The document presents a hierarchical activity model to infer a user's location, mode of transportation, and destinations over time using GPS sensor data. It uses a Rao-Blackwellized particle filter algorithm in a hierarchical model to estimate locations, modes, trip segments, and goals at each time step. The model can learn typical transportation modes and goals from data using expectation maximization. An evaluation shows it can accurately model activities and detect errors by identifying novel behaviors. The system is intended to provide predictive notifications and opportunities to users.