Catálogo de publicaciones - libros

Emergent interactions that are expressed by the movements of two agents are discussed in this paper. The common coding principle is used to show how the mirror neuron system may facilitate interaction behaviour. Synchronization between neuron groups in different structures of the mirror neuron system are in the basis of the interaction behaviour. The robotics experimental setting is used to illustrate the method. The resulting synchronization and turn taking behaviours show the advantages of the mirror neuron paradigm for designing of socially meaningful behaviour.

Within the framework of a model of vision-based robotic grasping inspired on neuroscience data, we deal with the problem of object orientation estimation by analyzing human psychophysical data in order to reproduce them in an artificial setup. A set of ANN is implemented which, on the one hand, allows to replicate some neuroscientific findings and, on the other hand, constitutes a tool for slant estimation that can improve the reliability of artificial vision systems, namely those dedicated to analyze visual data inherent to the interaction robot-environment, such as in grasping actions. The implementation confirms the hypothesis that integration of monocular and binocular data for the extraction of action-related object properties can provide an artificial system with improved pose estimation capabilities.

This paper presents a biologically-inspired artificial vision system. The goal of the proposed vision system is to correctly match regions among several images to obtain scenes matching. Based on works that consider that humans perceive visual objects divided in its cons-tituent parts, we assume that a particular type of regions, called curvilinear regions, can be easily detected in digital images. These features are more complex than the basic features that human vision uses in the very first steps in the visual process. We assume that the curvilinear regions can be compared in their complexity to those features analysed by the IT cortex for achieving objects recognition. The approach of our system is similar to other existing methods that also use intermediate complexity features for achieving visual matching. The novelty of our system is the curvilinear features that we use.

A supervised version of dFasArt, a neuronal architecture based method that employs dynamic activation functions determined by fuzzy sets is used for solving support of the problem of inter-vehicles collisions in roads. The dynamic character of dFasArt minimizes problems caused by noise in the sensors and provides stability on the predicted maneuvers. To test the proposed algorithm, several experiments with real data have been carried out, with good results.

The issue of collision avoidance in road vehicles has been investigated from many different points of view. An interesting approach for Road Vehicle Collision Assistance Support Systems (RVCASS) is based on the creation of a scene of the vehicles involved in a potentially conflictive traffic situation. This paper proposes a neuro-fuzzy approach for dynamic classification of the vehicles roles in a scene. For that purpose, different maneuver state models for longitudinal movements of road vehicles have been defined, and a prototype has been equipped with INS (Inertial Navigation Systems) and GPS (Global Positioning System) sensors. Trials with real data show the suitability of the proposed neuro-fuzzy approach for solving support to the problem under consideration.

The evolution of an effective central model of action selection and behavioral modules have already been revised in previous papers. The central model has been set to resolve a foraging task, where specific modules for exploring the environment and for handling the collection and delivery of cylinders have been developed. Evolution has been used to adjust the selection parameters of the model and the neural weights of the exploring behaviors. However, in this paper the focus is on the use of genetic algorithms for coevolving both the selection parameters and the exploring behaviors. The main goal of this study is to reduce the number of decisions made by the human designer.

In this paper a hybrid approach for solving a robot global localization problem in an office-like environment is presented. The global localization problem deals with the estimation of the robot position when its initial pose is unknown. The core of this system is formed by a virtual sensor, capable of detecting and classifying the corners in the room in which the robot acts, and an NSP (Neuro Symbolic Processor) control that infers and computes the possible robot locations. In this way, the whole global self localization problem is tackled with a hybrid approach: a classic neurosymbolic hybrid system, composed of a weightless neural network and a BDI agent (it processes the map and build the landmark connections), a neural virtual sensor (for detecting landmarks) and a unified neurosymbolic hybrid system (NSP) devoted to the computation of the robot location on the given map.

A kinematic adaptive neuro-controller for trajectory tracking of nonholonomic mobile robots is proposed. The kinematic adaptive neuro-controller is a real-time, unsupervised neural network that learns to control a nonholonomic mobile robot in a nonstationary environment, which is termed Self-Organization Direction Mapping Network (SODMN), and combines associative learning and Vector Associative Map (VAM) learning to generate transformations between spatial and velocity coordinates. The transformations are learned in an unsupervised training phase, during which the robot moves as a result of randomly selected wheel velocities. The robot learns the relationship between these velocities and the resulting incremental movements. The neural network requires no knowledge of the geometry of the robot or of the quality, number, or configuration of the robot’s sensors. The efficacy of the proposed neural architecture is tested experimentally by a differentially driven mobile robot.

The use of the vision in humans is a source of inspiration for many research works in robotics. Attention mechanisms has received much of this effort, however, aspects such as gaze control and modular composition of vision capabilities have been much less analyzed. This paper describes the architecture of an active vision system that has been conceived to ease the concurrent utilization of the system by several visual tasks. We describe in detail the functional architecture of the system and provide several solutions to the problem of sharing the visual attention when several visual tasks need to be interleaved. The system’s design hides this complexity to client processes that can be designed as if they were exclusive users of the visual system. Besides, software engineering principles for design and integration, often forgotten in this kind of developments, have been considered. Some preliminary results on a real robotic platform are also provided.

Bio-Inspired and Neuro-Inspired systems or circuits are a relatively novel approaches to solve real problems by mimicking the biology in its efficient solutions. Robotic also tries to mimic the biology and more particularly the human body structure and efficiency of the muscles, bones, articulations, etc. Address-Event-Representation (AER) is a communication protocol for transferring asynchronous events between VLSI chips, originally developed for neuro-inspired processing systems (for example, image processing). Such systems may consist of a complicated hierarchical structure with many chips that transmit data among them in real time, while performing some processing (for example, convolutions). The information transmitted is a sequence of spikes coded using high speed digital buses. These multi-layer and multi-chip AER systems perform actually not only image processing, but also audio processing, filtering, learning, locomotion, etc. This paper present an AER interface for controlling an anthropomorphic robotic hand with a neuro-inspired system.