Holonomic Collision Avoidance for Virtual Crowds

Abstract

All approaches to simulating human collision avoidance for virtual crowds make simplifications to the underlying behaviour. One of the prevalent simplifications is to ignore it's holonomic aspect (i.e. sidestepping, walking backwards). This does not, however, capture the full range of how humans avoid collisions. In real world scenarios we can often observe people sidestepping around each other and obstacles in their environment. In this paper we present a new holonomic collision avoidance algorithm for real-time crowd simulation. Our model is elaborated from experimental data, which allowed us to both observe the conditions under which holonomic interactions occur, as well as the strategies walkers use during such interactions to avoid collision. Our model is general enough to be used with other collision avoidance techniques. We validate our approach by reproducing situations from our experiments and we demonstrate several examples in which our method provides more plausible collision avoidance behaviour.