Institutional Repository
Technical University of Crete
EN  |  EL



My Space

Visual recognition and writing with the NAO humanoid robot

Kavroulakis Dimitrios

Full record

Year 2020
Type of Item Diploma Work
Bibliographic Citation Dimitrios Kavroulakis, "Visual recognition and writing with the NAO humanoid robot", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2020
Appears in Collections


Artificial intelligence aims to create behaviors, either by software agents or by robotic machines, which exhibit some basic intelligence. Today, humans can build different types of robots in order to perform some of their actions; however, each such robot needs to be properly trained to perform them accurately and sometimes it may fail. This diploma thesis presents the development of a "human" behavior on the humanoid robot NAO, which involves the visual recognition of a handwritten text (word) by the robot and the writing of the same text on a whiteboard using a marker held in his hand. More specifically, the NAO has been trained with Convolution Neural Networks and a dataset of uppercase handwritten letters of the Latin alphabet to recognize letters. NAO was also trained in the writing of uppercase letters of the Latin alphabet through a process of manual guidance of the arm joints by the man-trainer and recording of the timed trajectories of the joints through the specialized Choregraphe software. The individual components of the target behavior were integrated using the Python programming language. During the execution of the final behavior, the NAO robot receives an image from its camera, processes it with appropriate image processing algorithms from the OpenCV library, applies the trained neural network and detects sequences (words) of handwritten letters that it sees written on the board. Then, the robot writes one by one the letters that have been identified on the whiteboard, performing the appropriate movements with the arm holding the marker, shifting its position each time, so that the same sequence of letters is reproduced correctly. This approach was completed successfully, since the NAO robot achieves high percentages of accuracy in the correct visual recognition of the handwritten letters given to it, but also in their correct, legible writing with subtle differences. This behavior, which resembles children's learning processes, can be a point of interest as an interactive technological demonstration at STEM events for kids.

Available Files