Navigation By Combining Reactivity and Planning

x2 c3 c1 x1 Figure 4: Translating move operations in paths sponding sequence of move operations; this task can be performed with diierent modalities, as the same path can in general be described by several move operation sequences. We will investigate the extension of the interface in order to allow direct translation of move operation sequences into a format which is directable executable by the navigation system. 8 Conclusions The aim of this work is to build an intelligent system for navigation, featuring both planning capabilities and reactivity to external events and navigation failures. We have presented a exible two-layered framework where these capabilities can be combined and used by programming strategies for the reasoner. This provides abilities that are not available in most of the classical planners. By keeping the planning of paths (and move operators) independent of the particular low level system, we have deened a paradigm for controlling navigation which is actually independent of the low level features of the navigation module. The current implementation is at an experimental level. We plan to test performances of diierent strategies , to enhance the replanning activities, the reactions to failure, the reasoning capabilities and to tighten the communications between the navigation and the planning level to build a robust navigation system. References Bro86] R. A. Brooks. A robust layered control system for a mobile robot. The \high level" reactivity discussed in section 4 is a uniform tool for a exbile treatment of execution failures. In order to demonstrate this point let us consider the following global strategy for navigation control: rst, the position of the robot is acquired by issuing a query-position to the interface; then, the planner shortest-path is called, and the resulting path is given as input to the navigation command start-navigation. When navigation terminates, the robot position is queried again, and if it matches with the target position the strategies terminates. The interesting cases are when the robot position is diierent from the required one. If the robot is not on the spec-iied trajectory, it might have failed in recognizing a landmark, and lost its way: in this case a new path is determined to the target position from the current position, and then execution is activated. If the robot is on the speciied path, the interface is queried for the reason of the failure with query-failure-description. According to the result, diierent behaviours are possible. If …