On Wednesday April 2 at noon Prof. Manuela Veloso will give a presentation on Team Strategies. The presentation will take place in TSRB and lunch will be provided.

Selecting and Learning Multi-Robot Team Strategies

Manuela M. Veloso
Herbert A. Simon Professor
Computer Science Department
Carnegie Mellon University

Abstract

We will present different approaches to generating efficient multi-robot cooperative strategies. We first focus on multi-robot teams in games, which by definition, offer the challenge of the presence of an opponent. We discuss an adaptive playbook approach with implicit opponent modeling, in which multiple team strategies are represented as variable weighted plays. Different plays are suited for different opponents and weights can be dynamically adapted. Play selection proceeds probabilistically based on the play weights. We further present an MDP-based policy to specifically select strategies towards winning, based on the score and remaining time in a game. We lastly conclude by presenting recent results with successfully teaching multi-robot strategies by demonstration. The work presented is in conjunction with several of my students, in particular Colin McMillen and Sonia Chernova.