Sponsor:
Assess how research on the construction of cognitive functions in robotic systems is undertaken in Japan, China, and Korea Understand the relationship between robotics and the human-centered sciences (psychology, sociology, neuroscience)
Robotics as a Consumer of cognitive models and human data Data-driven methods Model-driven methods Collaborative approaches to cognitive modeling engage with human-centered disciplines Developmental approaches User-studies
Typically draw extensively from large data sets of human activity Sometimes generated by the individual research groups, sometimes from the larger community Often performed by non-experts in human experimentation Focused on particular robot applications/platforms Typically are one-way research projects Found throughout Japan, Korea, China
Intelligent environments Behavior Accumulation infers user goals and offers assistance Focus on sensing technologies Pictured (top) Robotic room 1 patient monitoring (bottom) Robotic room 3 clutter control
HRI/HCI component technology Focus on multimodal affect recognition from large Chinese corpus Pictured: (top) Automatic affect recognition system (bottom) vocal affect recognition pipeline
Cognitive functions to enhance the capabilities of service robotics Use of eye-tracking studies to guide robot performance Pictured HomeMate developed as a cognitive, consumer service robot
Typically exploit existing (published) models from human-centered disciplines Applications of models allow rapid development Typically no human participants until final system evaluation Sometimes novel approaches by linking models to other computational methods (e.g., crowdsourcing) Typically are one-way research projects Found throughout Japan, Korea, China
Expressive gestures and postures Emotion expression through music and performance Pictured (top) Kobian, an expressive humanoid and successor to the WABIAN-2R robot (inset) saxophone-playing robot; flute also performed
Robot Affective Mind: algorithms for emotion recognition, generation, and expression Draw from existing models of emotion Pictured: (top) Roti - a telepresence robot for English tutoring (bottom) Silbot - emotion generation
Focus on natural language understanding in complex domains Cognitive model of parsing, but boosted by web-based searching Pictured (top) KeJia a service robot with natural language processing (bottom) KeJia architecture External Knowledge Sources Sensors Human-robot dialogue Situated NLP Task Planning Motion Planning Domain KB Motion KB Robot Control
Focus on emergent properties of developmental systems Requires large, multi-faceted cognitive model Can be difficult to evaluate Often publish outside of robotics (Cognitive Science, ICDL, ICIS) Strong connections to human-centered research groups (primarily in the US) Most prevalent in Japan
Synergistic Intelligence experimentally understand and artificially construct human intelligence JST ERATO Asada Project ($21M USD) Pictured: Child-robot with Biomimetic Body (CB2 - top) M3-Kindy child robot (bottom)
Developmental Cognitive Robotics Goal: generate openended, emergent behavior for human social interaction Pictured: Noby, infant robot designed to display the cognitive capabilities and emergent behaviors of a 9-month-old
New research group Perceptual and cognitive functions for locomotion, and spatial navigation Inference and problem solving Pictured: PKU-HR3 humanoid robot
Build, refine, and evaluate cognitive models through direct user studies Often requires long-term deployments Uses complete operational system Often publish outside of robotics (CHI, CSCW, Human Factors) Strong connections to human-centered research groups (primarily in the US and Europe) Most prevalent in Japan
Deployment in public spaces (malls, train stations, schools) Algorithms for verbal and non-verbal communication and intention understanding Pictured: Robovie, which has been deployed for multiple years in public spaces
Robotics supporting Early Childhood Education (ECE) Cognitive models of learning and especially instruction Pictured: (top) QRIO interacting with children in a preschool (bottom) Model for a Care- Receiving Robot (CRR)
Collaborative physical tasks and remote instruction Collaborations with social scientists in Japan, UK, US Models of attention and behavior prediction Pictured: TalkTorque 2 most recent revision of the GestureMan series of robots
Cognitive modeling found in nearly every HRI research group But, there are a wide range of methodologies being used and an even wider set of approaches to these methodologies; very little standardization exists Extent of collaboration with human-centered disciplines varies significantly from very little contact to extensive collaborative projects Collaborators in the human-centered disciplines are primarily based outside of Asia