Process Planning and Self-Improvement in Cyber-Physical Systems

Biological organisms show a remarkable flexibility in how they organize their behavior and adapt it to changed circumstances. In this paper, we apply some of the more interesting concepts from biological theory to cyber-physical systems, especially those in such remote or hazardous environments that we cannot expect our control of them to be adequate for success or even survival. We propose a software architecture based on our Wrappings infrastructure, and show how it manages all of the resources necessary for autonomous operation, how it uses interacting planning and decision processes to organize its activity (determining that it cannot do something is one important aspect of the decision and planning processes), and how it uses various analyses of detailed behavioral instrumentation to improve that behavior or determine that improvement is not possible. We describe several difficult questions that arise when implementing our system architecture, and discuss how they might be addressed.