Merging plans with incomplete knowledge about actions and goals through an agent-based reputation system

In this paper, we propose and compare alternative ways to merge plans formed of sequences of actions with unknown similarities between the goals and actions. Plans are formed of actions and are executed by several operator agents, which cooperate through recommendations. The operator agents apply the plan actions to passive elements (which we call node agents) that will require additional future executions of other plans after some time. The ignorance of the similarities between the plan actions and the goals justifies the use of a distributed recommendation system to produce a useful plan for a given operator agent to apply towards a certain goal. This plan is generated from the known results of previous executions of various plans by other operator agents. Here, we present the general framework of execution (the agent system) and the results of applying various merging algorithms to this problem.