Briefly, Arizona State University’s School of Geographical Sciences has an interesting tool to model crowd behavior. This has interesting use in other areas.
There are two adaptations to this that I’d find most cool. Change the information input (car bomb / explosion causing the stampede) to be more subtle and over time (i.e. information and misinformation campaigns). Track physical and ideological movement. I know of some work in this space already, but this could build upon them.
Second, use these environments to test human interfaces with unmanned systems, both autonomous the teleoperated (remote controlled). That, to my knowledge, has not been done. I talked to some folks about doing this last year….