In this paper, we present our egocentric interaction model for recog nizing and supporting everyday human activities. We explain how it allows designers of ubiquitous computing systems to view physical (real) and virtual (digital) objects as residing in one single space and how sets of objects in the vicinity of a specific human actor can be classified based on human perceptual characteristics such as what can be observed and what can be manipulated. We also propose a wearable computer architecture that is based on the egocentric interaction model which potentially could facilitate the development of Ubiqui tous Computing applications by letting an operating system take care of main taining communication with worn and instrumented sensors as well as computing devices. Finally, we present our first steps in implementing an activ ity-aware wearable support system for people suffering mild dementia based on the proposed model and architecture.