Seminar - Organic Computing Self-Organisation of OC-Systems Markus Franke 25.01.2006 Typeset by FoilTEX
Timetable 1. Overview 2. Characteristics of SO-Systems 3. Concern with Nature 4. Design-Concepts 5. Example of a SO-System 6. Future Research 7. References Typeset by FoilTEX 1 von 39
Overview Why have Self Organized Systems arised? 1. administration of individual systems increasingly difficult 2. autonomic components needed with the ability of dynamic composition opportunistic interactions 3. heterogeneous systems becoming increasingly connected 4. architects cannot intricately plan interactions among components = fundamental change required in how applications are formulated Typeset by FoilTEX 2 von 39
Overview Nature scale complexity heterogeneity dynamism and unpredictability Can these strategies inspire solutions? Typeset by FoilTEX 3 von 39
Overview Sociology Computer Science Artificial Intelligence Robotic Organic Computing Bionik Information Technology Mechanical Engineering Biology Figure 1: Organic Computing - an interdisciplinary field Typeset by FoilTEX 4 von 39
Overview What means Self Organization? System designed to manage it self without external intervention Biological system: e.g. the human body, the autonomic nervous system Social society: insects, birds and human swarm examples for distributed self-organized Systems: Seti@Home, Anthill Project http://www.cs.unibo.it/projects/anthill/ Typeset by FoilTEX 5 von 39
Overview Figure 2: crowds of people Figure 3: swarm of birds Figure 4: shoal of fish Figure 5: a hive Typeset by FoilTEX 6 von 39
Characteristics of SO-Systems Typeset by FoilTEX 7 von 39
Self-Emergence (1) The Whole is more than the sum of it s parts. characterised by[5]: 1. interaction of large numbers of individuals 2. without central control 3. system behaviour which has not been programmed explicitly into the individuals typical bottom-up effect (randomness order) claim for controlled emergence Typeset by FoilTEX 8 von 39
Self-Emergence (2) Examples[5] resonant circuit: resonance frequency experiment: candle moving robots Figure 6: Candle Moving Robots developing order from random starting distributions and random rules. (Rolf Pfeifer,Zurich) Typeset by FoilTEX 9 von 39
Tasks of SO-Systems Self-X properties incorporate new elements detect unresponsive nodes geographic independence of accessability transparency Typeset by FoilTEX 10 von 39
Advantages/Disadvantages Advantages of SO-Systems: flexibility robustness decreasing design expenditures Disadvantages of SO-Systems: safety: appearance of failures during adaption security Typeset by FoilTEX 11 von 39
Concern with Nature Typeset by FoilTEX 12 von 39
Social Life bird swarm/ ant colony Limited local information Set of simple individual rules Global structures which emerges accomplish some function inspiration from mode of operation of social insects also observable at human beeings Typeset by FoilTEX 13 von 39
Swarm Intelligence population of simple agents interacting locally with one another no centralised control structure Two of the most successful swarm intelligence techniques 1. Ant Colony Optimization 2. Particle Swarm Optimization Typeset by FoilTEX 14 von 39
Design-Concepts Typeset by FoilTEX 15 von 39
Observer-Controller Architecture[5] classical top-down design process is not really suitable emergence characterized as a bottom-up phenomenon new system architectures needed 1. transition to goal setting methods 2. best effort - can deliver sub-optimal results 3. provisions to guide system towards the optimum = solution: Observer-Controller Architectures Typeset by FoilTEX 16 von 39
Observer-Controller Architecture analogy to human brain limbic system - emotional colouring (observer-controller plays role of limbic system) example: in a car (ABS controller) works that way Goals Observer/Controller stimuli Execution Unit Guard reaction Figure 7: Basic Observer/Controller structure Typeset by FoilTEX 17 von 39
Observer-Controller Architecture mechanisms to avoid undesirable behaviour 1. Assertions 2. Guard system behaviour: preset objectives own decisions Tasks of an Observer-Controller Architecture[5] 1. Observe production system for appropriate function / environment 2. Control parameters of the production system (reconfiguration?) 3. Control the Guard Typeset by FoilTEX 18 von 39
Design Approaches Top-Down vs. Bottom-Up Approach top-down: starts at top-level (general commands) bottom-up: what is needed in detail? introduction of new levels of abstraction example: bottom-up concept A motor cart without adaptive capabilities must drive autonomous with a constant speed. Typeset by FoilTEX 19 von 39
example: Bottom-Up (1) 1. Interleaved sensing actuation and Task specified by the desired position in shaft encoder counts for the motor PID: V = K p e + K i eδt + K d ė Figure 8: example: bottom-up concept, lowest level of abstraction Typeset by FoilTEX 20 von 39
example: Bottom-Up (2) 2. Interleaved sensing, reasoning and actuation PID-parameters are adjusted depending on the situation Figure 9: example: bottom-up concept, middle level of abstraction Typeset by FoilTEX 21 von 39
example: Bottom-Up (3) 3. More complex behaviour for the cart two driving motors for each of the rear wheels Figure 10: example: bottom-up concept, highest level of abstraction Typeset by FoilTEX 22 von 39
Architecture of Autonomous Systems Functional Architecture vs. Operational Architecture well-designed architecture: implementation functionality independent from should be general important for reuse medium to compare different systems 1. Hierarchical Approach 2. Behavioral Approach Typeset by FoilTEX 23 von 39
Hierarchical Approach abstract model of the world decisions based on this model translated through several layers advantage: transparent control structure disadvantage: overhead Typeset by FoilTEX 24 von 39
Behavioral Approach main idea: break up control problem into goals without central intelligence multiple parallel data-flows paths advantage: controller independency, easy to extend disadvantage: inefficiency and unpredictability Typeset by FoilTEX 25 von 39
Functional Architecture Figure 11: Hierarchical vs. Behavioral Approach Typeset by FoilTEX 26 von 39
Operational Architecture (1) environmental constraints = systems capabilities 1. time that is needed to perform an operation 2. ordering (list or sequence of operations) visual aid: precedence graph Figure 12: Example of a precedence graph Typeset by FoilTEX 27 von 39
3. tasks and subtasks Operational Architecture (2) reference to an abstract activity implicit explicit description representation: task tree, AND/OR Graph Figure 13: Task representation Typeset by FoilTEX 28 von 39
Operational Architecture (3) 4. synchronization of jobs being parallel executed 5. bindings of tasks weak vs. strong structured environment vs. unknown environment 6. interruption and exceptions uncertainty effort of acquiring additional information exception activation of plan generation system types of errors: software, hardware, external Typeset by FoilTEX 29 von 39
Example of a SO-System Typeset by FoilTEX 30 von 39
Example Adaptive Decentralized and Collaborative Control of Traffic Lights urban areas: congestion of traffic networks centralized control structure NP-Complete problems: computing power customization for each application Typeset by FoilTEX 31 von 39
Goals What should be achieved? 1. TCS without central components 2. global optimisation 3. adaption to different environments 4. dealing with changing traffic situations 5. stability Typeset by FoilTEX 32 von 39
Idea spread computing power each node: control of single junction, gathering data communication to adjacent nodes stable control rapid response to changes in the environment Figure 14: Schematic view of a traffic network Typeset by FoilTEX 33 von 39
techniques 1. Simulated Annealing 2. Genetic Algorithms 3. Classifier Systems (FCS) Adaptive algorithm Figure 15: Controller as an artificial life form Typeset by FoilTEX 34 von 39
Classifier System rules(classifiers): certain situations (input variables) actions genetic algorithm (exploration of search space) set of rules not fixed (evolves over time) additional value for each rule: the age of itself (long term memory) reduction of reaction times Typeset by FoilTEX 35 von 39
Classifier System - Tasks a-priori knowledge mappings for real-valued input data compliance with constraints Figure 16: Structural view of a Learning Fuzzy Classifier System Typeset by FoilTEX 36 von 39
Future Research[5] principles of self organization Exploitation by methods and tools Practical usage in technical applications Theory of emergent systems Safety and lateral limitations Interaction with the environment Typeset by FoilTEX 37 von 39
Thank you for your attention! Typeset by FoilTEX 38 von 39
References [1] Michael Beigl, Paul Lukowicz: Systems Aspects in Organic and Pervasive Computing - ARCS 2005 Springer Publishing Company (2005) ISBN 3-540-25273-8 [2] Klaus Mainzer: Self-Organization and Emergence in Complex Dynamical Systems (2004) [3] F. Rochner, C. Mueller-Schloer: Adaptive Decentralized and Collaborative Control of Traffic Lights (2004) [4] Travis C. Collier, Charles Taylor: Self-Organization in Sensor Networks (2003) [5] C. Mueller-Schloer: Organic Computing - On the Feasibility of Controlled Emergence (2004) [6] C. Mueller-Schloer, Christoph von der Malsburg, Rolf P. Wuertz: Organic Computing Informatik Spektrum (August 2004) [7] Dr.rer.nat.Christophe Bobda, Prof. Dr. Rolf Wanka: foils refering to the lecture: Organic Computing [8] Nicholas M. Avouris, Les Gasser: Distributed Artificial Intelligence: Theory and Praxis Kluwer Academic Publishers (1992) ISBN 0-7923-1585-5 [9] Edmund H. Durfee: Coordination of Distributed Problem Solvers Kluwer Academic Publishers (1988) ISBN 0-89838-284-X Typeset by FoilTEX 39 von 39