TURING PATTERNS IN RD-CNNs FOR THE EMERGENCE OF PERCEPTUAL STATES IN ROVING ROBOTS

Abstract

Behavior-based robotics considers perception as a holistic process, strongly connected to behavioral needs of the robot. We present a bio-inspired framework for sensing-perception-action, applied to a roving robot in a random foraging task. Perception is here considered as a complex and emergent phenomenon where a huge amount of information coming from sensors is used to form an abstract and concise representation of the environment, useful to take a suitable action or sequence of actions. In this work a model for perceptual representation is formalized by means of RD-CNNs showing Turing patterns. They are used as attractive states for particular set of environmental conditions in order to associate, via a reinforcement learning, a proper action. Learning is also introduced at the afferent stage to shape the environment information according to the particular emerging pattern. The basins of attraction for the Turing patterns are so dynamically tuned by an unsupervised learning in order to form an internal, abstract and plastic representation of the environment, as recorded by the sensors.

Reactive navigation through multiscroll systems: from theory to real-time implementation

“…The proposed reactive layer based on enslaving a chaotic attractor in periodic dynamics can be integrated in a complete architecture for the sensing-perception-action loop as already presented in Arena et al (2007), in which an adaptive and a contextual layer were developed to form a complete action-oriented perception methodology, based on complex dynamical system control. In particular the adaptive layer is used to create a plastic association between perceptual states and actions that can be modulated, during an unsupervised learning phase, by using a reward-based structure.…”

“…We have therefore, in the phase space, a kind of mirrored real environment, as measured by the sensors. The emerging controlled orbit will influence the behaviour of the robot by means of suitable actions, gained through the implementation of a simple unsupervised learning phase, that has already been presented in Arena et al (2007). Arena and co-workers developed a complete scheme, devoted not only to model the sensing-perception-action loop, but also to include memory of the past successful behaviours, in order to incrementally form a kind of contextual memory, helpful in case of time invariant environment.…”

The winnerless competition paradigm in cellular nonlinear networks: Models and applications

“…and the heavy regularity of the geometry of the classification regions ( Figure 9) suggests to use the FHN system for perceptual purpose [21]. The perceptual architecture is the same introduced in [17] and is made up of four main blocks (Figure 12): the sensing block that receives the external stimuli; the perceptual block that builds up a representation of the environment; the action selection table that triggers an action to the effectors; the difference of reward function (DRF) block, which evaluates the goodness of actions driving the learning process.…”

“…The output of the SNs is in any case bounded in the range [0, 0.1]. More details on the whole mathematical model are given in [17] and in [21].…”

“…In particular, the WLC network plays the role of perceptual core within an already developed control architecture [17], which models the sensing-perception-action loop, where, instead of using an RD-CNN, we use a WLC network to generate the perceptual states, and consider, in place of the emerging Turing patterns, the spatiotemporal patterns coded by the sequence of activated neurons.…”