Andreas Thom

Towards Intelligent Team Composition and Maneuvering in Real-Time Strategy Games

Players of real-time strategy (RTS) games are often annoyed by the inability of the game AI to select and move teams of units in a natural way. Units travel and battle separately, resulting in huge losses and the AI looking unintelligent, as can the choice of units sent to counteract the opponents. Players are affected as well as computer commanded factions because they cannot micromanage all team related issues. We suggest improving AI behavior by combining well-known computational intelligence techniques applied in an original way. Team composition for battling spatially distributed opponent groups is supported by a learning self-organizing map (SOM) that relies on an evolutionary algorithm (EA) to adapt it to the game. Different abilities of unit types are thus employed in a near-optimal way, reminiscent of human ad hoc decisions. Team movement is greatly enhanced by flocking and influence map-based path finding, leading to a more natural behavior by preserving individual motion types. The team decision to either attack or avoid a group of enemy units is easily parametrizable, incorporating team characteristics from fearful to daredevil. We demonstrate that these two approaches work well separately, but also that they go together naturally, thereby leading to an improved and flexible group behavior.

Intelligent moving of groups in real-time strategy games

This paper investigates the intelligent moving and path-finding of groups in real-time strategy (RTS) games exemplified by the open source game Glest. We utilize the technique of flocking for achieving a smooth and natural movement of a group of units and expect grouping to decrease the amount of unit losses in RTS games. Furthermore, we present a setting in which flocking will improve the game progress. But we also demonstrate a situation where flocking fails. To prevent these annoying situations, we combined flocking with influence maps (IM) to find safe paths for the flock in real time. This combination turns out to be an excellent alternative to normal movement in Glest and most likely in other RTS games.

Of motifs and goals: mining trajectory data

In response to the increasing volume of trajectory data obtained, e.g., from tracking athletes, animals, or meteorological phenomena, we present a new space-efficient algorithm for the analysis of trajectory data. The algorithm combines techniques from computational geometry, data mining, and string processing and offers a modular design that allows for a user-guided exploration of trajectory data incorporating domain-specific constraints and objectives.

To model or not to model: Controlling Pac-Man ghosts without incorporating global knowledge

The creation of interesting opponents for human players in computer games is an interesting and challenging task. In contrast to up-to-date computer games, e.g. real time strategy games, learning of non-player-character strategies for older games seems to be easier and not that time-consuming. This way, older games, like the famous arcade game Pac-Man, serve as a test bed for the creation of strategies that are fun to play against. The paper at hand uses computational intelligence methods to accomplish this challenge, namely evolutionary algorithms (EA) and artificial neural networks (ANN). The latter are trained on a model of the game whereas the EA learn good behavior by playing. The performance of these two approaches is compared on the original Pac-Man level as well as on other maps with different properties to test the ability of generalizing the learned strategies.

Making Sense of Trajectory Data in Indoor Spaces

The increasing prevalence of positioning and tracking systems has helped simplify tracking large amounts of, e.g., People moving through buildings or cars traveling on roads, over long periods of time. However, technical limitations of positioning algorithms and traditional sensing infrastructures are likely, especially indoors, to induce errors and biases in the resulting data. In particular, the resulting motion trajectories often do not conform perfectly to the underlying route network. As a consequence, analyses of trajectory sets are impeded by these phenomena, as it becomes hard to identify which route was taken in a particular travel instance or whether two travel instances followed the same route. In this paper, we present a bootstrapping approach and several algorithms to mitigate error biases and related phenomena, focusing on indoor scenarios. In particular, we are able to estimate and iteratively refine an underlying route network from a set of motion trajectories. Secondly, we represent sub trajectories, i.e., Movements on individual elements of the route network, by their median sub trajectory. The resulting aggregated and cleaned-up data set facilitates using further, domain-specific analysis tools. Additionally, it allows to predict the locally occurring expected positioning error biases. This in turn allows improved positioning, e.g., For real-time navigation assistance scenarios. We evaluate the proposed methods using trajectory data from employees at a large hospital complex. In particular, we show that we can reconstruct the hospitals route network accurately, and that we can furthermore extract median sub trajectories for almost all individual corridors. Finally, we illustrate that median trajectories deliver useful deviation maps to learn, and correct for, the expected local biases in positioning.

Acceleration of DBSCAN-based clustering with reduced neighborhood evaluations

DBSCAN is a density-based clustering technique, well appropriate to discover clusters of arbitrary shape, and to handle noise. The number of clusters does not have to be known in advance. Its performance is limited by calculating the e-neighborhood of each point of the data set. Besides methods that reduce the query complexity of nearest neighbor search, other approaches concentrate on the reduction of necessary e-neighborhood evaluations. In this paper we propose a heuristic that selects a reduced number of points for the nearest neighborhood search, and uses efficient data structures and algorithms to reduce the runtime significantly. Unlike previous approaches, the number of necessary evaluations is independent of the data space dimensionality. We evaluate the performance of the new approach experimentally on artificial test cases and problems from the UCI machine learning repository.

Deviation maps for robust and informed indoor positioning services

The ability to position and track people and assets has become increasingly widespread and important in business and personal life. The prevalent means for such tasks is signal-strength-based, prominently WiFi-based, positioning, together with GNSS positioning. The latter, however, is insufficient for the majority of indoor environments in which most of our work and personal lives takes place. Signal-strength-based positioning, though, too, is error-prone in real-life building environments, suffering from large biases induced by the often many and complex attenuating elements in the environment. Additionally, in the prevalent signal-strength-based positioning methods, which rely solely on signal pattern matching, such biases and errors are hard to assess and thus positioning quality and glitches hard to predict. We present an approach for assessing, visualizing, and counter-acting positioning biases and impairments in signal-strength-based positioning. This approach, centered around the notion of deviation maps, aim at improving positioning quality and predictability/reliability and, at the same time, at gaining knowledge and understanding of tracking quality. We seek to understand how the tracking quality is influenced by both positioning installation and building environment, and how the former may be altered to better suit the latter. We discuss results from applying our approach in a real-world large-scale work environment, a major hospital spanning 160,000 square meters, as well as lessons learned from the underlying experimentation-driven and use-centric design process. From these lessons we also derive directions for future work.

Intelligent group movement and selection in realtime strategy games

Movement of groups in realtime strategy games is often a nuisance: Units travel and battle separately, resulting in huge losses and the AI looking dumb. This applies to computer as well as human commanded factions. We suggest to tackle that by using flocking improved by influence-map based pathfinding which leads to a much more natural and intelligent looking behavior. A similar problem occurs if the computer AI has to select groups to combat a specific target: Assignment of units to groups, especially for multiple enemy groups, is often suboptimal when units have very different attack skills. This can be cured by using offline prepared self-organizing feature maps that use all available information for looking up good matches. We demonstrate that these two approaches work well separately, but also that they go together very naturally, thereby leading to an improved and—via parametrization—very flexible group behavior. Opponent AI may be strenghtened that way as well as player-supportive AI. A thorough experimental analysis supports our claims.