Automated planning is a much explored area in the field of artificial intelligence, with applications in robotics, complex simulation systems, complex infrastructure management, and so on. Emphasis on an actor, “capable of acting deliberately on its environment,” performing predetermined tasks is the essence of planning and acting. And the integration of automated planning with an actor is an analytical approach in this direction. This book addresses as well as advocates this approach.

The book spreads over eight chapters and two appendices, namely “Search Algorithms” and “Strongly Connected Components of a Graph.” The eight chapters are: “Introduction,” “Deliberation with Deterministic Models,” “Deliberation with Refinement Methods,” “Deliberation with Temporal Models,” “Deliberation with Nondeterministic Models,” “Deliberation with Probabilistic Models,” “Other Deliberation Functions,” and “Concluding Remarks.” The authors define a conceptual architecture of an actor using a set of deliberation functions and an execution platform, which are the main components of an actor.

Chapter 2 emphasizes incorporating planning into an actor; however, several kinds of problems might occur, such as execution failures, unexpected events, incorrect information, and partial information. The authors tend to solve these issues using repeated planning and replanning, and online planning incorporating subgoaling, limited-horizon planning, and sampling. Chapter 3 describes the refinement acting engine (RAE), which is inspired by a programming language and OpenPRS, open-source software used in robotics. The authors illustrate the features of RAE, such as controlling the progress of tasks, refining into concurrent subtasks, and choosing methods and stack ordering.

In chapter 4, the authors discuss planning and acting with temporal refinement methods and temporal models. Further, they present the planning problem with nondeterministic models and specify the relations between the different kinds of solutions such as safe and unsafe, cyclic and acyclic, strong and weak, and proper and improper solutions in chapter 5. In chapter 6, the authors gather stochastic shortest path problems and heuristic search algorithms, including a general heuristic search schema, best-first search, depth-first search, iterative deepening search, and heuristics and search control. Furthermore, the authors illustrate online lookaheads interleaving planning and acting in probabilistic domains using generative models, Monte Carlo rollout, sparse sampling, and UCT and Monte Carlo tree search.

Chapter 7 covers deliberation on sensing tasks such as modeling, controlling, recognition, objects detection, goal reasoning, and assessment. Further, the authors focus on learning and model acquisition techniques, in particular reinforcement learning and learning from demonstration approaches in planning and acting. Finally, the authors conclude the book.

This book is an interesting read for professionals, scientists, and doctoral students working in the area of automated planning and acting. Separate sections on discussion and historical remarks, as well as exercises in each chapter, make reading this book worthwhile.