Plan Recognition in Instrumented Environments

Plan recognition is the process of inferring a user's plans and goals by observing her behavior. With their sensors, instrumented environments provide a good infrastructure for the application of plan recognition, and in return can benefit from plan recognition knowledge by better supporting their users. This article addresses specific issues that arise if plan recognition is applied in physical environments. A plan recognition model based on dynamic Bayesian networks is presented and practically applied in an instrumented kitchen scenario. Furthermore, a utility function for sensor information is derived from this model, which can be used to solve the problem of sensor selection in real-world plan recognition applications.