Ubiquitous computing challenges in recognizing and predicting human activity

Ubiquitous computing offers a compelling vision, in which computers vanish into the environment, using sensors to infer their context, and then using dynamically available actuators to influence their environment. This vision has proven as difficult as it is compelling. In this paper, we first outline the vision, then highlight what we believe are the main hurdles preventing its realization, and then present a new paradigm to help overcome these hurdles within a particular subset of inferencing, inferring the activity of humans in their environment. Existing prototypes indicate that the paradigm is a fruitful one - the issues and challenges inherent in going beyond these first steps are described.