Constellation, le dépôt institutionnel de l'Université du Québec à Chicoutimi

Résumé

The emerging needs of human beings are pushing manufacturing companies from mass production to mass customization. The occurrence of these new challenges leads to a change of scenario where the robot no longer works isolated from human to a scenario in which the robot collaborates with the human in the same workspace (collaborative robotics). Wearable sensors using inertial measurement unit (IMU) are widely used to capture human upper body gestures in which the set of gesture being recognize is very large. However, foot gesture approach is starting to gain some places in applications where human’s hands are occupied when interacting with robots. This study presents an insole-based foot gesture recognition method for cobot operation mode selection. The insole is composed of an IMU and four force sensors. The classification algorithm uses a support vector machine (SVM) classifier based on features extracted by means of dynamic time warping (DTW) applied to only one reference gesture signal. Five human participants are used for the dataset. As a case study, the system was interfaced in real time (real-time classification algorithm) using a Simulink 2020a scheme with Universal Robots UR5 (5 kg payload). The worst-case recognition accuracy is around 88%. The algorithm is able to adequately discriminate between 10-foot gestures by means of a wearable insole sensor incorporated into the insole. Moreover, this study shows that, the control gesture can accurately be recognized from other current activities such as walking, turning, climbing the stairs, and similar.