Dynamically altering agent behaviors using natural language instructions.pdfVIP

Dynamically Altering Agent Behaviors Using Natural Language Instructions Rama Bindiganavale, William Schuler, Jan M. Allbeck, Norman I. Badler, Aravind K. Joshi, Martha Palmer University of Pennsylvania 200 S. 33rd St. Philadelphia, PA 19104 1-215-898-1976 {rama, schuler, allbeck, badler, joshi, mpalmer} @graphics.cis.upenn.edu ABSTRACT other things, any conditional instruction that references a hypothetical object in its condition. For example, a user could Smart avatars are virtual human representations controlled by not give the instruction “if a military vehicle enters the real people. Given instructions interactively, smart avatars can checkpoint, open the gate,” if there is no such vehicle to click on act as autonomous or reactive agents. During a real-time when the instruction is given. simulation, a user should be able to dynamically refine his or her avatar’s behavior in reaction to simulated stimuli without having A programming or scripting language such as that used in to undertake a lengthy off-line programming session. In this Improv [25] can provide a sufficiently powerful interface for paper, we introduce an architecture, which allow