Plan-directed architectural change for autonomous systems

Autonomous systems operate in an unpredictable world, where communication with those people responsible for its software architecture may be infrequent or undesirable. If such a system is to continue reliable operation it must be able to derive and initiate adaptations to new circumstances on its own behalf. Much of the previous work on dynamic reconfigurations supposes that the programmer is able to express the possible adaptations before the system is deployed, or at least is able to add new adaptation strategies after deployment. We consider the challenges in providing an autonomous system with the capability to direct its own adaptation, and describe an initial implementation where change in the software architecture of an autonomous system is enacted as a result of executing a reactive plan.