Planning and search techniques for intelligent behaviour of battlefield entities

Abstract

Computer Generated Forces (CGFs) are becoming increasingly important in battlefield simulations, in particular for reducing the number of operators required to control entities when training only a few military commanders. These entities must operate in a spatially and temporally continuous dynamic domain, react to any situation in a realistic but non-deterministic manner, using potentially uncertain information about the world. This paper describes our “broad agent” approach for implementing intelligent entities in battlefield CGFs, which attempts to avoid using rules covering every possible contingency but instead uses more fundamental principles to derive plans for achieving objectives. These agents are linked in a Command and Control (C2) hierarchy. An agent's behaviour is implemented as rulesets executed by a tool-kit, developed to support generic agent architectures and interactions between agents. When the agent receives an objective it generates alternative plans by considering possible sequences of actions and using high level principles selects the most appropriate plan for the current situation. This plan is then acted upon, and may include giving objectives to subordinate agents. The results presented throughout this paper have shown that by considering sequences from only a few actions the agents are capable of generating complex plans.KeywordsAgent ArchitectureRidge LineDelay OrderHigh Level CommanderLower Cost RouteThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.