Annual Review in Automatic Programming最新文献

It has been developed, built and tested an artificial vision based system to follow roads, which provides control signals, in a short time, by means of a joint of artificial neural nets. The image is segmented in “road” or “not road.” The obtained segmentation is the input for two neural nets, a classic architecture net (NN), and a TDNN (Time Delay Neural Network) one. The outputs provided by both nets, with a trajectory estimation, are introduced to a decision-making block, which selects the alternative containing less error. The classifier parameters are updated according to the current segmentation.

In the design of autonomous computer-based systems, we often face the embarrassing situation of having to specify, to the system, how it should carry out certain tasks, which involve computations known to be intractable or are suspected of being so. To circumvent such impasses, we resort to complexity-reducing strategies and tactics which trade some loss of accuracy for significant reductions in complexity. The term computational intelligence refers to such complexity reduction methods and to the research aimed at identifying such methods. In this paper we describe briefly some of our own work in this area and then develop a computational intelligence view of the task of process monitoring and optimization, as performed by autonomous systems. Some important current fields of discovery in computational intelligence include neural-net computing, evolutionary programming, fuzzy sets, associative memory and so on.

A concept of control for municipal waste water treatment plants is developed. The used fuzzy approach to real-time control aims at an operation of the plant that guarantees satisfactory performance of cleaning in different situations of load. The problem of nitrogen removal using flexible nitrification and denitrification volumes is tackled by means of fuzzy control. Important aspects of a simulation based analysis of the process including the structure of control are reported. Special emphasis is laid on the results obtained for conventional and fuzzy control in respect to a reduction of total inorganic nitrogen and lowering of running costs.

This text describes our generic approaches to monitoring and prognostic, emphasizing the application of learning techniques, and focuses on model-free specific supervision entities that can be realized by a learning-from-examples method. All necessary tools for the generation of a supervised learning of a process situation classifier will be outlined. Statistical feature selection and inductive numerical learning constitute the basis for the proposed architecture. A particular supervised nonparametric learning method, developed in-house, the Q^∗ -algorithm will be presented. Practical experiments for the monitoring of a lathe are carried out.

MORSAF is a real-time expert system designed to assist the operator of the SARAS oil refinery in the daily management of the topping plant. The main functionality provided by MORSAF is fault diagnosis, which is performed by an algorithm which reasons upon the faulty behaviour of the topping process represented through causal networks. Moreover, a prevention reasoning module has been designed to detect any unexpected behaviour by monitoring a set of significant process variables. Finally, a prediction functionality has been integrated to allow the filtering of redundant pre-alarms.

Since MORSAF has been developed as the demonstrator for the REAKT Esprit projects, its architecture strongly depends on the one of the REAKT toolkit, which is based on the blackboard model of multi-agents co-operation.

In this paper, we study the robust stability problem for a feedback system with a fuzzy controller. The problem is treated using the Small-Gain and Conicity Criteria for nonlinear stability. Robustness is addressed in terms of multiplicative perturbations and stability margins. Compensator design is based on an initial set of rules (the previous design), reflecting an approximate formulation of the desired performance. The previous design is refined and modified to achieve a prescribed robustness margin. We impose in the design procedure that the required changes have to be as small as possible. In this way, the final solution is the better compromise between performance and robust stability. The redesign is based on inequalities on the fuzzy parameters (ouput centroids) that are derived from Small-Gain stability conditions. The proposed method has been successfully tested on the simulated model of an inverted pendulum.

As part of the ESPRIT Projects HINT, a method for designing user interfaces, tailored for applications based on the HINT architecture, has been developed. The method consists of a set of steps, each with well defined objectives and results, and associated guidelines.

This paper presents an approach to integrated acquistion, evaluation, tuning, and execution of elementary skills in order to ease programming and employment of robots. To allow the use of machine learning techniques as well as real-time execution of the skills, a twofold representation is developed. The individual learning tasks involved in the generation and maintenance of skills are identified and related to proper learning techniques. A programming environment and a low level architecture taking care of the different requirements during all phases of skill acquisition and execution are described.

A Transputer based artificial vision system that allows estimating a mobile robot's absolute position (navigation), it's motion parameters (egomotion) and a depth map of the sight of view is presented. Navigation is achieved by tracking expected features present in basic 3D CAD information of the environment. Egomotion and relative depth are estimated using optic flow calculated on closed contours of points in the scene presenting high spatio-temporal gradients and require no a-priori knowledge.

It is widely accepted that visual programming techniques can offer increased productivity and reduced costs during the programming stage of real-time systems development and to a lesser extent at other phases in the system development life cycle. It is this author's belief, however, that while visual programming languages have much to recommend them, their current usage is erroneous. Current trends will prevent visual programming languages from being successfully applied to complex real-time systems and in the development of correct systems. We present a case for visual methods of real-time system development, that recognize visual programming as a single component of a more complex development method; we conclude by briely describing one such method.