Terrain generation using an Interactive Genetic Algorithm

Abstract

This paper introduces the Auto Terrain Generation System (ATGS), which is based on an Interactive Genetic Algorithm (IGA) that enables non-specialist users to rapidly generate terrains. The motivation for using an IGA is discussed, existing terrain generation techniques are described and a new approach, based on a fractal terrain engine, is outlined. Graphics engines allow terrains to be specified with over 800 floating point parameters, which can overwhelm nonspecialist users. These parameters also create a vast search space for auto-terrain generation systems that can complicate procedural techniques. ATGS addresses these issues via interaction with the user, which is implemented in ATGS by a web based user interface that allows users to rapidly indicate terrain preferences. These user preferences are used by a genetic algorithm to explore a multi-dimensional parameter space that satisfies the user's intuition and aesthetics. Proof of concept experiments are outlined, results are presented and future research work is projected.

Evolving patch-based terrains for use in video games

“…As for the use of the resultant terrain in video games, unlike Ong et al [10] and Walsh and Grade [20] our system does not require one single existing height-map that exhibits similar feature layout to the user's desired terrain. Instead we allow the users to supply multiple sample terrains and to also change their desired goal throughout the evolutionary process.…”

“…methods [6,10,17,20]. The use of patches along with the two-level parent selection mechanism offers a far greater level of interaction between the user and the evolutionary process, giving the user more control of feature arrangement.…”

“…Walsh and Grade [20] have used Genetic Algorithms to modify parameter values that describe the terrain, including feature height, feature roughness, water level, and atmospheric properties such as sun angle and cloud coverage. For the first implementation of their algorithm, Walsh and Grade use an interactive evolution mechanism to optimise Algorithm 1 Roof-Tiling 1: set vert pos to zero 2: while(vert pos+patch length)<heightmap length do 3: set hori pos to 0 4: while(hori pos+patch width)< heightmap width do 5:…”

“…Existing research in this area has produced systems that use evolutionary algorithms to vary the height of control points on the height-map [10], alter mathematical functions that define the height of each vertex on the height-map [6], to search for the optimal values of parameters that define general terrain properties [20], and to place terrain features and game resources on a map [17]. Our novel approach divides sample terrains into equal sized square patches and recombines patches from one or more of these samples to produce a new terrain.…”

Tester interactivity makes a difference in search-based software testing: A controlled experiment

“…Fatigue has already been identified as a major concern, and efforts to alleviate the problem have been proposed [17]. Alternatives have, therefore, been proposed that make it easier for the human user to interact, by selecting candidate solutions they favor and dismissing those they do not [18]; or focusing on the search objectives or the fitness values more than on the candidate solutions themselves [19,20].…”