Abstract. Mobile intention recognition is the problem of inferring an agent's intentions from the spatio-temporal behavior she shows. We present a framework for mobile intention recognition that covers all levels, from low-level position data to high-level intentions, with the detailed structure of the motion track as an intermediate level. Our framework is new in explicitly including the structure of space into the intention recognition process at each step. We demonstrate the benefit of adding spatial knowledge by an ex-post interpretation of the behavior tracked in a location-based game. Keywords. Spatial cognition, plan recognition, mobile computing `There is a great deal of work in low-level perception, and a great deal in high level recognition - including this thesis. The semantic gap between the output of the lowlevel processes and the high-level inference engines remains wide, and few have ventured to cross it.' H. Kautz (1987) [12, outlook (p. 127)]