Do You Want to Play with Me? Preliminary Study of a Motor Control Walking Game with Pepper Robot

This paper presents a preliminary study on the implementation of a motor control walking game with the Pepper robot, aimed at addressing the challenge of ecological transposition of cognitive tasks. Inspired by the popular game “Red Light – Green Light”, we designed a game to assess motor cognitive control using social robots. The study aimed to identify the optimal combination of algorithms, input data, and time windows to accurately classify game errors and performance. Our findings suggest that a time window of 125 ms and specific features from accelerometer and MediaPipe skeleton keypoints data yield the best results. Additionally, we highlight the importance of feature selection, particularly emphasizing the sagittal component of the accelerometer and the lower body region. Furthermore, our study compares 2-class and 3-class classification methods, indicating that the former generally outperforms the latter in terms of accuracy. We discuss the implications of our findings for advancing cognitive assessment in social robotics and emphasize the need for further research to validate automated assessment methods and establish their reliability in clinical settings.