Combining Symbolic and Statistical Knowledge for Goal Recognition in Smart Home Environments

An essential feature of pervasive, intelligent systems is the ability to dynamically adapt to their users' current needs. Hence, it is critical for such systems to be able to recognize the current goals and needs of the users based on observed past and current actions. This work addresses the problem of goal recognition in smart home environments. We investigate whether approaches for the plan recognition problem, which is a long-standing research area in the Artificial Intelligence community, can also be applied to the goal recognition problem in smart home environments. Therefore, we evaluate the application of a well-known symbolic plan recognition approach, which is based on classical planning methods, and propose to extend this approach through additional statistical knowledge to overcome some identified shortcomings of the planning-based approach. We show that the planning-based plan recognition approach indeed can be used to solve the goal recognition problem in smart home environments and show that the proposed extension outperforms the original approach as well as purely statistical goal recognition methods.