Multi-Agent Systems in Real-Time Applications



  • Prof. Dr. Jürgen Dix
  • Tristan M. Behrens


The first interactive application that could be called ''computer game'' was 1961's Space War, which combined simple graphics, simple newtonian physics and simple game-play to a very exciting space combat simulation. Two human players could steer their armed space-crafts in outer space and fire missiles in order to hit the opponent. The game was conceived at the MIT on a computer used for statistical computations.

In the seventies the first text-adventures appeared. The player found himself in environments that were described completely through texts and interacted with that environment using simple textual commands.

With the availability of high-quality graphical interfaces in the 1980s the visual aspects of computer games reached unprecedented heights. At the beginning of the 90s we saw the first major titles that entered the third dimension -- faster computers allowed the computations that are necessary for three dimensional scenes.

Over the years more calculations per second were possible which lead to a shift toward greater realism in graphics, which was even increased by the introduction of hardware-accelerated 3d graphics. With more and more main-CPU-power available due to the transfer of the rendering tasks to specialized GPUs, the use of physics engines to make complex physics available became an issue.

The game AI did hardly grow as fast as the graphical prospects did. Early titles like the before-mentioned Spacewar and the famous Pong were completely without AI. Hunt the Wumpus which was created in 1972 featured enemy movement based on patterns. Real-time strategy games (beginning of the 1990s) included solutions to several classic academic ai problems like pathfinding, decisions in real time, managing a multitude of objects, dealing with incomplete information and economic planning. Prime examples for real-time strategy games are Dune and Command & Conquer. More specialized techniques like artificial neural nets or machine learning were used in only a few game titles, e.g. Black and White and Creatures.


  • Develop low-level and high-level behaviors for agents controlling military units in the RTS scenario.
  • Introduce a military hierarchy in order to efficiently explore the terrain and solve pathfinding tasks.
  • Extend the techniques above to a warfare-scenario.


Please write an email to Tristan Behrens.

Example 1: A single unit controlled by an agent

Example 2: Several units controlled by agents

Example 3: A harvester exploring the world and gathering resources


Kontakt  Suche  Sitemap  Datenschutz  Impressum
© TU Clausthal 2020