Resumen de Grounding spatial awareness in sensorimotor representations: an interdisciplinary approach
Beata Joanna Grzyb
The scientific community has widely acknowledged the necessity of a new approach to artificial intelligence, and especially to cognitive robotics, and proposed a new developmental perspective in an attempt to overcome the current limitations in robot design. Developmental robotics, known also as epigenetic robotics or autonomous mental development methodology, is a relatively new and highly interdisciplinary subfield of robotics. This approach embraces ideas from artificial intelligence, developmental psychology, neuroscience, and dynamical systems theory. Although there is a growing interest of the neuroscientific and psychological communities in cross-disciplinary studies, the highly interdisciplinary nature of developmental robotics, as well as fundamental differences in research goals, methodologies and language prevent a more prolific collaboration. In this regard, the main achievement of this thesis is a unified framework, that integrates the contributions coming from different disciplines. The discussion on the role of the body in cognitive robotics has been mostly focused on the ¿static¿ body in itself, that is what kind of physical instantiation of the body is necessary. The important role of the ¿body-in-motion¿ for cognition, so far, has received little attention. On the contrary, the importance of locomotor experience as a crucial factor of developmental change has extensively been postulated, for example, in child psychology. We suggest that experience of self-produced locomotion should be addressed more thoroughly in current research on embodied cognition and artificial intelligence if they are to move beyond the present bounds of the static realization of embodiment. This thesis intends to contribute in this direction.
The main contribution of this thesis is a truly interdisciplinary approach to the problem of the space (and body) awareness, in which the gap between neuroscience, developmental psychology and robotics is bridged at the computational level. We address the problem of space representation at two different levels. At the lowest level, the investigation focuses on understanding the neuroscience mechanisms of sensorimotor transformations required for the planning and execution of gazing and reaching actions that are suitable for interaction in near space. At a higher level, the investigation focuses on understanding, first, how infants build their near space representation, and then how this representation changes with infants¿ increasing action capabilities, and with the onset of upright locomotion in particular. Our behavioral studies reveal that infants¿ perception of reachable space changes between 9 and 12 months, and that change is related to the experience of upright locomotion. These findings add to a growing body of literature on infant development highlighting the complex, changing, and reorganizational nature of development. The new developmental phenomenon documented in this thesis provides a new set of experimental questions through which we may better understand the complex system of skills that underlie a goal directed reaching. We offer three main hypotheses on the possible reasons underlying developmental change in the perceived reachability of objects in older infants: (i) the decreased ability to learn from negative outcome while reaching makes infants fine-tune their walking skill, (ii) the processes responsible for integration of different visual depth cues reorganize themselves at the onset of walking so as to incorporate information from self-motion-based depth cues, (iii) the representation of space changes with the onset of walking; near and far space are being integrated with the reaching and walking actions to constitute a coherent space representation. These hypotheses have been modeled and their plausibility subsequently tested in a robotic setup. The results of robot experiments showed that these hypotheses are not mutually exclusive and overlap in underlying mechanisms, providing further evidences that goal directed reaching is a complicated skill with a long and protracted developmental course. We advocate that new impetus to robotics can be given from these studies aiming at improving the efficacy of contemporary robotic systems. From a pragmatic point of view, a robot should be able to purposefully and consistently interact with its environment, by grounding its skills on the integration of different stimuli. Such skills could be based on building a representation of its nearby environment, representation which can be exploited for more precise and complex interactions with the environment components. The representation of space should be plastic, and change with the acquisition of new motor skills to properly reflect current robot action abilities. The final outcome is a robotic system with the ability to autonomously build a coherent representation of the environment for purposeful exploration and actuation in both peripersonal and extrapersonal space, through the active interaction with the environment in a similar way as infants do. Such joint studies should advance robotics, and give some insights for further understanding of human cognitive development, and the nature of embodied intelligence more generally.
