ScatterCrowdAI is an ongoing exploration of emergent AI crowd behaviors in a crisis situation. A generic scene is populated with a specified amount of agents, all of which have randomised stats (following a bell curve distribution) reflecting different basic archetypes, and roam, pause or interact with one another randomly. The player can then trigger an explosion at a desired location, triggering a wave of panic across the agents. Panicked agents then alert other agents, leading to an interesting observable spread of distress across agents until agents initially outside of the audible range of the explosion are in a state of panic only due to agents in their vicinity panicking and thus acting highly irrational.
The simulation speed can be freely altered by the player. The simulation can also be paused, in order to examine any individual agent closely. Plans for the future include adding some sort of sandbox gameplay mechanics, and implementing some fun ideas such as zombie agents. ScatterCrowdAI is being developed with Unity.