Developing a Robotic Tutor to Teach Math Skills to Children

INTRODUCTION Studies have shown that teaching methods can be facilitated with the use of robot-based technologies. Their effectiveness has been shown to approach, and often surpass, the effectiveness of human tutors. This can be attributed to their ability to achieve a higher level of engagement with children who respond better to technology-based teaching. This work discusses a novel approach for a math-teaching robotic platform with a simulation of the humanoid robot, named Robotis OP2. IMPLEMENTATION We developed a novel game-based math activity, targeting children between the ages of 4-7 years. This multi-level game presents questions ranging from number/shape identification to basic arithmetic to more difficult word problems as the child progresses from one level to the next. A deck of cards with all the possible answers to the questions is placed before the child. The child selects one and places it in front of the robot when he has an answer. The robot uses image processing (shape detection) to read this card, evaluate it and respond to the child. Shape detection module identifies the type and number of shapes on the card. It is implemented in C++ using Visual Studio. It implements the following steps: 1) convert image to gray scale 2) count number of edges using OpenCV’s findContours() function, 3) identify detected shapes as square, triangle or circle and 4) count the number of each shape. For enhanced attentional focus and a more natural social interaction, we also used a face detection to ensure the robot only asks a question when he child is present in its field of view. In addition, we also designed and implemented a novel Electrodermal Activity-based learning algorithm that allows the robot to estimate the child’s internal state. This, in turn, enables it to adapt its teaching method, depending on whether the child is nervous or calm, to achieve an intelligent, individualized teaching process. We have successfully implemented this interaction as a simulation. We have also proposed possible qualitative and quantitative measures that can be obtained to assess the child’s progress. We have also designed questionnaires to record the child’s feedback. This robot-based tutoring framework is developed using 3D graphic simulator, Webots, and screenshots and videos are captured to present our work. Future work includes deployment of our application to the robot, and conducting a full-scale user study to evaluate the effectiveness of our robot as a math tutor for young children