Daniela Corbetta - US grants

Proposal #: 12-29176
PI(s): Parker, Lynn E.
Arel, Itamar; Corbetta, Daniela; MAcLennan, Bruce J.; Reynolds, Gregory D.
Institution: University of Tennessee - Knoxville
Title: MRI/Acq.: Infant/Robot Grasp Learning Instrumentation
Project Proposed:
This project, acquiring equipment to create an instrument for social robot learning from human infant studies, aims to transform the development of human-interactive robots by applying fundamental scientific paradigms from developmental psychology studies in infants to new ways of designing autonomous robots. Specifically, the research utilizes data from human infant perceptual-motor learning studies in reaching and grasping to build comprehensive models of human sensorimotor control that achieve embodied learning like that exhibited by infants. These models will then be instantiated in a robotic setting to create a machine that can autonomously acquire shared object manipulation skills through bottom-up and top-down processes mimicking human infant learning. The equipment will facilitate multiple research thrusts within the grand scope of understanding and improving human-robot interaction. The thrusts include:
- Study of biologically-plausible visual attention mechanisms,
- Adaptive visually-guided motor skills, and
- Inference of human state.
Infant visual search patterns, from which models of perception will be developed, will be acquired via the eye tracking system. The robotic arm system will serve as a platform for extensive studies on perceptual motor skills. The instrumentation should enable the following outcomes:
- Yield robotic systems able to learn to physically interact with humans through shared object manipulation. These systems will learn skills that allow handling of previously unseen objects.
- Facilitate human-robot interaction, especially regarding manipulation, grasping, and handling capabilities, as well as eye-tracking, to better study the role of vision in infant grasp and reach learning.
- Contribute significantly to the integration of two bodies of research ? psychology and engineering ? coupled through the computational models built. These computational models will provide a two-way exchange of ideas between psychology and engineering,
- Lead both to the design of new experiments in psychology and to new mechanisms for interactive robots.
These systems will learn skills that allow handling of previously unseen objects. Facilitating robot interaction will permit to better study the role of vision in infant grasp and reach learning. These computational models will provide a two-way exchange of ideas between psychology and engineering leading to the design of new experiments in psychology and to new mechanisms for interactive robots.
Broader Impacts
The instrument, initially used by 7 faculty in 3 departments across 2 colleges, will offer new opportunities for cross-disciplinary training in the fields of cognitive psychology, developmental cognitive neuroscience, computer science, computer engineering, electrical engineering, and mechanical engineering, for both graduate and undergraduate students. The enabled research is also expected to offer significant practical societal impact, as many potential applications of human-robot interaction involve the exchange of objects (e.g., an assistive robot picking up a dropped TV remote control for a disabled person, a delivery robot handing a package to a human, or a therapeutic robot handing a toy to an autistic child). The work may have implications for addressing developmental problems in children, emanating from the increased understanding of the perceptual and motor learning processes in infants. Moreover, the instrumentation provides experiental and cross-disciplinary opportunities that buttress classroom theory, thus providing a positive impact to education, students, faculty, K-12 teachers, museums, etc.

DESCRIPTION (provided by applicant): The major objective of the proposed research is to develop and test a completely novel methodology to study infant cognition and action using multiple levels of recording and analysis. Infant visual attention toward graspable objects will be measured using eye-tracking technology combined with event-related potential (ERP) measures during visual inspection and immediately prior to reaching for the object. Of particular interest will be the relationship between neural activity and visual scanning in planning and executing successful reaches. 4.5- and 9-month-old infants will be tested to provide insights into the early development of controlled reaching. Aim 1 examines interrelationships between infant visual scanning patterns (measured with eye-tracking) and event- related potential correlates of infant visual attention during visual inspection of an object prior to reaching. It is predicted that olde infants will demonstrate greater integration of looking and reaching behavior than younger infants, and that infants will demonstrate greater consistency in looking and reaching behavior on attentive trials than inattentive trials (as indexed with ERP measures). Aim 2 examines the impact of visual scanning, and attention on the planning and execution of goal-directed behavior by identifying pre-reaching ERP components occurring immediately preceding the initiation of reaching movements. It is predicted that a pre- reaching component associated with voluntary initiation of reaching will be identified in the ERP that will be greater in amplitude on trials in which infants demonstrate consistency in looking and reaching behavior (representing planned reaches) than on trials in which looking and reaching are inconsistent (representing reactive reaching). Aims 1 and 2 will be addressed in a single experiment composed of simultaneous measurement of infant visual scanning, ERPs, and analysis of reaching behavior. Research utilizing these approaches has traditionally been conducted across parallel but separate research programs. Integrating these approaches into a single study will produce findings which will shed light on the attentional processes involved in action planning and execution, an aspect of infant behavior that has never been investigated before. The development of a feasible approach to simultaneous measurement of infant visual scanning, neural activity, and motor behavior will represent a major advance in the sophistication of psychophysiological tools available for research with preverbal infants.