RATMAN and its relation to other multi-agent testbeds

Abstract

RATMAN (Rational Agents Testbed for Multi Agent Networks) is a workbench for the definition and testing of rational agents in multiagent environments. The special feature of RATMAN is the specification of such agents with hierarchical knowledge bases comprising all knowledge levels from sensoric knowledge to learning capabilities. In all levels only knowledge representation languages have to be used which are based on logic. On each knowledge level the designer may choose the granularity of knowledge for the agent to be designed and moreover he may decide whether the agent should have a certain skill at all. Thus it will be possible to construct a society of very heterogeneous agents from expert systems on one side of the spectrum to simple reactive agents on the other side. Since the aim of such a testbed is to get more insight in the behavior of intelligent agents'; cooperating actions, RATMAN is providing a set of statistical and documentational features. In the second part other approaches to multiagents environments are presented. AF, MACE, AGORA and MAGES are first characterized by their main features. Then their specialities are discussed and finally the boarders with respect to RATMAN are considered.