This paper represents the first steps in investigating issues emerging from a scenario where a robot interacts with a group of people around an interactive tabletop. In particular, the impact that a humanoid robot, acting as an avatar for a remote member of the group, has on the collaboration between the members of the group is investigated. This is carried out in the context where the avatar is depicting the actions being carried out by the remote user during a game. A preliminary study was performed to find out how the users interact when they are all co-located. The experiment is then carried out in two variants, in the first, a member of the group is moved to a remote location and allowed to interact with the other members using audio support alone. In the second, a robot is used to represent the actions of the remote user to the co-located users. The results obtained indicate that the addition of an avatar, used in such a way, has a positive impact on group interaction and allows for the task to be completed more successfully than when the avatar is not used. Further, comparative analysis of videos of the group behavior with a co-located team and group behavior with one member represented by the robot avatar, showed similar cooperative action.

Reinforcement learning (RL) is learning how to map states to actions so as to maximise a numeric reward signal. Fuzzy Q-learning (FQL) extends the RL technique Q-learning to large or continuous problems and has been applied to a wide range of applications from data mining to robot control. Typically, FQL uses a uniform or pre-defined internal representation provided by the human designer. A uniform representation usually provides poor generalisation for control applications, and a pre-defined representation requires the designer to have an in-depth knowledge of the desired control policy. In this paper, the approach taken is to reduce the reliance on a human designer by adapting the internal representation, to improve the generalisation over the control policy, during the learning process. A hierarchical fuzzy rule based system (HFRBS) is used to improve the generalisation of the control policy through iterative refinement of an initial coarse representation on a classical RL problem called the mountain car problem. The process of adapting the representation is shown to significantly reduce the time taken to learn a suitable control policy.

We present an experimental comparison between two approaches to optimization of the rules for a fuzzy controller. More specifically, the problem is autonomous acquisition of an “investigative” obstacle avoidance competency for a mobile robot. We report on results from investigating two alternative approaches to the use of a Learning Classifier System (LCS) to optimize the fuzzy rule base. One approach operates at the level of whole rule bases, the “Pittsburgh” LCS. The other approach operates at the level of individual rules, the “Michigan” LCS. In this work, both of these Fuzzy Classifier Systems were designed to operate only on the rules of fuzzy controllers, with predefined fuzzy membership functions. There are two main results from this work. First, both approaches were capable of producing fuzzy controllers with subtle interactions between rules leading to competencies exceeding that of the hand-coded fuzzy controller presented in this article. Second, the Michigan approach suffered more seriously than the Pittsburgh approach from the well-known LCS “cooperation/competition” problem, which is accentuated here by the structural combination of Evolutionary Computation and a fuzzy system. This problem was alleviated a little by the combination of a clustered subpopulation niche system and a fitness-sharing scheme applied to the Michigan approach, but still remains. © 2007 Wiley Periodicals, Inc. Int J Int Syst 22: 993–1019, 2007.

The issue of finding fuzzy models with an interpretability as good as possible without decreasing the accuracy is one of the main research topics on genetic fuzzy systems. When they are used to perform online reinforcement learning by means of Michigan-style fuzzy rule systems, this issue becomes even more difficult. Indeed, rule generalization (description of state-action relationships with rules as compact as possible) has received a great attention in the nonfuzzy evolutionary learning field (e.g., XCS is the subject of extensive ongoing research). However, the same issue does not appear to have received a similar level of attention in the case of Michigan-style fuzzy rule systems. This may be due to the difficulty in extending the discrete-valued system operation to the continuous case. The intention of this contribution is to propose an approach to properly develop a fuzzy XCS system for single-step reinforcement problems.

An analysis was performed to evaluate the strength of pin-loaded composite and aluminum joints. The analysis involved using three classifiers: decision tree, adaptive neuro fuzzy inference system and the combination of two. By using the well-known C4.5 algorithm, as a quick process, the structure of fuzzy inference system (number of membership functions and fuzzy rules) could be roughly estimated. Then, the parameter identification is carried out by adaptive neuro-fuzzy system. The comparison of performance of three methods indicates that mentioned hybridization speeds up learning processes and reduced errors.

In this article, we describe some of the important currently used methods for solving classification problems, focusing on feature selection and extraction as parts of the overall classification task. We then go on to discuss likely future directions for research in this area, in the context of the other articles from this special issue. We propose that the next major step is the elaboration of a theory of how the methods of selection and extraction interact during the classification process for particular problem domains, along with any learning that may be part of the algorithms. Preferably this theory should be tested on a set of well-established benchmark challenge problems. Using this theory, we will be better able to identify the specific combinations that will achieve best classification performance for new tasks.

This paper offers an experimental study of the influence of learning on evolution in populations of neural networks in which evolutionary and learning fitness surfaces are set and known in advance. Although not biologically plausible, this allows us to investigate various hypotheses regarding the interaction between evolution and learning in neural networks, such as "neighbourhood correlation" and "relearning", in easily controlled conditions. Experimental results are presented comparing the evolution of neural networks, with and without learning and on similar and dissimilar tasks. The results chart the evolutionary progress of neural network populations in terms of fitness at birth and fitness after lifetime learning on the different tasks presented and with different selection pressures.

This paper reviews experience in using a temporal fuzzy classifier system which explicitly represents time in the classifier syntax by augmenting individual classifiers with temporal tags. The contribution of each activated classifier to the composite system output is modulated in time according to the parameters of the fuzzy temporal tag associated with that classifier. This feature allows the learning algorithm—in this case, the genetic algorithm—to explore and exploit temporal features of the environment in which the classifier system might be expected to operate. Experimental results in applying the fuzzy temporal classifier system to control of a time-delayed plant and to the distributed problem of adaptive routing in communication networks are presented. © 1998 John Wiley & Sons, Inc.