This is a comprehensive volume of tutorial notes on metareasoning for all kinds of robotics-based systems. It should interest readers working in robotic data analytics. Each of its five chapters ends with a list of references.
The first chapter, “Introduction to Metareasoning,” revisits key concepts such as robot, autonomy, reasoning, rationality, and metareasoning. The author’s attempt to provide precise definitions for key technical terms is met with limited success. Next is a discussion on metareasoning, including its benefits, the related engineering, notes on what is not metareasoning, and further reading.
Chapter 2, “Metareasoning Design Options,” starts with a discussion of the requirements of an engine/algorithm for metareasoning. It covers how to control metareasoning, the different possible modes of metareasoning, and the various policy options available for different stages, including optimizations, scheduling, heuristics, and learning. The chapter also introduces a structure for a multi-robot system, that is, a flexible manufacturing system (FMS).
Chapter 3, “Implementing Metareasoning,” discusses the software architecture and location of metareasoning in a robotic system. A case study related to FMS follows. The fourth chapter, “Synthesizing Metareasoning Policies,” discusses how to compose and synthesize a tool for a robotic system. This is followed by further elaboration of the FMS case study. Finally, chapter 5, “Testing Metareasoning Policies,” covers performance metrics, test planning, visualizing metareasoning, data analysis, assurance, and explicability, all using the FMS case study.
This study of metareasoning in robotics is a fairly easy read. However, most of the analysis is based on neural network paradigms. I should also note that some aspects are missing, including human-guided knowledge collection and discoveries in noisy environments.