We present an Agent-Based model called ProtestLab for the simulation of street protests, with multiple types of agents (protesters, police and ‘media’) and scenario features (attraction points, obstacles and entrances/exits). In ProtestLab agents can have multiple “personalities” (implemented via agent subtypes), goals and possible states, including violent confrontation. The model includes quantitative measures of emergent crowd patterns, protest intensity, police effectiveness and potential ‘news impact’, which can be used to compare simulation outputs with estimates from videos of real protests for parametrization and validation. ProtestLab was applied to a scenario of policemen defending a government building from protesters (typical of anti-austerity protests in front of the Parliament in Lisbon, Portugal) and reproduced many features observed in real events, such as clustering of ‘active’ and ‘violent’ protesters, formation of moving confrontation lines, occasional fights and arrests, ‘media’ agents wiggling around ‘hot spots’ and policemen with defensive or offensive behaviour.

--