Multi-robot systems (MRS) are a very active and important research topic nowadays. One of the main problems of these systems is the large number of variables to take into account. Due to this, robot behaviors are sometimes learnt instead of calculated via analytical expressions. A typical learning mechanism, specially for biomimetic robots, is Learning from demonstration (LfD). This paper proposes a LfD approach for implicit coordinated navigation using combination of Case-Based Reasoning (CBR) behaviors. During a training stage, CBR is used to learn simple behaviors that associate positions of other robots and/or objects to motion commands for each robot. Thus, human operators only need to concentrate on achieving their robot's goal as efficiently as possible in the operating conditions. Then, in running stage, each robot will achieve a different coordinate navigation strategy depending on the triggered behaviors. This system has been successfully tested with three Aibo-ERS7 robots in a RobCup-like environment.

This work presents a new approach to proactive collaborative wheelchair control. The system is based on estimating how much help the person needs at each situation and providing just the correct amount. This is achieved by combining robot and human control commands in a reactive fashion after weighting them by their respective local efficiency. Thus, the better the person drives, the more control he/she is awarded with. In order to predict how much help users may need in advance rather than waiting for them to decrease in efficiency, their skills to deal with each situation are estimated with respect to a baseline driver profile to increase assistance when needed. Situations are characterized at reactive level to keep a reduced set. This profile has been extracted from real traces of more than 70 inpatients presenting different physical and cognitive skills via clustering. The approach has been successfully tested at Fondazione Santa Lucia (FSL) in Rome.

Localization in multi-robot systems is a key problem in multi agent systems. In many cases, specially involving legged robots, like the Robocup soccer competition, it relies on predefined landmarks at known locations. However, when several objects are in motion, vision occlusions due to agents in the field of view make this kind of localization unfeasible for mild time periods. This also happens if landmarks are not within the field of view. This paper presents a technique to let a robot estimate its position with respect to objects or robots by sharing whatever visual data it has with its teammates. Shared data is used to estimate the relative positions of robots watching the same object via stereoscopy. The rest of the robots can be localized via triangulation and acquire information on the position of hidden or unknown objects that other members of the team can see. The system has been tested with two Aibo ERS 7 robots from Sony and an Aiball.

This work presents a new approach to shared control to assist wheelchair driving. Rather than swapping control from human to robot either by request or on a need basis, the system estimates how much help is needed in a reactive fashion and continuously produces an emergent motory command in combination with human input. To provide time stability and integration, instant commands are modulated by a factor depending on human efficiency in a shifting time window. Thus, the better the person drives, the more control he/she is awarded with. The approach has been tested at Fondazione Santa Lucia (FSL) in Rome with volunteers presenting different disabilities. All volunteers managed to finish a mildly complicated trajectory with door crossing and major turns and the proposed system increased efficiency in all cases.

Hybrid maps combine metric and topological information for efficiently managing large-scale environments. In a feature-based mapping framework, this paper describes the application of a spectral clustering approach for automatically detecting the transitions between subsequently traversed local maps. Contrary to recently proposed approaches, this algorithm considers each individual map feature as a node of a graph whose edges link two nodes if they are simultaneously observed. Thus, given a sequence of observations, an auxiliary graph is incrementally built whose edges carry non-negative weights according to the locality of the features. Given a feature, its locality defines the set of features that has been observed simultaneously with it at least once. At each execution of the mapping approach, the feature-based graph is split into two subgraphs using a normalized spectral clustering algorithm. If the graph partition is validated, the algorithm determines that the robot is moving into a new area and a new local map is generated. We have tested the proposed approach in real environments where features are obtained using 2D laser sensors or vision. Experimental results demonstrate the performance of the proposal.

In this paper, we describe a new approach which uses scale-invariant image features to estimate the motion of a stereo head. These point features are matched between pairs of frames and linked into image trajectories at video rate, generating what it is called visual odometry, i.e. motion estimates from visual input alone. With respect to previously proposed approaches, the main novelty of our proposal is that the matching between sets of features associated to stereo pairs and between sets of image features associated to consecutive frames are conducted by means of a fast combined constraint matching algorithm. Besides, the efficiency of the approach is increased by using a closed-form solution to estimate the final robot displacement between consecutive acquired frames. We have tested the proposed approach for navigational purposes in a real environment. Experimental results demonstrate the performance of the proposal.

Autonomous robots are capable of navigating on their own. Shared control approaches, however, allow humans to make some navigation decisions. This is typically executed either by overriding the human or the robot control at some specific situations. In this paper, we propose a method to allow cooperation between humans and robots at each point of any given trajectory so that both have some weight in the emergent behavior of the mobile robot. This is achieved by evaluating their efficiencies at each time instant and combining their commands into a single order. In order to achieve a seamless combination, this procedure is integrated into a bottom-up architecture via a reactive layer. We have tested the proposed method using a real robot and several volunteers, and results have been satisfactory both from a quantitative and qualitative point of view.

In this paper, an approach for fitting a circle to 2-D data which represent only a small part of the curve is described. This approach deals with the particular case where data is specified in terms of its Cartesian coordinates and the errors in both coordinates are not independent. Besides, an associated uncertainty ellipse which describes the measurement error and the variance matrix associated to the estimated parameters are obtained. This method is particularly well designed to fit a circle to a set of measured range readings provided by a 2-D laser range finder when these range readings are specified in terms of its Cartesian coordinates. Therefore, it has been successfully applied to acquire circle-shaped landmarks in a mobile robotics navigation task.

This paper presents a new CBR navigation scheme based on three behaviours: wall following, corridor following and door crossing. The system switches among CBR behaviours depending on a local grid built around the robot from sonar sensor readings. The main advantage of the proposed system is that it can easily adapt to any robot and environment, learning, both by training and by its own experience. The method has been successfully tested in real environments with a Pioneer robot equipped with 8 frontal sonar sensors.

Autonomous robots are capable of navigating on their own. In some cases, though, it is interesting to allow humans to influence navigation. Shared control is typically achieved either by giving control to the human or the robot at some specific situations. In this work, we propose a method to share control between humans and robots at each point of a given trajectory, so that both have weight in the resulting behavior of the mobile. This is achieved by estimating their respective local efficiencies at each time instant and combining their commands into a single order. In order to achieve a seamless combination, this procedure is integrated into a bottom-up architecture via a reactive layer. The system is meant to be used in wheelchair control for people with disabilities. Thus far, we have tested the proposed method using a Pioneer AT robot driven by several volunteers. Results have been satisfactory both from a quantitative and qualitative point of view.

This paper presents a geometrical feature detection system to use with conventional 2D laser rangefinders. This system consists of three main modules: data acquisition and pre-processing, rupture and breakpoint detection and feature extraction. The novelty of this system is a new efficient approach for natural feature extraction based on curvature estimation. This approach permits to extract and characterise line segments, corners and curve segments from the laser scan. Experimental results show that the proposed approach is very fast and permit to verify its effectiveness in indoor and outdoor environments

This paper presents a hybrid layered architecture for intelligent transportation. The system relies on integrating odometry, DGPS and compass readings for localization and deliberative planning, whereas the path is tracked at low level by means of a CBR based reactive navigation algorithm capable of adapting to the vehicle kinematics and environment conditions through learning. The system has been successfully tested using a 4-wheeled outdoors robot Pioneer AT

This paper presents a visual attention mechanism for the acquisition of landmarks in an arbitrary scene. The proposed mechanism consists of two consecutive selection stages. The first one employs classical preattentive saliency computa tions to select a reduced set of interest regions from the whole input image (data-driven stage). The second stage selects from the output of the first selection stage the region that can be considered as a potential landmark (model-driven stage). This potential landmark is the input of the attentive stage, that must characterize it and finally determine if this object is a real landmark. The used imaging sensor is a stereo vision system which is capable of providing depth data as well as color images. This stereo vision system is mounted on an autonomous mobile robot and serves map-building and localisation purposes. We present results achieved by applying the proposed visual attention scheme to on-line acquired stereo pairs of indoor and outdoor scenes.

This paper introduces a novel real-time human motion analysis system based on hierarchical tracking and inverse kinematics. This work constitutes a first step towards our goal of implementing a mechanism of human-machine interaction that allows a robot to provide feedback to a teacher in an imitation learning framework. In particular, we have developed a computer-vision based, upper-body motion analysis system that works without special devices or markers. Since such system is unstable and can only acquire partial information because of self-occlusions and depth ambiguity, we have employed a model-based pose estimation method based on inverse kinematics. The resulting system can estimate upper-body human postures with limited perceptual cues, such as centroid coordinates and disparity of head and hands.

In this letter, the ability of higher-order Hopfield networks to solve combinatorial optimization problems is assessed by means of a rigorous analysis of their properties. The stability of the continuous network is almost completely clarified: (1) hyperbolic interior equilibria, which are unfeasible, are unstable; (2) the state cannot escape from the unitary hypercube; and (3) a Lyapunov function exists. Numerical methods used to implement the continuous equation on a computer should be designed with the aim of preserving these favorable properties. The case of nonhyperbolic fixed points, which occur when the Hessian of the target function is the null matrix, requires further study. We prove that these nonhyperbolic interior fixed points are unstable in networks with three neurons and order two. The conjecture that interior equilibria are unstable in the general case is left open.

In object tracking, change of objects aspect is the most important cause of failure. This paper proposes a modified template matching approach to solve this problem without a priori learning of object views. This method permits to track non-rigid objects in real-time by employing a weighted template, which is dynamically updated, and a hierarchical framework that integrates all the components of the tracker. The capability of the tracking system to handle partial occlusions and target distortions is demonstrated for several video sequences.

In this work the authors show how the behaviour of Web users strongly affects the dynamics of TCP connections in Internet. Analysing actual and systematically generated HTTP traces, it is proved that the time between the download of two pages is critical to determine the re-utilisation of TCP connections. On the other hand, the study also shows that the utilisation of version 1.1 of the HTTP standard does not significantly affect the traffic generated by HTTP 1.0. In this sense, the heavy-tailed nature of the size of HTTP connections can be considered as an invariant property.

Research in mobile robot navigation has been based in approaches that integrate the metric and topological paradigms for mapping indoor environments. While metric methods produce accurate environment representations, their huge data volume and time complexity often prohibits efficient planning in large-scale indoor environments. On the other hand, topological maps can be used in a more efficient way, but accurate and coherent topological maps are often difficult to learn and maintain in large-scale environments. The paper describes an approach that integrates both paradigms: metric and topological. The metric map is learnt using a very simple scheme based on the works of Moravec and Elfes (1988). Then, a topological map is generated on top of the metric map by using a hierarchical structure. The main advantage of the proposed structure is that the partitioning of the metric map into coherent regions is achieved in an unsupervised manner with a low computational time. The paper gives results for autonomous exploration, mapping and planning of a Nomad200 mobile robot in large indoor environments.

This paper presents both a new sonar based place learning method and a global localisation algorithm based on finite Markov chains for autonomous robots. Landmarks are calculated by projecting the Fourier transform of the depth map obtained from a ring of equally spaced sonar sensors onto a bidimensional base of its vectorial subspace. Resulting landmarks can be acquired at any position of the environment and they do not depend on the robot orientation. Localisation relies on segmenting available landmarks into homogeneous regions and calculating transition matrices between them. Then, the position of the robot is estimated according to finite Markovian chains, but the probability of occupying a given region is modulated by the most recently acquired landmark. The method is valid for complex unstructured environments and it has experimentally proven to be fast, reliable and computationally cheap.

Presents a fast localisation algorithm for autonomous mobile agents in dynamic environments based on the definition of a very small sized landmark type. These landmarks are calculated by obtaining the coordinates of the circular depth function obtained from a ring of equally spaced sonar sensors projected on a bi-dimensional base of a vectorial space. Finally, a pyramidal structure is used to enhance and fasten the performance of the localisation algorithm.

This paper describes reconfigurable geometries for image sensors based on a concentric cartesian multiresolution lattice modified by four configuration parameters. They allow one, without moving the image sensor, to examine any region of the field of view with the highest available resolution, as well as to select the acuity profile for the regions surrounding the fovea. The efficient processing of the multiresolution images obtained requires discrete shifts of fovea and rings whose magnitudes are calculated. Real time foveal images have been preprocessed and examples are given.

Models for motion detection with artificial neural networks are inspired in physiological data of simple visual systems. Local speed estimation is a problem that involves more complicated architecture than a single lateral connection. We proposed an extension of these pioneer models. It considers bigger receptive fields that provide a larger zone of influence from receptors, allowing speed estimation on a wide range of velocities. The essence of this neural model is to estimate local speed from comparison between edge detection at the current position and delayed activities of neighboring input receptive fields.>