Chapter 11 introduces perhaps the most important concept of this book: the information space. If there is uncertainty in sensing the current state, then the planning problem naturally lives in an information space. An analogy can be made to the configuration space and motion planning. Before efforts to unify motion planning by using configuration space concepts [588,657,852], most algorithms were developed on a case-by-case basis. For example, robot manipulators and mobile robots have very different characteristics when defined in the world. However, once viewed in the configuration space, it is easier to consider general algorithms, such as those from Chapters 5 and 6.
A similar kind of unification should be possible for planning problems that involve sensing uncertainties (i.e., are unable to determine the current state). Presently, the methods in the literature are developed mainly around individual models and problems, as basic motion planning once was. Therefore, it is difficult to provide a perspective as unified as the techniques in Part II. Nevertheless, the concepts from Chapter 11 are used to provide a unified introduction to many planning problems that involve sensing uncertainties in Chapter 12. As in the case of the configuration space, some effort is required to learn the information space concepts; however, it will pay great dividends if the investment is made.
Chapter 12 presents several different problems and solutions for planning under sensing uncertainty. The problems include exploring new environments with robots, playing a pursuit-evasion game with cameras, and manipulating objects with little or no sensing. The chapter provides many interesting applications of information space concepts, but it should also leave you with the feeling that much more remains to be done. Planning in information spaces remains a challenging research problem throughout much of robotics, control theory, and artificial intelligence.
Steven M LaValle 2020-08-14