DERVISH - An Office-Navigating Robot

ion The real world is so finely grained that any attempt to plan using a highly detailed model of reality is doomed because of enormous computational complexity. DERVISH avoids this pitfall by reasoning about an abstraction of the real world that is based on a quantization of the world into a set of states; each state corresponds to a node on the topological map or a hallway segment between two nodes. Figure 3 depicts the quantization of a hallway fragment. The area encompassing a topological node is labeled with a number, whereas the area between two topological nodes is labeled with a dashed number. The state set represents DERVISH’s positional uncertainty because it captures all the actual positions at which the robot believes it could be located. (We assume no uncertainty about DERVISH’s orientation once it has left the initial room.) Associated with each state is a certainty factor, which represents the relative likelihood of the robot being in the corresponding state. Note that the robot’s notion of state consists only of information about its own location. For example, the state does not contain information about whether particular doors Articles SUMMER 1995 55 Dictator Hallway Navigator Enter Room Escape Room AT_DOORWAY BLOCKADE ELSEWHERE LOST