Robot navigation in unknown terrains via multi-resolution grid maps

Abstract

The authors propose techniques for navigation in an unexplored terrain that is arbitrarily populated with both concave and convex curved obstacles of unknown dimensions and locations. A method of robot navigation is presented which does not require a prelearned model and which can escape from concave dead-end corridors in local planning. It contains a component for learning that allows for continuous transition from local path planning to global path planning. The navigation process is composed of a number of traversals: each traversal is from an arbitrary source point to an arbitrary destination point. Initially, the terrain is explored using sensors, and the paths of the traversals may be suboptimal. A multi-resolution grid map is used to model the obstacle terrain, being incrementally constructed by integrating all relevant information. >