ABSTRACT
Crowd simulation is the process of generating moving trajectories of many agents starting from initial conditions driven by rule-based models [1-4], flow models [5,6], logical formalisms [7,8], or prior observations [9-11]. The simulator usually has a number of parameters to tune and the quality of simulation often depends on the choice of parameter values. Even though many crowd simulation results are generally satisfactory, such visible artifacts as awkward behaviors and bottleneck situations may arise in a small portion of the simulation. An animator may need to tune the parameters carefully and rerun the simulator repeatedly until an overall satisfactory result is achieved. Sometimes, manual parameter tuning and repeated runs could be tedious and time consuming.
