|
|
|
PROJECTSPHASE III
Digital on-demand Computing Organism: Stability and Robustness
The project was proposed by:
Summary:
The Digital on-Demand Computing Organism (DodOrg) project addresses the basic concepts of a new
computer principle inspired by biological concepts and their specific properties with respect to adaptability
and self-organization.
Continuing the work of the previous project phases, the focus of this application lies on stability and robustness:
while a stable system remains functional in the presence of normal changes during operation
(e.g. self-adaptation or self-optimization due to some changes in the environment) a robust system additionally
- to a certain extent - is able to tolerate even harsh changes like malfunctions, attacks, or corrupted data
through means of self-healing and self-protection.
In biological organisms, interacting closed control loops on cell level, tissue level, body level, or partial
system level are a major mean to achieve robustness. DodOrg adheres to this model by employing several
control loops, i.e. intra-cell loops, neighbor- and general inter-cell loops, intra-organ loops, and system wide
control loops. Like in biological counterparts these individual loops autonomously control different features
of the overall system, therefore seamless interaction of these control loops is a major research issue of this
funding proposal.
Robustness demands for novel techniques to enable our system to evaluate previous system states and decisions
to automatically and proactively tune individual components in the context of an overall objective. Additional
robustness is provided through an active immunology-inspired process, securing the system against
malicious malfunctions and deliberate attacks in a way comparable to dealing with infections and malicious
diseases in biology.
The proposed project comprises of four research groups and a biological expert: Prof. Becker (hardware
architectures), Prof. Brinkschulte (middleware), Prof. Henkel (thermal management), Prof. Karl (monitoring),
and Prof. Brändle (bio-inspired principles).
Emergent radio: Emergent strategies to optimise collaborative transmission schemes
The project was proposed by:
- Prof. Dr.-Ing. Michael Beigl, Braunschweig
Distributed and Ubiquitous Systems, Institute for Operating systems and Computer Networks, Technical University of Braunschweig
- Christian Decker, Karlsruhe
Telecooperation Office, University of Karlsruhe, Karlsruhe
Project website
Summary:
Cooperative and collaborative strategies for transmission in wireless sensor networks
enable transmission range restricted nodes to reach distant receivers by superimposing transmission
signals. This addresses an important practical problem of wireless sensor networks. In this
proposal we extend this strategy by emergent properties: We establish a method to adapt the
collaborative emergent optimisation process by a) remembering previous behaviour from similar
situations, b) using this information to adapt the current optimisation run by using randomised
and feedback-based approaches to determine an optimally pre-synchronised set of nodes for transmission
and c) optimising and learning observed optimisation behaviour for the random process,
which is better than the behaviour we had in memory so far. Using feedback information is a
natural and intuitive approach to adapt to the the scenario's dynamics without the requirement
for external intervention. Our approach will therefore show both emergent and self-organisation
properties. We will demonstrate the suitability of the method by implementing and deploying a
sensor network in an office setting. The demonstration will show how to globally minimise and
equalise the energy used for collaborative transmission.
The Bio-Chemical Information Processing Metaphor as a Programming Paradigm for Organic Computing III
The project was proposed by: PD Dr. Peter Dittrich, Jena
Bio Systems Analysis Group, Department of Computer Science and Mathematics, Friedrich-Schiller-Universität Jena
Project website
Summary:
This project develops a theoretical and practical framework to exploit the principles of biochemical
information processing as a programming paradigm for organic computing. By
doing so, we expect to make a technology available that allows to create computational
systems with the self-* properties of their biological counterpart. So far, we have designed
a workbench for chemical computing, developed a theoretically well grounded chemical programming
technique (called "organization oriented programming"), evaluated this technique
qualitatively on various problems like a chemical FLIP-FLOP and the maximum independent
set problem, studied structured molecules, and began quantitative evaluations. In the third
phase of the project we will switch our focus from design time to run time. That is, we
will develop methods that allow a user to influence an intrinsically self-organizing artificial
chemical system at run time. We will develop methods that consider continuous reference
values (e.g., fraction of cluster heads) as well as discrete goals to be provided by the
user at runtime. We will construct prototypic systems: one system will use a language like
Fraglets for structured molecules. Furthermore we plan to implement a chemical service on
a middleware like OC. Experimental studies should show how our methods scale to such
complex environments. Finally, we study apoptosis as a mechanism for organic computing
and show how our approach can be used to design a chemical program implementing an
artificial apoptosis decision mechanism.
Embedded Performance Analysis for Organic Computing
The project was proposed by: Prof. Dr.-Ing. Rolf Ernst, Braunschweig
Institute for Computer and Communication Network Engineering, Technical University of Braunschweig
Project website
Summary:
The project develops and investigates methods for performance analysis and optimization of
organic systems that are at least in part subject to real-time requirements. Compositional
analysis and optimization techniques that were successfully applied at design time are taken as a
basis to develop online methods that continuously observe and control the real-time constraints
in a dynamically changing system.
The proposed framework combines local analysis and optimization to an emergent, self-organizing
global observation and control network that scales with the system architecture and load situation.
The architecture is automatically determined while new sets of tasks are required to
announce their resource requirements. The framework uses a variety of techniques to avoid overload
and system failure, such as performance prediction, watch dogs, and shaping functions. The
techniques must be effective, even if the announced requirements are incorrect.
There are numerous challenges, from the framework architecture itself over computation time
bounds to algorithm iteration control to resource overhead minimization. A complete demonstrator
will be developed on a configurable set of microprocessor boards.
AUTONOMOS: A Distributed and Self-Regulating Approach for Organizing a Large System of Mobile Objects
The project was proposed by:
- Prof. Dr. Sandor Fekete, Braunschweig
Institut für Mathematische Optimierung Fachbereich Mathematik und Informatik Technische Universität Braunschweig
- Prof. Dr. Stefan Fischer, Lübeck
Institut für Telematik Technisch-Naturwissenschaftliche Fakultät Universität zu Lübeck
Project website
Summary:
The objective of our project AUTONOMOS is the development of a distributed and self-regulated approach for the self-organization of a large system of many self-driven, mobile objects. Based on methods for mobile ad-hoc networks using short-distance communication between vehicles, and ideas from distributed algorithms, local data clouds are formed in reaction to specific traffic structures (e.g., traffic jams). These Hovering Data Clouds (HDCs) are used for forming Organic Information Complexes (OICs) that are functional entities within the traffic flow, hosted by - but independent of - the individual moving vehicles (e.g., for detecting, indicating and monitoring the structure of a traffic jam that continues to exist in place, even as the involved vehicles axe replaced.) Using HDC-based OICs, we develop Adaptable Distributed Strategies (ADSs) for dealing with complex and changing traffic situations. A final goal is the extension to achieve Global Objectives for vehicles and traffic flow. Our project is based on a well-established interdisciplinary cooperation and combines practical know-how from the field of mobile ad-hoc networks with theoretical expertise'from a wide range of algorithmic topics. While the first three years of our six-year project dealt with largely local phenomena like traffic flow and their improvement, the goal for the second half is the extension of our work to even more complex concepts, methods, and structures, with an emphasis on overall scenarios.
Model-Driven Development of Self-Organizing Control Applications
The project was proposed by:
- Prof. Dr. Hans-Ulrich Heiss, Berlin
Fachgebiet Kommunikations- und Betriebssysteme, Fakultät für Elektrotechnik und Informatik, Technische Universität Berlin
- PD Dr.-Ing. Gero Mühl, Berlin
Fachgebiet Kommunikations- und Betriebssysteme, Fakultät für Elektrotechnik und Informatik, Technische Universität Berlin
- Dr.-Ing. Jan Richling, Berlin
Fachgebiet Kommunikations- und Betriebssysteme, Fakultät für Elektrotechnik und Informatik, Technische Universität Berlin
- Dr. Arno Wacker, Duisburg
Lehrstuhl für Verteilte Systeme, Fakultät Ingenieurwissenschaften, Universität Duisburg-Essen
- Prof. Dr.-Ing. Torben Weis, Duisburg
Lehrstuhl für Verteilte Systeme, Fakultät Ingenieurwissenschaften, Universität Duisburg-Essen
Project
website
Summary:
The MODOC project develops self-organizing algorithms and development tools for organic computing
(OC) applications. The goal of the project is to simplify the development of self-organizing
and self-stabilizing applications. To achieve this, MODOC follows a model-driven approach shifting
the inherent complexity of OC application development from the application developer to the tool
chain. The application domain of MODOC are actuator/sensor network-based control applications
which read data from sensors and in response control actuators covering a wide range of pervasive
computing (e. g., home automation and facility management). In the third phase of the project,
MODOC will focus on safety properties, on increasing the efficiency, and on extending the range
of covered applications. So far, MODOC-based applications give no guarantees while the system is
stabilizing or organizing. Now, we want to consider safety constraints required by many applications
to be satisfied all the time even in case of failure. Furthermore, we want to ease the development and
debugging of OC applications by round-trip modeling based on a reflective and adaptive middleware.
The range of adaption will newly cover distributed event composition and hybrid routing algorithms.
Finally, a more efficient application execution platform based on a self-stabilizing virtual machine
will be introduced.
Architecture and Design Methodology for Autonomie System on Chip (ASoC)
The project was proposed by:
- Prof. Dr. Andreas Herkersdorf, München
Institut für System- und Schaltungstechnik, Fakultät für Elektrotechnik und Informationstechnik, Technische Universität München
- Prof. Dr. Wolfgang Rosenstiel, Tübingen
Wilhelm-Schickard-Institut, Technische Informatik, Fakultät für Informations- und Kognitionswissenschaften, Universität Tübingen
Project
website
Summary:
In the first two phases we have shown that organic (hardware) systems composed of
cooperating autonomic subsystems can be designed and implemented such that they
have lifelike properties. We have demonstrated that organic computing has the potential
to provision VLSI systems with the capabilities of self-organization, selfhealing,
and self-optimization, thereby allowing them to adapt to their environment
and improve their functionality through online learning. The overall goal of phase 3 is
to prove the viability of the developed ASoC approach for designing and building selforganizing
and self-optimizing VLSI systems by means of operational prototypes
running real-world applications.
In phase 3 of our Autonomic System on Chip (ASoC) project, we will demonstrate
that our concepts are applicable with affordable overheads to the design and implementation
of dependable and robust ASoC prototypes for real-world applications. We
will advance the state of the art of reinforcement learning methods applied to VLSI
systems by theoretical and empirical analysis, which will allow us to direct the emergent
processes on the chip towards desired goals. Based on several use cases, we
will develop ASoC prototypes, which will be based on simulation as well as on FPGA
technology, and which will show the effectiveness and scalability of our approach.
Smart Teams: Local, Distributed Strategies for Self-Organizing Robotic Exploration Teams
The project was proposed by:
Project website
Summary:
The overall goal of the Smart Team project is to provide the algorithmic foundation for a selforganizing
robot team with local capabilities. The team's objective is to explore the terrain il is
deployed to and to execute some predefined work.
To do so, the team members have to communicate with each other, tasks have to be assigned and
the energy consumption of the team needs to be controlled. This must happen without central control
in a self-organizing way. Applications for a Smart Team are for instance expeditions in oceans or on
outer planets. Also possible are rescue actions in hazardous areas which are inaccessible for humans.
We assume that the robots have only a limited view on their environment and neither the robots
nor any other central instance knows the global state of the system at any time. We seek for local,
distributed strategies that lead to a good global behavior conceming all the tasks the Smart Team has
to perform. This is the main algorithmic challenge in this project. Furthermore, the strategies executed
by the robots are required to be robust such that minor failures are compensated. The tasks of such a
Smart Team are similar to the fundamental challenges of all social life forms: Explore, (self-)organize,
communicate and act jointly.
During the second phase, we have extended our work from the first phase in several directions:
We have developed the first asymptotically optimal strategy - the Hopper Strategy - for keeping an exploring
robot connected to its base camp with the help of mobile relays. In addition, we have designed
first strategies dealing with multiple explorers. Thanks to the development ofthe Smart Teams Simulator
in the second phase, we will be able to experiment with geographic exploration strategies in the
third phase. We have furthermore extended and wrapped up our work on collective graph exploration
and presented first models and complexity theoretic characterizations for the problem of assigning
tasks to robots under locality constraints. Another restraint for autonomous robot movement are energy
limitations. We have introduced algorithms to find a path minimizing the energy consumption
of movement and communication. Finally, we have started to investigate the localization of mobile
targets using only local information.
In the new phase, we plan to extend, evaluate, and unify our work. More specifically, we will
tackle the following problems:
- Exploration: We plan to figure out how to adapt our insights into graph exploration to geographic
exploration. This work will partly be experimental, using our Smart Teams Simulator.
- Maintaining Communication and Formations: We plan to further investigate how to keep multiple
explorer robots connected using mobile relay robots in terrains with obstacles. In addition,
we plan to develop new, more realistic cost models for maintaining communication. We have
observed that our communication problem for many explorers is closely connected to maintaining
formations in flocks of moving objects. We will therefore extend our work in this direction.
- Assignment: We plan to develop local algorithms that are able to assign groups of robots to
tasks such that the robots are collectively able to solve them. Our complexity theoretic insights
show that these problems are hard. Therefore we will try to develop local approximations and
will employ new variants of locality models to describe our assignments problems.
- Energy Consumption: We plan to further study the optimization of the joint cost for communication
and motion that involves multiple mobile relays. Moreover, we want to develop local
algorithms that exploit controllable mobility of robots to achieve an overall energy-optimal
configuration in Smart Teams.
- Unification: In the last phase we will work on combining our results on exploration, communication,
assignment, and energy efficiency. E.g., we plan to figure out how far the knowledge
about specific movement pattems, as they are given for example by (geographic) exploration
strategies, can be used to simplify strategies for maintaining a communication network.
Organic Traffic Control Collaborative
The project was proposed by:
- Prof. Dr.-Ing. Christian Müller-Schloer, Hannover
Fachgebiet für System- und Rechnerarchitektur, Fakultät für Elektrotechnik und Informatik, Universität Hannover
- Prof. Dr. Hartmut Schmeck, Karlsruhe
Institut für Angewandte Informatik und Formale Beschreibungsverfahren(AIFB), Universität Karlsruhe(TH)
- Prof. Dr. Jürgen Branke, Warwick
Operational Research & Management Sciences, Warwick Business School, The University of Warwick
- Prof. Dr. Jörg Hähner, Hannover
Fachgebiet für System- und Rechnerarchitektur, Fakultät für Elektrotechnik und Informatik, Universität Hannover
Project website
Summary:
Organic Traffic Control (OTC3) aims at the realisation of an organic traffic control system for
urban road networks. In the previous phases of the project, an adaptive learning intersection
controller has been developed that is equipped with collaboration capabilities to allow for selforganised
traffic light control in urban networks. In OTC3, the underlying observer/controller
architecture will be further advanced and refined. Additionally, the existing system will be extended
and improved with a novel route guidance and driver information component: Road
users will be provided with route recommendations that consider the current and predicted
state of the road network. Recommendations aim at minimising the individual travel times
of road users while preventing the formation of congestions. Route guidance and traffic
light control components will be integrated, so traffic lights can pro-actively adapt to recommended
routes.
Decentralised and hierarchical system architectures will be investigated and compared to
obtain a deeper understanding of the relationship between the local behaviour of a population
of system elements and the global behaviour of the system, which is a central question for
many Organic Computing systems. Both variants will be evaluated and compared, their
working principles and benefits will be demonstrated. In the evaluation, a special focus
will be on the robustness of the organic system with respect to unforeseen events and its
flexibility to incorporate the goals of a human planner.
OCCS - Observation and Control of Collaborative Systems
The project was proposed by:
- Prof. Dr. Hartmut Schmeck, Karlsruhe
Institut für Angewandte Informatik und Formale Beschreibungsverfahren(AIFB), Universität Karlsruhe(TH)
- Prof. Dr.-Ing. Christian Müller-Schloer, Hannover
Fachgebiet für System- und Rechnerarchitektur, Fakultät für Elektrotechnik und Informatik, Universität Hannover
- Prof. Dr. Jürgen Branke, Warwick
Operational Research & Management Sciences, Warwick Business School, The University of Warwick
- Prof. Dr. Jörg Hähner, Hannover
Fachgebiet für System- und Rechnerarchitektur, Fakultät für Elektrotechnik und Informatik, Universität Hannover
Project website
Summary:
In Phase III the project will refine and extend concepts and tools for the design of distributed
observer/controller architectures. These are necessary in order to design more complex and
collaborative self-organising technical systems, which are at the same time safe, robust, and
adaptive. In Phase I we have developed a generic observer/controller architecture, which
allows for controlled self-organisation in technical scenarios. By observing and analysing
the current state of the system under observation and control (SuOC), this architecture can
be used to prevent unwanted emergent behaviours and to encourage or enforce the desired
ones. A centralised observer/controller architecture was investigated in this context to validate
and refine the resulting concepts and methods. In Phase II we focused on complex
distributed scenarios. There, we refined the generic architecture into a distributed multilevel
observer/controller architecture to support the control of complex self-organising and
adaptive technical systems and investigated collective learning as part of the distributed
controllers. In Phase III we will extend this systematic approach by refining components of
the generic observer/controller architecture, not studied so far, and we will explore complex
collaboration mechanisms. We will investigate metrics to quantify robustness and flexibility
of OC systems, and we analyse and demonstrate systematically the improvements obtained.
For validation purposes and for the demonstration of practical usability, we will develop selected
multi-agent/multi-robot scenarios as testbeds. Moreover, we will apply our results to
the distributed application Organic Traffic Control, which will be developed in the corresponding
project OTC3, and the multi-robotics scenario from the project OC-TESTLAB.
Multi-Objective Intrinsic Evolution of Embedded Systems (MOVES)
The project was proposed by:
Prof. Dr. Marco Platzner, Paderborn
Institut für Informatik, Universität Paderborn
Project website
Summary:
This project aims at the investigation and development of intrinsically evolvable embedded systems. Simulated
evolution will provide such systems with a means to react properly to unforeseen changes in the
environment and the system resources. In an intrinsically evolved system the evolutionary process runs
together with the function under evolution on the same target platform. This is a necessary precondition
for autonomous operation. While evolutionary techniques have already been applied to the design of software
and hardware, intrinsic evolution as an adaptation method is a novel approach. We view intrinsic
evolution as a promising mechanism to provide autonomous embedded systems with self-adaptation and
self-optimization capabilities. We will achieve our goals by combining research in bio-inspired computing
with modern embedded system architectures.
The main research contributions of project phase III are the investigation of methods for parallel evolution
within a group of autonomous systems and the development of an architecture framework for evolutionon-
chip. Further, we will experimentally evaluate the developed methods and architectures on real-world
systems, including groups of autonomous robots, evolvable cache controllers, and active suspension systems.
The vision behind this project is that novel bio-inspired algorithms paired with modern system-onchip
architectures will allow us to construct future embedded systems that exhibit intelligent behavior.
A Modular Approach for Evolving Societies of Learning Autonomous Systems
The project was proposed by:
Summary:
In the previous funding period we developed a modular approach for realizing self-organizing
autonomous systems that show emergent behavior in societies of such systems. We investigated
how a system can learn to use its capabilities in changing environments while at the same
time paying attention to the overall group behavior. To improve the learning speed we combined
individual exploration with imitation of successful behaviors of teammates and demonstrated
our approach by simulation and with our Paderkicker robots.
In the next funding period we plan to use our modular approach to explore how the group
behavior can be modeled in such a way that the learning speed and the overall performance
can be improved while paying attention to the safety of the robots. This does not only include
the improvement of the usual behavior performance and safety. It also means to enable the
robot society to develop qualitatively new behaviors that were impossible to learn by the robots
individually, namely coordinated behavior without using user-defined coordination rules only
relying on subjective perception. This will be done using the previously developed approach for emergent behavior in order to extend it with the techniques developed for imitation by mechanisms
for coordination of robot societies in real world and simulation.
Formal Modeling, Safety Analysis, and Verification of Organic Computing Applications - SAVE ORCA
The project was proposed by: Prof. Dr. Wolfgang Reif, Augsburg
Institut für Informatik, Universität Augsburg
Project
website
Summary:
Original goals The overall goal of this project is to develop a method for the systematic,
top-down design, construction and analysis of highly reliable and adaptive Organic Computing
applications. Reliability here means preservation of functional correctness, safety under
unexpected disturbances and component failures. Adaptability in the context of this project
is related to adaptive system behavior under changing requirements and modified tasks. The
aim is to provide a formal framework for building self-x systems, which will make design and
construction of future organic systems easier and safer. As an integral part of the framework,
formal analysis will allow for behavioral guarantees despite the fact that the system
self-organizes. This will improve the quality, reliability, stability, and adaptability of such
systems. The approach combines safety analysis, formal specification and verification with
state-of-the-art software engineering methods and deals with the three major challenges of
construction and verification of self-x systems:
1. Reconcile self-x properties and guaranteed behavior, safety, etc.
2. Enable top-down engineering for modularity and reuse to achieve a cost-effective construction
principle
3. Measure the self-x capacity and compare it to classical solutions
State of project after the second phase The project is well on track. The basic results
of the first phase have been enhanced and based on a formal semantics during the second
phase. As a consequence, the system class has been substantiated with a precise semantics.
This allowed for (1) a rigorous definition of different self-x properties, (2) a generic implementation
of the reconfiguration paradigm with an off-the-shelf constraint solver and (3) first
steps towards a generic software engineering process for the construction of organic computing
applications. Concepts and ideas to solve all three challenges have been developed and
incorporated in the methodology. All new results have been applied to case studies. This has
also led to good and mutually helpful cooperations with other projects within the SPP. For
a more detailed overview of the results of the second phase please refer to Appendix A.
Goals for the next phase It is now necessary to combine and harmonize the results of
formal modeling and the software engineering approach. Furthermore, new aspects which
allow for better and more efficient system behavior - like e.g. intelligent strategies for role selection - will be investigated. The third focus will be on the decentralization of the observer/
controller that detects errors in the system and reconfigures it. So far, it is possible to
model and verify a centralized observer/controller and these techniques will be applied to a
decentralized one. Finally, the existing analysis methods will be extended to encompass additional
self-x capabilities, more specifically self-adaptation and self-optimization. All newly
developed concepts and the refinements of existing ones will be applied to a case study which
is based on an extensive platform of tools.
On-line Fusion of Functional Knowledge within Distributed Sensor Networks
The project was proposed by: Dr. Bernhard Sick, Passau
Fakultät für Mathematik und Informatik, Universität Passau
Project
website
Summary:
Humans learn from other humans { and intelligent computer systems should do that likewise! Humans
do not only learn facts but also rules. The latter can be seen as a more generic, abstract kind of
knowledge. We refer to these kinds of knowledge as \descriptive" and \functional" knowledge,
respectively. In the world of distributed, intelligent sensor networks, the exchange of descriptive
knowledge between sensor nodes, i.e. knowledge describing what is seen in a node's local environment,
has been investigated since several years. We will go a step further and investigate the exchange
of functional knowledge that re
ects how this local environment is observed and how a node reacts
on certain observations. In a steadily changing environment, where new knowledge arises or old
knowledge becomes obsolete, intelligent nodes adapt on-line to their environment by means of selflearning
mechanisms. The exchange of functional knowledge that is locally acquired but of global
importance will lead to a certain kind of self-optimization of the overall sensor network. Moreover, the
overall system will exhibit an emergent behavior. Within the scope of our project is the development of
techniques for the extraction, fusion, and insertion of functional knowledge. Additionally, methods
for the assessment of the quality, the novelty, and the reliability of functional knowledge will be
provided. Techniques for functional knowledge exchange are applicable in areas such as distributed
intrusion detection or driver assistance systems. Techniques for the detection of novel and obsolete
knowledge are not only necessary for knowledge exchange. They can also be used to realize various
kinds of self-x properties in organic systems.
Organic Computing Middleware for Ubiquitous Environments
The project was proposed by:
- Prof. Dr. Theo Ungerer, Augsburg
Lehrstuhl für Systemnahe Informatik und Kommunikationssysteme, Institut für Informatik, Universität Augsburg
- Dr. Wolfgang Trumler, Augsburg
Lehrstuhl für Systemnahe Informatik und Kommunikationssysteme, Institut für Informatik, Universität Augsburg
Project
website
Summary:
The rising complexity of smart environments in ubiquitous computing requires
new techniques for manageabihty. The Organic Computing (OC) Initiative
identified self-x properties as crucial for future ubiquitous embedded systems.
The proposed research project is based on the organic ubiquitous middleware
OC^^ designed and implemented mainly during the first two phases of
the DFG-SPP 1183. OC/A incorporates the self-x features of self-configiuation,
self-optimization, self-heafing, and self-protection.
So far, the self-x features of OC/i do not interfere or cooperate with each
other. A reorganization of the OC/i architecture allows a better integration of
the self-x features and the distinction of fast reflexes and slower planned reactions.
Therefore we want to develop and evaluate a two-level planning engine
to control OC systems in a generic, scalable, and goal-driven way. Furthermore,
we plan to pursue and extend the cooperation with other members of the priority
program, and to deliver a final version of the OC/i toolkit containing all
investigated features which may be used by other researchers.
Learning to Look at Humans
The project was proposed by: Dr. Rolf P. Würtz, Bochum
Institut für Neuroinformatik, Lehrstuhl für Systembiophysik der Ruhr-Universität, Bochum
Summary:
Artificial vision is a typical application domain of organic computing. Learning to interpret visual
input is still a great challenge. We will develop a system that can learn to find and track humans
in video sequences and to recognize individuals. The system will be based on a generic data
format specialized on the task only after learning and on a general matching mechanism. It
will segment and extract models from video sequences based on a schematic definition of
human figures and fuse these individual models to gradually self-organize a generic articulated
model. In the process, it will construct specialized subsystems for the recognition of body parts
and will integrate these subsystems with each other. To recognize individuals, the system will
form person-specific models emphasizing significant differences. While achieving the specific
vision goals, the project will realize many of the important objectives of organic computing in an
exemplary way.
Organic Fault-Tolerant Control Architecture for Robotic Applications
Prof. Dr.-Ing. Erik Maehle,
Lübeck
Institute of Computer Engineering, Universität Lübeck
together with:
Project website
Summary:
Mastering complexity is one of the greatest challenges for future dependable information
processing systems. Traditional fault tolerance relying on explicit fault models seems to reach its
limits to meet this challenge. During their evolution living organisms have, however, developed very
effective and efficient mechanisms like the autonomic nervous system or the immune system to make
them adaptive and self-organizing also in case of new unforeseen situations. These systems operate
unconsciously and in an emergent way to make the body self-protecting, self-healing, self-optimizing
and self-configuring. Inspired by these organic principles the control architecture ORCA
(Organic Robot Control Architecture) shall be developed in this project. Target platforms are complex
distributed embedded real-time systems, in particular autonomous mobile robots. In contrast to defining
fault models explicitly, the "health status" of the system is continuously monitored by so called
OCUs (Organic Control Units) which are closely attached to BCUs (Basic Control Units) implementing the
regular behaviors. Based on techniques like rule-based hybrid crisp-fuzzy systems or adaptive filters,
the OCUs are able to learn on-line and thus adapt to new unforeseen (fault-)situations in an emergent way.
To evaluate this approach, simulations as well as experiments with real climbing robots will be carried out.
Organic Self-organizing Bus-based Communication Systems
Prof. Dr.-Ing. Jürgen Teich, Erlangen
Department of Computer Science 12 (Hardware-Software-Co-Design), University of Erlangen-Nuremberg, Erlangen, Germany
Project website
Summary:
Today's electronic systems comprise a multitude of complex components interacting over communication networks and bus systems.
In this project, an organic approach for the analysis, design and optimization of bus-based communication systems is investigated.
The goal of our approach is to overcome drawbacks of today's pure offline designs that are based on worst-case estimations, are
not expandable, and may easily degenerate when the environment or requirements change at run-time. In contrast, a decentralized
approach using online self-organization would be able to monitor the actual traffic of the communication system and adapt either
sending rates, probabilities, priorities, etc. accordingly.
This project intends to provide a) theoretical foundations on self-organization for bus-based communication architectures based
on game theory and utility functions, b) models as well as a design methodology including learning techniques to implement such
properties for conflicting requirements, such as deadlines and bandwidth, and c) a simulation testbed. Finally, d) a hardware
demonstrator shall be developed in order to prove the benefits of the investigated approaches in a realistic environment.
|
|
|