A theory of sensor-based robot navigation using local information

Abstract

A definition of the general, multi- robots navigation problem is introduced. Successively, algorithms are described able to solve some subclasses of the problem, describing a new algorithm which guarantees a solution for the simplest case. This algorithm (the Wild Rover Algorithm), as well as the entire approach, stands on the idea that it is possible to integrate the analogical representation (easy to operate on-line on real sensors, with no problems of local consistency, but which does not guarantee, per se, about finding a solution) with a symbolic representation, on which sound search algorithms (particular forms of graphsearch) can carry out the planning itself and, eventually, optimization. With respect to the existing literature, our approach gives two relevant results: i) it defines an algorithm that, at the same time, under defined assumptions, is complete (it finds a solution if it exists), and is suitable for operating on a real robot in a real world; ii) generalizes navigation problem showing a formal way to face the multi robot problem, with possible partial solutions based on heuristics.

A well founded formalization of the navigation problem exists in Robotics for the Configuration Space approach to robot movement. However, we consider it too specific, as well as its related algorithms, to be used in the general case.