Chapter 19 – Robotics

Publisher Summary A difference between search in robotics and typical search testbeds in artificial intelligence, like the Eight puzzle or the Rubik's cube, is that state spaces in robotics are continuous and need to be discretized. Another difference is that robots typically have a priori incomplete knowledge of the state spaces. In these cases, they can predict with certainty neither which observations they will make after they have moved nor the feasibility or outcomes of their future moves. Thus, they have to search in nondeterministic state spaces, which can be time consuming due to the large number of resulting contingencies. Complete AND-OR searches minimize the worst-case trajectory length but are often intractable since the robots have to find large conditional plans. Yet, search has to be fast for robots to move smoothly. This chapter gives a general introduction to search problems in robotics, including search problems that arise if a robot has incomplete knowledge of its environment or its location in the environment, as is the case for mapping, localization, and goal-directed navigation in unknown terrain. It also discusses different ways of discretizing continuous state spaces.