Using Dynamic Field Theory to extend the embodiment stance toward higher cognition

Learning to reach after learning to look: A study of autonomy in learning sensorimotor transformations

“…From computational perspective, DNFs are attractor dynamics, which stabilise the neural states against noise, fluctuations, and other states, competing for activation. These continuous in time and in space dynamics allow to couple the controlling architecture to real physical sensors and motors, while providing an interface to discrete, or symbolic, cognitive representations [8]. Using DNFs, the whole robotic architecture, including the perceptual and memory systems, behaviour organisation (or planning), learning, and motor control, is formulated as a single (but modular) dynamical system, leading to natural and seamless integration of different components.…”

“…Lateral interactions of the DNFs bring about the existence and stability of a localised-peak solution that is the computational basis for modelling elementary cognitive processes of categorisation, detection, selection, and memory in the DNF framework [8].…”

“…Recently, we have introduced a neural-dynamic architecture, based on the Dynamic Field Theory [8], that introduces learning and adaptation of the sensorimotor transformations in an architecture, capable to autonomously generate saccadic eye movements from visual input. We have demonstrated how the involved sensorimotor mapping between the retinal positions of the selected objects and the motor commands, needed to saccade toward these objects may be initially learned and constantly updated [9], [10], [11].…”

Learning Sensorimotor Transformations with Dynamic Neural Fields

“…Such peaks of activation are units of representation in Dynamic Field Theory [15]. Because of the stability and attractor properties of the DNF dynamics, cognitive models formulated in DFT may be coupled to real robotic motors and sensors and were shown to generate cognitive behavior in autonomous robots [14].…”

“…Dynamic Field Theory (DFT) is a particular flavor of the dynamical systems approach, which has been successful in application of the cognitive models to control of robotic behavior [14,11,19]. The core element in this framework are the Dynamic Neural Fields (DNFs) -activation functions defined over topological spaces, which characterize the state of the behaving agent and its representation of the environment.…”