Demonstrating Social Error Recovery with AgentFactory

Abstract

In real world applications, agents - be they software agents or autonomous robots - inevitably face erroneous situations that have not been planned for. Re-planning can sometimes provide solutions to problems, but it is computationally expensive and rarely practical. We argue that replanning is not necessarily the last resort. Instead, during erroneous circumstances, agents should always take advantage of other agents in their environment. In this paper we report on early work that looks at Social Error Recovery as a particular class of exception handling that allows agents to resolve erroneous situations that are beyond their direct control. We also show how the AgentFactory Framework and its language AF-APL have been directly extended to support a basic model of Social Error Recovery.