Feasibility Study on Eye Gazing in Socially Assistive Robotics: An Intensive Care Unit Scenario

Recently, there has been an increasing interest in the adoption of socially assistive robots to support and alleviate the workload of clinical personnel in hospital settings. This work proposes the adoption of a socially assistive robot in Intensive Care Units to evaluate the criticality scores of bedridden patients. Within this scenario, the human gaze represents a key clinical cue for assessing a patient’s conscious state. In this work, a user study involving 10 participants role-playing 4 levels of consciousness is performed. The collected videos were manually annotated considering 6 gazing directions and an open-source automatic tool was used to extract head pose and eye gazing features. Different feature sets and classification models were compared to find the most appropriate configuration to detect user gaze in this scenario. Results have suggested that the most accurate gazing estimation is obtained when the head pose information is combined with the eye orientation (0.85). Additionally, the framework proposed in this study seems to be user-independent, thereby encouraging the deployment of appropriate robotic solutions in real assistive contexts.