S. Adarsh,∗ K. I. Ramachandran,∗∗ and Binoy B. Nair∗
Accuracy, neuro-fuzzy systems, membership functions, regression analysis, ultrasonic sensors
Ultrasonic are widely used for obstacle detection and range estima- tion. They offer a cost-effective solution compared with a camera, infrared, and RADAR-based sensing systems over short ranges. The sensor HC-SR04 is one of the most popular and commercially available ultrasonic sensors, widely used to address the problems in obstacle detection and ranging in robotics applications. In this paper, we propose a neuro-fuzzy based system that can improve the accuracy of the range estimated by the ultrasonic sensor across its measurement range. The proposed system could reduce the root mean square error associated with the range estimation of the sensor by a factor of 4. A brief discussion on the observed improvement in accuracy, sensitivity, and the calibration process undertaken is also presented. A simplified regression model is proposed as an out- come of this experiment. The simplified model is derived from the neuro-fuzzy system and is observed to be capable of offering lower error, compared with the conventional linear/non-linear regression models. The designed neuro fuzzy system was compared with other techniques such as Support Vector Regression and Artificial Neural Network. The proposed algorithm can be used to improve the accuracy of any range sensor used for mobile robot navigation.
Important Links:
Go Back
