Building Environment Maps using Neural Networks

N. Achour, R. Toumi, and N.K. M’sirdi


Sonar map, ultrasonic sensor, mobile robot, neural networks


The authors present an efficient and cheap approach to build a sonar- based mapping for autonomous mobile robot in indoor environments. The system uses one ultrasonic sensor (emitter and receiver are separated) mounted on a motorization constituted by two motors allowing horizontal and vertical scanning. The reading is modelled as probability profiles projected on a two-dimensional map. These readings provide information concerning empty and occupied areas in the sensor cone. The latter is the geometric representation of the sensor beam; it is composed of data allowing recognition of the environment structure. The world is modelled as a grid of cells. Two classifying neural nets are used to build the environment map, the first one for robot navigation in order to scan the workspace to obtain a database that is used by a second one to build the map. The result is a faithful representation of the real environment when the inputs are cones of data. The map can be used to plan paths and for navigation. Results from actual runs are presented at the end.

Important Links:

Go Back