Recognition of plants with CTFM ultrasonic range data using a neural network

The success of any mobile robot is dependant on its ability to perceive the environment in which it is working. This paper describes some work to verify the suitability of a new sensor as a landmark detector for a mobile robot. The landmarks that we are investigating are plants. An ultrasonic sensor continually transmits a frequency modulated signal. Echoes detected by a second transducer are demodulated with the transmitted signal to produce audio tones which are proportional to range. The spectrum of these tones is determined by the geometry of the object. An artificial neural network (ANN) is used to recognise plants. This paper discusses a series of experiments that prove that the machine perception system is independent of the range, size, and the orientation of the plant. This system has potential applications in industrial and office robots.