On Weighted Regions and Social Crowds: Autonomous-agent Navigation in Virtual Worlds

Abstract

Virtual environments have gained in importance in many aspects of the world we live in today. Immersive virtual worlds are ubiquitous in modern movies, video games, and online communities. They are also important in non-entertainment applications such as training and education software, simulations of mass events, evacuation scenarios, human factor analysis, and urban city planning. Training and education software itself comprises various application areas, ranging from teaching children with the help of virtual characters, to training policemen and firefighters, or training soldiers in virtual environments for military operations. Other applications are online mapping services such as Open Street Map, Google Street View, or Mapillary. A key aspect of creating an immersive virtual world is the development of algorithms that handle the navigation of its virtual inhabitants. This involves the creation of believable paths that are smooth, do not contain unnecessary detours, keep clearance from obstacles, respect terrain and region information, and avoid collisions with other moving entities. Furthermore, it involves the coordination of large virtual crowds in both sparse and dense situations, and the generation of social behavior among virtual groups. State-of-the-art algorithms and crowd-simulation models struggle with such tasks, and consequently, the range of possible character behaviors is still limited up to the present day. This thesis focuses on three computational tasks, with which state-of-the-art algorithms still struggle: Region-based path planning, region-based path following, and coordinating dense virtual crowds and social groups. We show why these tasks are difficult to solve with existing algorithms when using grids or graph-based representations of the traversable space in a virtual environment. Furthermore, we show that these tasks can be solved efficiently on a surface-based representation when using novel methods. These novel methods on region-based planning and coordinating crowds and social groups are presented in detail, and they form the main contributions of this thesis.