons 1,2 1 The Graduate Center, City University of New York 365 Fifth Avenue, New York, NY 10016, USA 2 Brooklyn College, The City University of New York 2900 Bedford Avenue, Brooklyn, NY 11210, USA 3 Hunter College, City University of New York 695 Park Avenue, New York, NY 10065, USA 4 College of Staten Island, City University of New York 2800 Victory Boulevard, Staten Island, New York 10314 sklar@sci.brooklyn.cuny.edu, aozgelen@gc.cuny.edu, nitsuga@pobox.com, michael.costantino@cix.csi.cuny.edu , jpablomch@gmail.com, susan.epstein@hunter.cuny.edu, parsons@sci.brooklyn.cuny.edu Abstract. Our research involves application of methods well-studied in virtual multiagent systems (MAS) but less well-understood in physical multi-robot systems (MRS). This paper investigates the relationship be- tween performance results collected in parallel simulated (multiagent) and physical (multi-robot) environments. Our hypothesis is that some performance metrics established in simulation will predict results in the physical environment. Experiments show that some performance metrics can predict actual values, because data collected in both simulated and physical settings fall within the same numeric range. Other performance metrics predict relative values, because patterns found in data collected in the simulated setting are similar to patterns found in the physical set- ting. The long term aim is to establish a reliability profile for comparing different types of performance metrics in simulated versus physical envi- ronments. The work presented here demonstrates a first step, in which experiments were conducted and assessed within one parallel simulated- physical setup. 1 Introduction Our research investigates issues that are well-studied in virtual multiagent sys- tems (MAS) but present particular difficulties in physical multi-robot systems (MRS). These issues center around how to coordinate activity and allocate tasks to team members in real-time, dynamic, noisy, constrained environments. An overarching goal is to identify which MAS approaches are feasible for multi- robot teams and transfer well in terms of performance to an MRS setting. This work is motivated by two related application areas: urban search and rescue (USAR) [1, 2] and humanitarian demining [3, 4]. In both instances, teams