next up previous
Next: Theory Up: Introduction Previous: Limited Rationality

The Pursuit Task

To evaluate our algorithm, we have automated the construction of payoff matrices for the pursuit task, in which four predators try to surround a prey which moves randomly. The agents move simultaneously in a two-dimensional square grid. They cannot occupy the same square, and the game ends when all four predators have surrounded the prey on four sides (``capture''), when the prey is pushed against a wall so that it cannot make any moves (``surround'') or when time runs out (``escape''). The pursuit task has been investigated in Distributed AI (DAI) and many different methods have been devised for solving it [10] [17] [11]. These either impose specific roles on the predators, spend much time computing, or fail because of lack of coordination. RMM provides another method for providing this coordination but, to avoid the ``time-consuming'' aspects of game-theoretic approaches, we must devise a theory to support selective expansion of the hierarchy.

Jose M. Vidal
Sun Mar 10 12:52:06 EST 1996