Enhanced Robotic Teleoperation in Construction Using a GAN-Based Physiological Signal Augmentation Framework

Abstract

Communication channels between humans and robots can alleviate safety hazards (collisions, interference) during human–robot collaboration (HRC) at construction sites. Recently, the authors demonstrated the feasibility of improving HRC by using electroencephalogram (EEG) signals to establish hands-free, nonverbal communication. Despite the potential of EEG to provide reliable means of communication, there is a concern regarding the quality of the collection of EEG signals, especially low spatial resolution. EEG signals collected from wearable devices suffer from this problem because the number of electrodes (5–32) is much lower than traditional clinical EEG systems (64–256). More importantly, the low spatial resolution may reduce the reliability of human–robot communication driven by EEG signals. To address this challenge, this study seeks to increase the spatial resolution of EEG signals by proposing a generative adversarial network (GAN)-based data augmentation framework. In it, artificial EEG signals will be produced from actual signals. To examine the feasibility of increasing the spatial resolution of EEG signals for an improved HRC, the EEG dataset of four subjects was collected using a wearable 32-channel device. The results show that the framework can enhance the spatial resolution of the collected dataset by 39.1% by generating realistic artificial signals. It also increased the accuracy of the EEG-based robotic teleoperation by 4.9%. It is expected that increasing the spatial resolution of EEG signals will improve the reliability of EEG-based human–robot communication.