Artificial Contender™ is an instance-based active learning system using a proprietary method for representing, storing and retrieving knowledge.
The main principle behind the Artificial Contender™ engine
is non-synchronous decision-making about each AC agent’s actions, using the agent’s up-to-date knowledge about a game. Knowledge is organized hierarchically in a flexible number of levels of abstraction – zoom levels. Each zoom level corresponds to a set of specific game situation representations. Game situation representations are game dependent and flexible.
An AC agent’s knowledge about what is happening in the game is represented by semantic directed graphs – acting graphs and generalization trees both growing and being analyzed in real time. Acting graphs store “game situation -> action -> game situation” – type data. Generalization trees provide means for structured hierarchical analysis of game data – the physical organization of “zoom levels” .
Agents’ actions are represented hierarchically to differentiate between strategic and tactical actions of different levels. Each action of a higher level is interpreted as a series of actions of a lower level. AC technology provides a framework for a rich set of flexible game-dependent hierarchies of game situations and actions.
AC Knowledge Viewer tool screen shot (the knowledge graph has been automatically generated during training of AC agent for This Is Football 2005)
Artificial Contender™ technology consists of:
AC scientific fundamentals
Artificial Contender™ SDK lays a solid ground for development of AC agents. The SDK contains reliable and thoroughly tested platform-independent C++ code. AC SDK implementation uses modern metaprogramming techniques based on C++ templates.
Artificial Contender™ tools simplify the development of AC agents, providing a standardized way to test and analyze AC agents' behaviors. With AC tools it is possible to see grounds for every decision that an AC agent makes. It is possible to browse AC knowledge and visually analyze its structure. AC decisions are always transparent even for non-programmers.
Artificial Contender™ documentation provides detailed step by step instructions on the creation of AC agents. A developer can decide whether to use a recommended structure of AC components or assemble a customized solution using flexible AC SDK blocks.
The future of gaming industry lies in introducing ever more sophisticated AI technologies and engines into gameplay. Both customers and developers expect game agents to match real humans in terms of playing style and behavior. Using advanced behavior-capture AI principles, Artificial Contender™ creates a solid ground for this next step of gameplay evolution. With the possibility of human-like interaction and tools that have never been available before, AC opens new territory both for end users and game developers.