Hanson Robotics Ltd, Shatin, Hong Kong
We describe a novel experimental framework, exploring humanlike cognition as a holistic, embodied phenomenon. The framework integrates a neuro-symbolic AI dialog ensemble, various soft robotics hardware with humanlike expressive face for social learning & communications, arms & grasping hands, locomotion, and many sensors, machine perception, & motion control tools. We present an alpha-platform architecture, and results from a variety of experiments including human-robot interaction, ensemble verbal & nonverbal dialogue interactions, mechanical tasks such as facial and arms controls, and in uses in the arts, therapeutic healthcare, and telepresence. The framework builds on the authors’ prior cognitive robots research, including Bina-48, PKD android, Zeno, Sophia, and others, adding new tools for designing cognitive AI, character animation, improved robotic embodiment, and a stable, mass-manufacturable hardware, and many open interfaces & extensions, to empower research into the complex, multidimensional factors giving rise to human cognition as physically-embodied phenomenon. The framework bridges fields from arts to mathematics, computer science, to psychology, integrating robotics tools from ROS and gazebo, computer animation tools including Blender and Unity, and various standard AI tools including many standard open source neural network and symbolic AI toolsets. We show the framework used in several intelligent humanlike robots with naturalistic expressive faces and bodies, and abilities to openly converse with people, see faces, learn, and build social relationships with people in HRI experiments. We describe new breakthroughs in soft robotics materials for actuation of expressive faces, and microfluidic force sensing in soft skin materials. We describe a novel cognitive ensemble verbal and non-verbal dialogue model along with basic emotional parameters for driving the agent’s goal pursuit and making the agent more intelligible and appealing to users. Using the framework to perform machine learning and reasoning in social interactions with users, we evaluate possible consciousness in the AI by assessing Tononi Phi values in the data as the system learns and pursues goals, and find the data show clear Phi signals during problem solving. The many results demonstrate progress in facilitating experiments in living AI, and uses in arts, science, healthcare, and educational outreach. We propose such an embodied human emulation robots and AI framework may serve next-gen breakthroughs in the study of human-level intelligence in AI and people.