Mixed teams of humans and robots - information in motion and decision dynamics in search tasks

Engineering systems so that mixed teams of human operators and robots are able to cooperatively achieve tasks is of considerable current research interest. Such systems require not only robust and scalable algorithms that enable autonomous interaction between robots, but also an understanding of the decision dynamics of the human operators involved. In this work, we present the results of our investigation into two aspects of this effort: communication through motion, and human decision dynamics in search tasks.;Communicating through motion forms a common thread in a surprisingly large variety of contexts, such as gesturing and team play in sports such as soccer or football. We present control laws and communication protocols by which such signaling can be achieved between nonholonomic planar mobile robots. The notion of a context for communication is presented. We extend this idea to the state trajectories of controllable finite dimensional linear time invariant systems, and formulate a joint optimal control communication problem. The solution to this problem is derived for a few interesting special cases, and future directions are presented.;The decision dynamics of human operators is a critically important component of large engineered systems, especially ones that have humans in either supervisory or direct controller capacities. Evidence is mounting that decision styles vary greatly among individuals in the performance of mission tasks. We present research aimed at understanding the decision dynamics involved in search, surveillance and reconnaissance tasks. We have designed and conducted human subject trials with a variety of games: a game of robotic search/reconnaissance in potential fields, a game of reward seeking, and a game of search. We present the results of our analysis of the games of search and reward seeking. In the game of search, the user is required to count the number of real roots that a given random polynomial has in a fixed interval [-1,1]. This game is administered in three different forms: with a flat reward structure, with a variable reward structure, and a two-player variation with partial peer feedback. The distinct play styles that emerged from the analysis are presented. We develop Markov machine models that play indistinguishably from individual players. A formal notion of boredom is defined. We also present discussions of the game of reward-seeking, and comparisons with the results of the robotic field search game.