الفهرس 
AbstractOnly 14 pages are availabe for public view
 |  | from | 93 |  |  |
| | | from | 93 | | |
Abstract
Humans can physically interact with the surrounding environments through their sense of touch. In addition to the image perceived by their vision, humans have the advantage of precisely recognizing the nearby items by using their fingers to explore, touch, and feel the surroundings.
Therefore, simulating the sense of touch in robotics is incredibly crucial for improving robotic artificial intelligent systems since robots play such a significant role in everyday human life. However, the approaches for obtaining accurate haptic perception continue to be a major study topic that challenges researchers in the robotic field.
In this thesis, research on detecting distinct 2D uniform shapes using various machine learning algorithms is conducted, with the optimal predicting approach being chosen by evaluating its efficiency, accuracy, and time. A reasonably affordable type of tactile sensor is examined in comparison to the other tactile sensors used before, as well as its efficiency in identifying the actual shape of objects.
To do so, the tactile sensor is attached to the arm robot`s fingertip, which encloses some pressure sensor elements known as Taxels. The collected datasets have been used as inputs to three different classification algorithms are called K-Nearest Neighbors Classifier (KNN), Naïve Bayes, and Support Vector Machine (SVM). Then, those supervised learning algorithms have been implemented to recognize the desired object shape from the collected data; where the best performance obtained with SVM using the Radial Basis Function (RBF).
One advantage of this study is the Taxel number within the sensor, which is fewer than any other type of tactile sensor used in the previous related studies as well as the new proposed exploratory technique followed to construct the scanned objects such as square, circle, triangle, and pentagon. As a consequence, the work reported here contributes to the whole set of tools required to develop the abilities of the artificial fingertip in terms of learning and perceptual dexterity.
Overall, the experimental methodology, using the sense of touch, described in this thesis has been validated in real situations and has proven its robustness and accuracy in autonomous perception and exploring the desired shape.