Multimodal Episodic Analysis of Human Personality Traits for Personalized Behavior in Social Robots

This paper presents a longitudinal analysis of human personality for socially interactive robots. A combination of visual, verbal, and paralinguistic cues has been employed to predict human personality and generate a robot's personalized behavior in an episodic manner. Low-level perceptual features have been extracted with a stereo vision camera sensor and a wireless microphone. A higher-level interpretation of these features provides us with suitable cues for personality assessment. A Random Forest supervised learning algorithm has been applied to classify the personality label of each individual interacting with the social robot, Emah. Three personality traits: Extraversion-Introversion, Agreeableness-Self centered, and Openness-Traditionalist are considered in this paper. We utilized the long-term memory of the interactive robot to enhance the likelihood of personality estimation. Suitable personalized robot behavior has been generated in the form of speech, gestures, and facial expressions. The proposed system has been validated in real-world human-robot interaction scenarios. The evaluation results show promising improvement in the adaptability of the robot's behavior, provided multimodal as well as episodic personality assessment is employed.