ABSTRACT
In this section, based on physical reservoir computing, we propose a new concept of a wearable tactile sensor suit as a network of tactile sensors that can be used as a computational resource. As a case study, we exploited the network to emulate the periodic motions of a robot wearing a sensor suit with a nine-channel fabric-based sensor on its left arm. The collected data of the robot’s posture and tactile sensor states were used to train a linear regression (LR) model of the sensor states as readout modules that predict the next wearer’s movement using the current sensor data. Our findings illuminate that the LR algorithm’s performance is comparable with alternative methodologies that employ recurrent neural networks. This suggests that a fabric-based tactile sensor network can proficiently monitor the natural body dynamics of a robot. Furthermore, these spontaneous bodily dynamics can be harnessed as an effective computational resource.
