Improved Obstacle Relevancy, Distance, and Angle for Crowds Constrained to Arbitrary Manifolds in 3D Space

Abstract

Recent work has proposed crowd simulation algorithms on arbitrary manifolds in 3D space. These algorithms simulate crowds on far more realistic surfaces than previously possible, including multi-story structures, science fiction scenarios, and habitats for insects and other animals that can walk on walls. However, current implementations can have distinct artifacts, including collision false positives and false negatives. Also, current implementations fail to account for the cylindrical shape of the characters being simulated. The resulting crowds move unnaturally and have obvious collisions. After identifying the cause of these artifacts, we propose an algorithm that does not struggle from these false positives or false negatives and correctly accounts for the non-spherical shape of agents. The resulting crowds move on large surfaces (over 100k triangles) running with a thousand agents in real-time.