Improved method for action modeling using Bayesian probability theory

Abstract

The technical development of service robots has enhanced the variety of services provided by them to human beings. Service robots need to interact with human beings; hence, they require considerable learning time. The learning time can be reduced by adopting a learning approach in a virtual environment. To this end, it is necessary to describe a human being's movements in the virtual environment. In this paper, we propose a method to generate an action model of a virtual character by calculating the probability of human movements using Bayesian probability. The virtual character selects actions based on the action model, and it executes these actions. Using the proposed method, the path of a virtual character was decreased by around 74 %, as compared to related methods based on Bayesian probability. © 2013 Springer Science+Business Media.