Gaze calibration for human-android eye contact using a single camera

In the same way as between humans, eye contact between humans and robots (especially androids) is an important element in nonverbal communication. To achieve eye contact in androids, we need to control the android's gaze to follow the planned trajectory. To do this, we have to estimate the relationship between the gaze direction and the input values of the actuator controller in an android's eyes. We call this estimation gaze calibration. In this paper, we propose a gaze calibration method for an android using a single camera as the gaze direction target in order to accurately adjust the android's gaze. First, we estimate the android's gaze direction from the orientation of its head and the camera, only using the relationships between each coordinate system. Second, we create datasets to estimate the parameters of the modeled relationship. We also propose a method to evaluate the effectiveness of the gaze calibration method. For the evaluation, we compare the subjects' perception of the calibrated android's gaze with a human gaze. Experimental results with this evaluation method show that the gaze calibration method makes the android's eye contact similar to a human gaze control for most of the test cases.