Development of an EtherCAT-based motion control system and a semantic segmentation yaw control algorithm for quadruped robots

Abstract

This thesis consists of two parts. In the first part, upgrades were made to the electronic and electrical subsystems of the quadrupedal robot Laelaps. The second part involved the development of an algorithm for controlling the direction of an agricultural robot using deep learning neural networks. The robot is being developed under the supervision of Professor Evangelos Papadopoulos, in the Automatic Control Systems Laboratory of the School of Mechanical Engineering National Technical University of Athens. Regarding the first part of the thesis, the process of designing and implementing part of the new electronic and electrical subsystems of the quadruped robot is analyzed. The main design goal of the new architecture was improving response times of the EtherCAT system, as well as improving the electrical stability and the reliability of the motion control system. The aim of the second part of the thesis is to describe the design and implementation process of an autonomous navigation system using deep learning for agricultural robots in vineyards. In short, an image of the vine in front of the robot is captured and transformed in real time, through a deep learning model, into a segmented image used to control its direction. The machine learning model was implemented in the popular machine learning environment PyTorch. Then various techniques were applied in order to improve the accuracy of the model. To validate the correct operation of the system, through novel techniques, a dataset of images was created, from a specially configured room with a synthetic vineyard, on which the model was trained. Then, an algorithm to extract the robot path deviation error from the segmented image was designed and validated. This algorithm also controls the robot’s direction based on this error. Its development was mainly done on a pre-existing robotic platform and with the help of the synthetic vineyard. Finally, a simple procedure for identifying and changing the row in a vineyard was implemented.