The trajectory motion of a robot can be a valuable information to estimate the localization of an autonomous robotic system, especially in a very dynamic but structurally-known environments like water pipes where the sensor readings are not reliable. The main focus of this research is to estimate the location of meso-scale robots using a deep-learning-based motion trajectory segment detection system from recorded sensory measurements while the robot travels through a pipe system. The idea is based on the classification of the motion measurements, acquired by inertial measurement unit (IMU), by exploiting the deep learning approach. Proposed idea and utilized methodology are explained in the related sections and it is observed that convolutional neural network approach is quite powerful to overcome the unreliability of IMU data.
Cihan Uyanik, Erdem Erdemir, Erkan Kaplanoglu, Ali Sekmen, "A deep learning approach for motion segment estimation for pipe leak detection robot", Procedia Computer Science, Volume 158, 2019, Pages 37-44, ISSN 1877-0509, https://doi.org/10.1016/j.procs.2019.09.025.