A disaster robot is used for crucial rescue, observation, and exploration missions. In the case of implementing disaster robots in bad environmental situations, the robot must be equipped with appropriate sensors and good algorithms to carry out the expected movements. In this study, a neural network-based terrain classification that is applied to Raspberry using the IMU sensor as input is developed. Relatively low computational requirements can reduce the power needed to run terrain classification. By comparing data from the Accelerometer, Gyroscope, and combined Accelero-Gyro using the same neural network architecture, the tests were carried out in a not moving position, indoors, on asphalt, loose gravel, grass, and hard ground. In its implementation, the mobile robot runs over the field at a speed of about 0,5 m/s and produces predictive data every 1,12s. The prediction results for online terrain classification are above 93% for each input tested.
Copyrights © 2023