Gábor Samu’s research while affiliated with Budapest University of Technology and Economics and other places

Anytime systems can very advantageously be used when the resource and/or data availability is changing during the operation and some kind of intelligent reconfiguration of the system is needed to cope with the temporal data and resource access conditions. These schemes may provide an optimal tradeoff between the time/resource need and computational complexity and the quality (accuracy) of the results. Such systems can be achieved by using special types of models, methods, and algorithms together with applying active monitoring being able to supervise the operation of the system on-line and making intelligent decisions based on the sensory information of the so-called shortage indicators. Since the monitor operates at prescribed response time requirements and the number and complexity of the executable tasks can be very high especially in case of complex systems new considerations are needed to achieve optimal or acceptable performance. At computer level it means to application of special compilation methods dealing also with timing considerations and constraints of the underlying operating system and even their own run-time characteristics. This can also be supported by anytime development tools and special anytime description languages. In this paper a hierarchical compilation method is introduced together with theoretical considerations about a possible anytime development tool and the basics of the ATDL anytime description metalanguage are presented.

Anytime systems provide a tradeoff between the time resource, and computational performance, namely the quality of the results. Intelligent monitoring schemes are needed to achieve such a property and these schemes are supported by special compilation methods. Besides the global and local compilations, hierarchical compilation provides a theoretical base for the applicability of active monitoring of anytime systems represented by a directed acyclic graph. Active monitoring methods have to deal with timing considerations and constraints of the underlying operating system and even their own run-time characteristics. The theoretical basics of a new compilation method, the hierarchical compilation and the active monitoring scheme based on this proposed techniques are introduced.

We present a block library for Matlab/Simulink that allows fast prototyping of reconfigurable DSP systems. Up till now no similar software package was available. The block library supports the construction of reconfigurable discrete time linear and nonlinear systems from reconfigurable digital filters using various filter structures, state-space form implementations, polynomial filters, and PID controllers. We list the requirements for the block library and introduce the main implementation related decisions that allows the block library to meet these requirements. An example illustrates the usage of the block library.

Agents operating in real-time domain have to encounter resource i nsufficiency in some cases and applications even with deliberative and aware design. The res ource demands may vary depending on the operation phases, change in/of the environment and especially in alert conditions and in situations caused by sudden events. A s trong hardware base can be required for operating in dynamically changing environment. Some real-t ime agents applying sophisticated AI algorithms have to comply with timing constrai nts and continuous operation as well, leaving them to make a compromise between the consumption of the continuous and sufficient computation time and one or more performance properties. Anytime algorithms provide a tradeoff between t he time resource and computational performance, namely the quality of the results. This kind of behaviour is favorable in some domains, i.e. a less optimal decision can be better than a missing one. In case of complex systems usually a decomposition is applied during the design process, thus for the construction of a usable system the compilation of the operational unit s (modules) is also needed. The use of anytime algorithms as operational units assumes a compl ex software frame, an operation system which contains a so-called scheduler capable of distributing the computation time among the anytime and non-anytime algorithms. Thi s scheduler which is called as monitor in anytime systems has to deal wi th the timing considerations of many portions of the underlying operation system and even its own run- time characteristics to determine the time allocations while trac king the performance of the algorithms and the state of the environment. There is a gap between the im plementation of these anytime systems and the theoretical results as the compilation and the monitoring have. A survey of a new compilation technique, a monitoring scheme, and an implementation approach are covered by this paper.

The development of techniques for autonomous mobile robot navigation has been in focus for several decades. The main objective of this paper is to present a neuro-fuzzy algorithm. This algorithm can approximate the basic guiding models. In this paper the approximation of the vector field model is shown. The vector field is an extension of the potential based method. With the help of this neuro-fuzzy technique we can reduce the calculation complexity of the guiding algorithm.

Authors present an analysis about the consequences of complex measurement and signal processing tasks. They show that measurement and signal processing problems of now-a-days open new dimensions for the interpretation of the basic concepts of measurement and signal processing and make the reevaluation of these concepts necessary. Traditional methods fail in many cases to yield useful solutions, especially when measurement and signal processing problems reveal considerable complexity, involve a wide spectrum of various disciplines and require a multitude of components and methods. Since traditional methods, without a proper resource management supported at the system level, seem inappropriate to solve such problems, qualitatively new methods are needed. The present paper seeks answers to these problems. It gives a brief overview of various imprecise computational methods and discusses their applicability to treat complex measurement and signal processing problems.

This paper presents the design considerations of a modular hardware-software platform devoted to serve education, research and development in the fields of ambient intelligence, ubiquitous/pervasive computing, and networked embedded systems. The primary goal of this new platform is to support Embed-ded Information Systems Engineering education at the Budapest University of Technology and Economics, one of the few attempts world-wide to cover and in-tegrate the wide range of knowledge required in the field of embedded systems.

A kutatási program során intelligens jel- és képfeldolgozó, modellezési valamint diagnosztikai elemző módszerek kerültek kifejlesztésre, amelyek lehetővé teszik egy olyan automatikus ütközés elemző rendszer megvalósítását, amely baleseti szituációkról készült digitális fényképek alapján autonóm módon - emberi beavatkozás nélkül - képes a 3D modell, a deformációs energia és ezen keresztül az ütközés irányának és sebességének meghatározására. A rendszer alapját a digitális (jel- és) képfeldolgozás intelligens módszerekkel történő ötvözése képezi, ami lehetőséget nyújt a számítási idő csökkentésére és a pontosság növelésére. Alkalmazása (illetve továbbfejlesztése összetettebb esetek vizsgálatára) új információkhoz vezet, melyek ismeretében egy baleset automatikusan rekonstruálható, valamint a jármű biztonsági berendezései úgy alakíthatók, hogy az utas védelem minél hatékonyabb legyen. A kitűzött és megvalósított célok szorosan kapcsolódnak a számítógépes grafika illetve a számítógéppel segített intelligens modellezés és tervezés feladataihoz. A kutatásnak a modellezésben, jel- és képfeldolgozásban, a képminőségjavítás, információkiemelés, és 3D rekonstrukció terén elért eredményei a konkrétan megjelölt jármű-ütközés analízisen túlmenően olyan diagnosztikai, monitorozó és elemző rendszerek kialakításában is komoly szerephez juthatnak, amelyek széles körben alkalmazhatók a közlekedés, gépészet, robotika, építészet, egészségügy legkülönbözőbb területein. | As a result of the research project, new signal- and image processing, modeling, and diagnostics methods have been developed which together offer a way for automatic car crash analysis. This system is able to determine the 3D model of the deformed car, the deformation energy, the direction and speed of the impact based on digital photos made of the crashed car. The system combines digital signal- and image processing techniques with intelligent methods. By this, the computational time can de decreased while the reliability of the results increased. Using the car crash analysis system we can get into possession of new information by which the circumstances of accidents can automatically be reconstructed, and furthermore, safe car-body design can be achieved. The planned and implemented aims of the project are strongly related to the problems of computer graphics and computer aided modeling and design. Beyond car crash analysis, the results of the research in the fields of modeling, signal- and image processing, image quality improvement, information enhancement, and 3D reconstruction may get a significant role in developing diagnostics, monitoring, and analysis systems applicable at a wide range of other engineering problems, e.g. in transportation, mechanical engineering, robotics, architecture, and medical care.