Adaptive Behavior current issue

Editorial: Behavior and Mind as a Complex Adaptive System

Nolfi, S., Ikegami, T., Tani, J. — 2008-03-14

Perception for Action: Dynamic Spatiotemporal Patterns Applied on a Roving Robot

Arena, P., Fortuna, L., Lombardo, D., Patane, L. — 2008-03-14

In this article, we apply a bio-inspired control architecture to a roving robot performing different tasks. The key of the control system is the perceptual core, where heterogeneous information coming from sensors is merged to build an internal portrait representing the current situation of the environment. The internal representation triggers an action as the response to the current stimuli, closing the loop between the agent and the external world. The robot's internal state is implemented through a nonlinear lattice of neuron cells, allowing the generation of a large amount of emergent steady-state solutions in the form of Turing patterns. These are incrementally shaped, through learning, so as to constitute a "mirror" of the environmental conditions. Reaction—diffusion cellular nonlinear networks were chosen to generate Turing patterns as internal representations of the robot surroundings. The associations between incoming sensations and the perceptual core, and between Turing patterns and actions to be performed, are driven by two reward-based learning mechanisms. We report on simulation results and experiments on a roving robot to show the suitability of the approach.

Enactive Robot Vision

Suzuki, M., Floreano, D. — 2008-03-14

Enactivism claims that sensory-motor activity and embodiment are crucial in perceiving the environment and that machine vision could be a much simpler business if considered in this context. However, computational models of enactive vision are very rare and often rely on handcrafted control systems. In this article, we argue that the apparent complexity of the environment and of the robot brain can be significantly simplified if perception, behavior, and learning are allowed to co-develop on the same timescale. In doing so, robots become sensitive to, and actively exploit, characteristics of the environment that they can tackle within their own computational and physical constraints. We describe the application of this methodology in three sets of experiments: shape discrimination, car driving, and wheeled robot navigation. A further set of experiments, where the visual system can develop the receptive fields by means of unsupervised Hebbian learning, demonstrates that the receptive fields are consistently and significantly affected by the behavior of the system and differ from those predicted by most computational models of the visual cortex. Finally, we show that our robots can also replicate the performance deficiencies observed in experiments of motor deprivation with kittens.

Microslip as a Simulated Artificial Mind

Ogai, Y., Ikegami, T. — 2008-03-14

A microslip is a type of action hesitation we experience in everyday life, which highlights the gap between human action and machine action patterns. By proposing a simple computational model for microslips, we examine the microslip as an implicit parallel dynamics underneath human cognition. Here, an agent, given as a dynamical system of a simple neural architecture, takes one of two choices, whose neural net is evolved using a genetic algorithm. An evolved agent often shows a hierarchy of action primitives and intentionality, and the agent is sensitive to the subtle differences of the object's layout, which results in a complex basin structures in the action-selection landscape.

On the Coupling Between Agent Internal and Agent/ Environmental Dynamics: Development of Spatial Representations in Evolving Autonomous Robots

Gigliotta, O., Nolfi, S. — 2008-03-14

In this article we describe how a population of evolving robots can autonomously develop forms of spatial representation which allow them to self-localize and to discriminate different locations of their environment by integrating sensory-motor information over time. The evolving robots also display a remarkable ability to generalize their skill in new environmental conditions that they have never experienced before. The analysis of the obtained results indicates that the evolved robots come up with simple and robust solutions that exploit quasi-periodic limit cycle dynamics emerging from the coupling between the robot/environmental dynamics and a robot's internal dynamics. More specifically, the variations of a robot's internal states are governed by transient dynamical processes originating from the fact that these internal states tend to slowly approximate fixed attractor points, corresponding to different types of sensory states that last for a limited time duration and alternate while the robot moves in the environment.

Learning Multiple Goal-Directed Actions Through Self-Organization of a Dynamic Neural Network Model: A Humanoid Robot Experiment

Nishimoto, R., Namikawa, J., Tani, J. — 2008-03-14

We introduce a model that accounts for cognitive mechanisms of learning and generating multiple goal-directed actions. The model employs the novel idea of the so-called "sensory forward model," which is assumed to function in inferior parietal cortex for the generation of skilled behaviors in humans and monkeys. A set of different goal-directed actions can be generated by the sensory forward model by utilizing the initial sensitivity characteristics of its acquired forward dynamics. The analyses on our robotics experiments show qualitatively how generalization in learning can be achieved for situational variances, and how the top-down intention toward a specific goal state can reconcile with the bottom-up sensation from reality.

On Cognition as Dynamical Coupling: An Analysis of Behavioral Attractor Dynamics

Montebelli, A., Herrera, C., Ziemke, T. — 2008-03-14

The interaction of brain, body, and environment can result in complex behavior with rich dynamics, even for relatively simple agents. Such dynamics are, however, often difficult to analyze. In this article, we explore the case of a simple simulated robotic agent, equipped with a reactive neurocontroller and an energy level, which the agent has been evolved to recharge. A dynamical systems analysis shows that a non-neural internal state (energy level), despite its simplicity, dynamically modulates the behavioral attractors of the agent—environment system, such that the robot's behavioral repertoire is continually adapted to its current situation and energy level. What emerges is a dynamic, non-deterministic, and highly self-organized action selection mechanism, originating from the dynamical coupling of four systems (non-neural internal states, neurocontroller, body, and environment) operating at very different timescales.

Attractor Landscapes and Active Tracking: The Neurodynamics of Embodied Action

Negrello, M., Pasemann, F. — 2008-03-14

Behavior is the product of three intertwining dynamics: of the world, of the body and of internal control structures. Neurodynamics focuses on the dynamics of neural control, while observing interfaces with the world and the body. From this perspective, we present a dynamical analysis of embodied recurrent neural networks evolved to control a cybernetic device that solves a problem in active tracking. For competent action selection, agents must rely on the attractor landscapes of the evolved networks. Insights into how the networks achieve this are given in terms of the network's dynamical substrate, which highlights the role of the network's inherent attractors as they change as a function of the input parameters (sensors). We introduce some terminological extensions to neurodynamics to allow for a more precise formulation of how attractor changes influence behavior generation: in particular, attractor landscapes, which are the space of all attractors accessible through coherent parametrizations of the network (input stimuli), and the meta-transient, which resolves behavior by approaching attractors as they shape-shift. We apply these concepts to the analysis of interesting behaviors of the tracking device, such as temporal contextual dependency, chaotic transitory regimes in moments of ambiguity, and implicit mapping of environmental asymmetricities in the response of the device. Finally, we discuss the relevance of the concepts introduced in terms of autonomy, learning, and modularity.