Using trails to improve map generation for virtual agents in large scale, online environments

Samperi, Katrina (2016). Using trails to improve map generation for virtual agents in large scale, online environments. University of Birmingham. Ph.D.

Preview

Samperi15PhD.pdf
PDF
Download (9MB)

Abstract

This thesis looks at improving the generation of maps for intelligent virtual agents in large scale environments. Virtual environments are growing larger in size and becoming more complex. There is a major challenge in providing agents that are able to autonomously generate their own map representations of the environment for use in navigation. Currently, map generation for agents in large scale virtual environments is performed either by hand or requires a lengthy pre-processing step where the map is built online. We are interested in environments where this process is not possible, such as those that encourage user generated content.
We look at improving map generation in these environments by using trails. Trails are a set of observations of how a user navigates an environment over time. By observing trails an agent is able to identify free space in an environment and how to navigate between points without needing to perform any collision checking. We found that trails in a virtual environments are a useful source of information for an agent's map building process. Trails can be used to improve rapidly exploring randomised tree and probabilistic roadmap generation, as well as being used as a source of information for segmenting maps in very large scale environments.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):