The Autonomous Robot Architecture (AuRA) provides multi-level representation and planning capabilities. This paper addresses the task of navigational path-planning, which provides the robot with a path guaranteed to be free of collisions with any modeled obstacles. Knowledge supporting visual perception can also be embedded, facilitating the actual path traversal by the vehicle.
A multi-level representation and architecture to support multi-sensor navigation (predominantly visual) are described. A hybrid vertex-graph free-space representation based upon the decomposition of free space into convex regions capable for use in both indoor and limited outdoor navigation is discussed. This “meadow map” is produced via the recursive decomposition of the initial bounding area of traversability and its associated modeled obstacles. Of particular interest is the ability to handle diverse terrain types (sidewalks, grass, gravel, etc.) “Transition zones” ease the passage of the robot from one terrain type to another.
The navigational planner that utilizes the data available in the above representational scheme is described. An A* search algorithm incorporates appropriate cost functions for multi-terrain navigation. Consideration is given to just what constitutes an “optimal” path in this context.