Online Terrain Classification Using Neural Network for Disaster Robot Application

A disaster robot is used for crucial rescue, observation, and exploration missions. In the case of implementing robots in bad environmental situations, must be equipped with appropriate sensors good algorithms to carry out expected movements. this study, a neural network-based terrain classification that applied Raspberry using IMU sensor as input developed. Relatively low computational requirements can reduce power needed run classification. By comparing data from Accelerometer, Gyroscope, combined Accelero-Gyro same network architecture, tests were carried not moving position, indoors, on asphalt, loose gravel, grass, hard ground. its implementation, mobile runs over field at speed about 0,5 m/s produces predictive every 1,12s. The prediction results online are above 93% each tested.