Robotics technology has made progress on a number of important issues in the last decade. However many challenges remain when it comes to the development of systems for human-robot interaction. This paper presents a case study featuring a robust dialogue interface for human-robot communication onboard an intelligent wheelchair. Underlying this interface is a sophisticated software architecture which allows the chair to perform real-time, robust tracking of the dialogue state, as well as select appropriate responses using rich probabilistic representations. The paper also examines the question of rigorous validation of complex human-robot interfaces by evaluating the proposed interface in the context of a standardized rehabilitation task domain.