Recognition and Expression of Emotions by a Symbiotic Android Head Daniele Mazzei , Abolfazl Zaraki , Nicole Lazzeri and Danilo De Rossi Presentation by: Kaixi Wu “Social Robots” • Humans are fascinated by robots that can understand and express emotions • Social Robots should be believable and acceptable, so as not to betray the expectations of the humans interacting with them • The FACE robot conveys emotion through facial expressions The FACE Humanoid Project This paper examines the research conducted on the FACE humanoid robot. The robot has sensors to perceive its outside world and follow conversations, and various expressions and behaviours to react to social cues. The FACE Robot The FACE Robot • FACE: Facial Automaton for Conveying Emotions • Built by David Hanson • Skull is 3-D printed in ABS (Acylonitrile Butadiene Styrene, a plastic material) • Skin is made of Frubber (A skin-like silicone) • Motor anchor points placed under skin, connected to metal cables and servo motors to control motion 32 Total Servo Motors Servo motors are actuators that can precisely control rotational or linear motion • Face: 25 servo motors • Neck: 4 servo motors • Eyes: 3 servo motors Facial Action Coding System • By psychologists Paul Ekman and Wallace Friesen in 1978 • Action Units represent points of contraction or relaxation of one or more muscles • Distinct configurations for the six universally accepted emotions (happiness, anger, sadness, disgust, fear, and surprise) AU Configuration for FACE Hybrid Engine for Facial Expressions Synthesis • HEFES is a service of FACE that takes the basic emotion inputs interpolates these on the “emotional plane” • Allows for generation of more realistic and less stereotypical facial expressions Gaze Controlling System • Human gaze is a strong non-verbal cue to a person’s mental and emotional states • Attention module selects most prominent target • Gaze-control system continuously adjusts the head and eye movements • To mimic human motions, the eyes always start to move slightly before the head does. FACE Cognitive Architecture • Based on Antonio Damasio’s Theory: • Inputs from sensors are converted to knowledge structures • Knowledge structures allow reasoning • Reasoning processes result in internal or external actions and new generated knowledge • Actions and new knowledge drive emotions and behaviors • CLIPS: an intuitive rule-based planning system that allows for quick reactions Experiment #1: FACE Expressive Believability Key Questions • “Is a humanoid robot able to convey expressions as well as humans?” • “Are there differences between facial expressions observed as 2D photos, 3D models or performed by a physical humanoid robot?” • “Studies investigating the recognition of different facial expressions state that positive emotions are recognized faster and may be visually simpler than negative facial expressions. Is this theory still valid with a humanoid robot?” • “Does the interpretation of humanoid robot expressions induce different psychophysiological state in comparison with 2D photos and 3D models?” The Experiment • 15 subjects aged 19 − 31 years were recruited for the experiment. • Stepwise Protocol: Subjects exposed to gradually more realistic stimuli (2D pictures, 3D models, and the actual robot) • 2D photos and 3D models were created for FACE robot and a human face, for each of the 6 expressions • Subjects were asked to recognize the emotional states • The subjects’ psychophysiolgical signals were also analyzed for hints of nervous system activity indicating challenging or strenuous tasks The Results • There is a better tendency to recognize the physical robot’s expressions than the 2D photos and 3D models of the robot or human • Interpretation of facial expressions of the FACE robot does not alter the subjects’ psychophysiological states differently from interpreting those of human 2D photos and 3D models Experiment #2: FACE Gaze and Tracking The Experiment • 11 subjects, aged 22-35, were recruited for the experiment • Showed participants videos of two-person social interactions • Tracked eye movements with professional eye tracker • Obtained gaze points with corresponding times • Same videos were shown to FACE robot • Compared human gaze behavior with FACE gaze behavior The Results The FACE gaze control system was able to replicate the human gaze 89% of the time. Experiment #3: FACE Behavioral Control The Experiment Preliminary tests conducted in various social scenes, to see the robot’s reaction to social cues Behavioral Model • “If no subjects are present in the robot’s field of view, the robot is annoyed and looks at the most salient point (in term of colours and shapes) of the perceived scene” • “If someone is present in the scene, the robot’s facial expression becomes neutral and the robot starts to follow the most important subject identified according to a ranking of the following social cues: hand gesture, distance, speaking probability, facial expressions” • “If someone invades the robot’s intimate space, it changes its facial expression to dislike keeping the attention on the subject” Demo Video https://www.youtube.com/watch?v=-6FVZsaDLVg Conclusion & Future Developments Conclusion: The Development Process • Sensory apparatus to perceive and interpret social world and social cues • Robot facial expression generation system to convey emotion through facial expressions • Human-inspired gaze model to track subjects in a scene • Hybrid control center that allows real-time control of robot behavior and a user-friendly interface Conclusion: Remaining Challenges • Limitation of input sensor used • Environmental noise • Data communication and synchronicity Conclusion: Future Developments • Better perception through improved sensors: • Wide FOV (Field of View) Vision Sensors • Touch sensors • A social behavioral interpretation system for better human-robot interaction • User assessment of robot behavior to see how natural the robot is perceived by humans • Integrating the FACE head with a robot body to show emotion through gestures Any questions?