Designing human-robot interaction algorithms for improving metacognitive skills

In today's technology, robots are now involved in every aspect of life. Therefore, it is unthinkable that robots should not affect the field of education. Robot simulators can help developers to test and improve intelligent systems before putting them into real environments. In this study, the use of robots in education is specifically addressed in terms of the development and prediction of metacognitive abilities. This thesis is organized around two research problems. First, a simulation environment is provided for testing and developing methods that follows learning models in the field of robot-human interaction. The robot simulation is fully programmed in its native language (i.e. LUA) of the CoppeliaSim software. The simulation is about a humanoid robot making lesson plans. The study group of the research consists of 64 (32 experimental, 32 control) academicians working at Zonguldak Bülent Ecevit University. The implementation process covers an eight-week course period. Some of the data were collected from the experimental and control groups through the MCQ-30 (Metacognition Questionnaire-30) scale. The other part was obtained only from the metacognitive strategy questions answered by the experimental group within the scope of lesson plan preparation with robot simulation. In this direction, results were obtained regarding the effect on the metacognition scores of the instructors who prepared lesson plans by receiving metacognitive strategy support and the opinions of the instructors about the robot simulation usage process. According to the results, the posttests of the instructors who were given metacognitive strategy use support with robot simulation increased compared to the pretest scores. On the other hand, there was no significant difference between the pre-test and post-test scores of the instructors in the control group who did not use robot simulation and followed the traditional process. It is seen that the applications organized with robot simulation are effective in the metacognition scores of the people in the experimental group. In the second research problem, a machine learning model was developed using a data set containing 250 participants and 32 features. The purpose of this model is to decide whether a person who performs the test consisting of 30 questions mentioned above needs to receive metacognition training, planning, monitoring and evaluation training with the simulation developed in the first stage of the thesis. The classification model mentioned in the thesis is tested with many algorithms such as CNN, Decision Trees and SVM results are presented comparatively.