Evaluating the robustness of an appearance-based gaze estimation method for multimodal interfaces.