Gaze Trajectory Prediction in the Context of Social Robotics

Social robotics is an emerging field of robotics that focuses on the interactions between robots and humans. It has attracted much interest due to concerns about an aging society and the need for assistive environments. Within this context, this paper focuses on gaze control and eye tracking as a means for robot control. It aims to improve the usability of human-machine interfaces based on gaze control by developing advanced algorithms for predicting the trajectory of the human gaze. The paper proposes two approaches to gaze-trajectory prediction: probabilistic and symbolic. Both approaches use machine learning. The probabilistic method mixes two state models representing gaze locations and directions. The symbolic method treats the gaze-trajectory prediction problem similar to how word-prediction problems are handled in web browsers. Comparative experiments prove the feasibility of both approaches and show that the probabilistic approach achieves better prediction results.