Obstacle Identification by an Ultrasound Sensor Using Neural Networks
This paper presents a method for obstacle recognition to be used by a mobile robot. Data are made of range measurements issued from a phased array ultrasonic sensor, characterized by a narrow beam width and an electronically controlled scan. Different met
- PDF / 939,771 Bytes
- 5 Pages / 595.278 x 793.684 pts Page_size
- 24 Downloads / 219 Views
1
D. Diepl, A. Johannet 1 , P. Bonnefoy2 and F. Harroy2 LGI2P - EMA/EERlE, Parc Scientifique G. Besse, 30000 Nimes, FRANCE. 2 IMRA Europe, 220 rue Albert Caquot, 06904 Sophia Antipolis, FRANCE Email: [email protected]
Abstract
2
This paper presents a method for obstacle recognition to be used by a mobile robot. Data are made of range measurements issued from a phased array ultrasonic sensor, characterized by a narrow beam width and an electronically controlled scan. Different methods are proposed: a simulation study using a neural network, and a signal analysis using an image representation. Finally, a solution combining both approaches has been validated.
Ultrasound sensors are usually used as proximity sensors, but they lack bearing directivity which generally prevents us from obtaining any accurate information. In order to reduce this drawback we have proposed an original sensor including several individual ultrasound emitter-receivers [3,4]. The ultrasonic sensor concerned consists of an array of 7 transmitters simultaneously emitting acoustic waves at the frequency of 40 kHz (Figure 1). The phase of each emitter can be adjusted individually, so that the beam width of the resultant wave will have a restricted size, and its bearing direction may be fixed (Figure 2). Echoes coming from reflectors are detected by two receivers, and the reflectors' range and orientation can be determined by measuring the time of flight, i.e. the "time duration between the transmission and the reception of a signal. The sensor is thus analogous to a sonar system, upon whose main principles the ultrasound system was developed.
1
Introduction
The development of an autonomous mobile robot is still a difficult task. Generally three types of problems are studied: the first deals with locomotion (stability, efficiency) the second deals with reflex actions (obstacle avoidance) and the third with navigation in order to reach a goal. The major difficulties encountered in such a task is the extreme variability of the environment with which the robot interacts, and the noise inherent in the real world. Obviously nobody tries to develop a robot able to evolve in all types of environment but. the variability intrinsic to even a specific type of environment is sufficient to lead to a relative failure of the traditional methods of modelling [1]. In this context, the neural networks approach appears to be an alternative solution in which the robot learns to adapt to the environment rather than learns all the reactions to each possible event. Within the wide field of research dealing with the development of mobile robots, starting from works centred on obstacle avoidance [9], this study focuses on the neural identification of obstacles using an original ultrasound sensor.
G. D. Smith et al., Artificial Neural Nets and Genetic Algorithms © Springer-Verlag Wien 1998
3
The Ultrasonic Sensor
Simulation Study
The first part of the work consists of modelling the sensor and the echoes in order to find out by simula-
,,1 I:t:l{*...
gem
...
tran
Data Loading...
