Implementation of Artificial Neural Networks for Localization System on Rescue Robot

Abstract

Localization system is one of the most important parts in mobile robot especially in rescue robot. Because the rescue robot is used to search, help and guide disaster victims through the safest path to the evacuation point. So a good localization system will determine the robot's success in order to find a safe path when it comes to branching. This localization system usually consists of sensor systems and pattern recognition algorithms. The catastrophic conditions cause the camera sensor, RFID, or odometers readings to be less than the optimum due to the lighting factor and the number of obstacles from the ruins. Under these conditions the ultrasonic rangefinder and compass sensors are good enough to use because it have a bit of interference. The use of these sensors as localization systems produces a difficult data patterns to identify. To ease the identification or pattern recognition problem, so this research is proposed a back propagation neural network algorithm. The neural network is used to process the input of a robot (distance to the wall), and the direction of the robot to produce the predicted robot position in the room. The neural network consists of two hidden layers that have 10 input variables and 6 output variables. In this study the robot is tested in a labyrinth with 6 branches. From the obtained test results, the neural network is able to identify the robot position with an accuracy of 80.62%.