Human mobility data is valuable for many applications, but it can pose a significant privacy risk for individuals. A person's daily movements are closely linked to their sociodemographic characteristics and their points of interest (POI), which can reveal sensitive information about them, such as their religion or educational background. Researching and mitigating these risks requires realistic, semantically rich data, which is often unavailable or lacks semantic features. We introduce ASTRA, an agenda-based approach for generating synthetic mobility data with semantic attributes. Using real travel surveys and census data, ASTRA simulates artificial agents with sociodemographic features that follow daily activity agendas. It maps activities to semantically similar POIs, and projects them onto real-world locations within a user-defined geographical region. To model movement, ASTRA extends the exploration and preferential return (EPR) model with spatio-temporal and semantic constraints as imposed by an agenda. 
Our evaluation against real check-in data and the EPR baseline model shows that ASTRA can generate realistic mobility data at scale, preserving important characteristics of human movement.