Organizing communication and introspection in a multi-agent blocksworld scenario

Abstract

The implementation of a simple blocksworld-scenario simulation-program is described. The blocksworld is modeled according to the multi-agent paradigm of distributed artificial intelligence. Each block is viewed as an agent. The agents have capabilities like to move, to communicate, to plan or to gain a small amount of introspective knowledge which are necessary to transform the initial scene of a problem into the goal scene. The structure of the system is oriented along the ideas of the specification of RATMAN described in (BMS91). RATMAN was reduced to its two central modules and their concepts were implemented with means as simple as possible. The result was a system, that allows to experimentally develop concepts for communication, planning and introspection, that are (for this simple toy-domain) sufficient to solve the problems without any global problem solver, but by the cooperative behavior in the society of agents.