An Interaction Protocol for Mutual Assistance in Agent Teamwork

An interaction protocol for bidirectional deliberation on direct help in agent teamwork.

“…Helpful behavior in agent teamwork refers to a collaborative act performed by one agent in order to assist its teammate. Different mechanisms for assistance between teammates by executing actions or providing information have been proposed in the last two decades [Itoh, 1991, Miceli et al, 1994, Yen et al, 2004, Cao et al, 2005, Fan et al, 2005, Kamar et al, 2009, Polajnar et al, 2012.…”

“…Agent interaction protocols specify interactive behavior of individual agents dealing with each other. Nalbandyan [2011] and Polajnar et al [2012] introduce the M utual Assis tance Protocol (MAP), with its subsequent elaboration in [Nalbandyan et al, 2013], which specifies a direct help mechanism for an agent team with heterogeneous skill profiles, situated in a dynamically changing environment. The mechanism is based on a bilateral distributed agreement in which both requester and helper agents jointly decide on performing a help act, in contrast with unilateral approaches as in [Ka-mar et al, 2009], in which only one side of the help transaction makes the decision.…”

“…Simulation experiments in [Polajnar et al, 2012] suggest behavioral advantages of employing Action MAP over unilateral help protocols when communication cost is relatively low compared to agents' action costs.…”

“…Simulation experiments in [Nalbandyan et al, 2013] show th a t the Requester In this thesis, we approach this research question by designing new protocols, developing simulation models for them, exploring their impact on team performance through simulation experiments, and adjusting our model parameters for optimum performance. We study agent interaction protocols for direct help in the context of a board game microworld, inspired by the Colored IVails game [Gal et al, 2010], and used in previous studies of MAP [Polajnar et al, 2012, Nalbandyan et al, 2013. For modeling and simulation, we use the Agent Interaction Modeling Simulator (AIMS) framework introduced by Alemi et al [2014], which allows concurrent simulation of multiple teams in identical dynamic environments.…”

“…In summary, Nalbandyan [2011] and Polajnar et al [2012] introduce MAP as a better performing interaction protocol in comparison to the unilateral protocols in situations when the communication costs are relatively low compared to the actions costs. RIAMAP and HIAMAP are introduced in [Nalbandyan et al, 2013] as two ad vanced variations of MAP, which outperform the basic MAP version and demonstrate complementary behaviors over the param eter space.…”