1 Distributed Motion Coordination: From Swarming to Synchronization Ali Jadbabaie Department of Electrical and Systems Engineering and GRASP Laboratory.

Slides:



Advertisements
Similar presentations
Partitional Algorithms to Detect Complex Clusters
Advertisements

Coordination of Multi-Agent Systems Mark W. Spong Donald Biggar Willett Professor Department of Electrical and Computer Engineering and The Coordinated.
Polynomial dynamical systems over finite fields, with applications to modeling and simulation of biological networks. IMA Workshop on Applications of.
Midwestern State University Department of Computer Science Dr. Ranette Halverson CMPS 2433 – CHAPTER 4 GRAPHS 1.
Motion Planning for Point Robots CS 659 Kris Hauser.
INTRODUCTION to SWITCHED SYSTEMS ; STABILITY under ARBITRARY SWITCHING
Kick-off Meeting, July 28, 2008 ONR MURI: NexGeNetSci Distributed Coordination, Consensus, and Coverage in Networked Dynamic Systems Ali Jadbabaie Electrical.
THE ROLE OF LIE BRACKETS IN STABILITY OF LINEAR AND NONLINEAR SWITCHED SYSTEMS Daniel Liberzon Coordinated Science Laboratory and Dept. of Electrical &
Flocks, Herds and Schools Modeling and Analytic Approaches.
COORDINATION and NETWORKING of GROUPS OF MOBILE AUTONOMOUS AGENTS.
Ch 9.1: The Phase Plane: Linear Systems
Graph Laplacian Regularization for Large-Scale Semidefinite Programming Kilian Weinberger et al. NIPS 2006 presented by Aggeliki Tsoli.
Basic Feasible Solutions: Recap MS&E 211. WILL FOLLOW A CELEBRATED INTELLECTUAL TEACHING TRADITION.
Javad Lavaei Department of Electrical Engineering Columbia University Various Techniques for Nonlinear Energy-Related Optimizations.
Chapter 4: Network Layer
Direction Election in Flocking Swarms Ohad Ben-Shahar, Shlomi Dolev Andrey Dolgin, Michael Segal Ben-Gurion University of the Negev.
Lecture #13 Stability under slow switching & state-dependent switching João P. Hespanha University of California at Santa Barbara Hybrid Control and Switched.
10/11/2001Random walks and spectral segmentation1 CSE 291 Fall 2001 Marina Meila and Jianbo Shi: Learning Segmentation by Random Walks/A Random Walks View.
Entropy Rates of a Stochastic Process
N.E. Leonard – U. Pisa – April 2007 Slide 1 Cooperative Control and Mobile Sensor Networks Cooperative Control, Part II Naomi Ehrich Leonard Mechanical.
Centre for Autonomous Systems Petter ÖgrenCAS talk1 A Control Lyapunov Function Approach to Multi Agent Coordination P. Ögren, M. Egerstedt * and X. Hu.
Zhixin Liu Complex Systems Research Center,
ONR MURI: NexGeNetSci From Consensus to Social Learning in Complex Networks Ali Jadbabaie Skirkanich Associate Professor of innovation Electrical & Systems.
Network Coding Project presentation Communication Theory 16:332:545 Amith Vikram Atin Kumar Jasvinder Singh Vinoo Ganesan.
Some Fundamentals of Stability Theory
EE 685 presentation Optimization Flow Control, I: Basic Algorithm and Convergence By Steven Low and David Lapsley Asynchronous Distributed Algorithm Proof.
Hypercubes and Neural Networks bill wolfe 10/23/2005.
Tutorial 10 Iterative Methods and Matrix Norms. 2 In an iterative process, the k+1 step is defined via: Iterative processes Eigenvector decomposition.
From Consensus to Social Learning Ali Jadbabaie Department of Electrical and Systems Engineering and GRASP Laboratory Alvaro Sandroni Penn Econ. and Kellogg.
Differential Equations
Domain decomposition in parallel computing Ashok Srinivasan Florida State University COT 5410 – Spring 2004.
Manifold learning: Locally Linear Embedding Jieping Ye Department of Computer Science and Engineering Arizona State University
CIS 540 Principles of Embedded Computation Spring Instructor: Rajeev Alur
2 Introduction: phase transition phenomena Phase transition: qualitative change as a parameter crosses threshold Matter temperature magnetism demagnetism.
A Framework for Distributed Model Predictive Control
TUTORIAL on LOGIC-BASED CONTROL Part I: SWITCHED CONTROL SYSTEMS Daniel Liberzon Coordinated Science Laboratory and Dept. of Electrical & Computer Eng.,
1 Burning a graph as a model of social contagion Anthony Bonato Ryerson University Institute of Software Chinese Academy of Sciences.
Average Consensus Distributed Algorithms for Multi-Agent Networks Instructor: K. Sinan YILDIRIM.
Consensus in Multi-agent Systems with Second-order Dynamics Wenwu Yu Department of Mathematics Southeast University, Nanjing, China Supervisor: Guanrong.
A Practical Approach to Robotic Swarms IASTED Conference on Control and Applications May 2008 Howard M. Schwartz and Sidney N. Givigi Jr.
Synchronization in Networks of Coupled Harmonic Oscillators with Stochastic Perturbation and Time Delays 尚轶伦 上海交通大学 数学系.
Stable, Circulation- Preserving, Simplicial Fluids Sharif Elcott, Yiying Tong, Eva Kanso, Peter Schröder, and Mathieu Desbrun.
Synchronization in complex network topologies
EE 685 presentation Optimization Flow Control, I: Basic Algorithm and Convergence By Steven Low and David Lapsley.
Spectral Sequencing Based on Graph Distance Rong Liu, Hao Zhang, Oliver van Kaick {lrong, haoz, cs.sfu.ca {lrong, haoz, cs.sfu.ca.
Kinematic Redundancy A manipulator may have more DOFs than are necessary to control a desired variable What do you do w/ the extra DOFs? However, even.
Lecture #14 Computational methods to construct multiple Lyapunov functions & Applications João P. Hespanha University of California at Santa Barbara Hybrid.
COMMUTATION RELATIONS and STABILITY of SWITCHED SYSTEMS Daniel Liberzon Coordinated Science Laboratory and Dept. of Electrical & Computer Eng., Univ. of.
Review of Matrix Operations Vector: a sequence of elements (the order is important) e.g., x = (2, 1) denotes a vector length = sqrt(2*2+1*1) orientation.
Miniconference on the Mathematics of Computation
Asymptotic behaviour of blinking (stochastically switched) dynamical systems Vladimir Belykh Mathematics Department Volga State Academy Nizhny Novgorod.
Decentralized Control of Mobile Sensor Networks Kevin M. Lynch Laboratory for Intelligent Mechanical Systems Mechanical Engineering Department Northwestern.
Mathematical Tools of Quantum Mechanics
Signal & Weight Vector Spaces
Chapter 2-OPTIMIZATION G.Anuradha. Contents Derivative-based Optimization –Descent Methods –The Method of Steepest Descent –Classical Newton’s Method.
1 Distributed Motion Coordination: From Swarming to Synchronization Ali Jadbabaie Department of Electrical and Systems Engineering and GRASP Laboratory.
STATIC ANALYSIS OF UNCERTAIN STRUCTURES USING INTERVAL EIGENVALUE DECOMPOSITION Mehdi Modares Tufts University Robert L. Mullen Case Western Reserve University.
Napa Valley August 3, M. Cao A. S. Morse B. D. O. Anderson Yale University Yale University Australian National University Vicsek’s System with Integer.
Lecture #7 Stability and convergence of ODEs João P. Hespanha University of California at Santa Barbara Hybrid Control and Switched Systems NO CLASSES.
Giansalvo EXIN Cirrincione unit #4 Single-layer networks They directly compute linear discriminant functions using the TS without need of determining.
A Tutorial on Spectral Clustering Ulrike von Luxburg Max Planck Institute for Biological Cybernetics Statistics and Computing, Dec. 2007, Vol. 17, No.
UCLA March 2, 2006 IPAM Workshop on Swarming by Nature and by Design thanks to the organizers: A. Bertozzi D. Grunbaum P. S. Krishnaprasad I. Schwartz.
Ch 9.6: Liapunov’s Second Method In Section 9.3 we showed how the stability of a critical point of an almost linear system can usually be determined from.
Review of Matrix Operations
Consensus in Random Networks
Degree and Eigenvector Centrality
Autonomous Cyber-Physical Systems: Dynamical Systems
Singular Value Decomposition SVD
Stability Analysis of Linear Systems
On Topological Entropy and Stability of Switched Linear Systems
Presentation transcript:

1 Distributed Motion Coordination: From Swarming to Synchronization Ali Jadbabaie Department of Electrical and Systems Engineering and GRASP Laboratory University of Pennsylvania Hamilton Institute Seminar 06/24/2005

SWARMS 2 Distributed Coordination in nature Flocks, swarms and schools exhibit coordinated group behavior although each animal acts completely autonomously How do these behaviors emerge? How are they sustained? How do individual decisions lead to collective group behavior?

SWARMS 3 Complexity, Statistical Physics, emergence of collective behavior

SWARMS 4 r agent i neighbors of agent i Vicsek’s kinematic model How can a group of moving agents collectively decide on direction, based on nearest neighbor interaction? How does global behavior emerge from local interactions?

SWARMS 5 = heading = speed MAIN QUESTION : MAIN QUESTION : Under what conditions do all headings converge to the same value and agents reach a consensus on where to go? Distributed consensus algorithm

SWARMS 6 Multi-agent Representations: Proximity Graphs We use graphs to model neighboring relations V: A set of vertices indexed by the set of mobile agents. E: A set of edges the represent the neighboring relations. W: A set of weights over the set of edges. Agent i’s neighborhood The neighboring relation is represented by a fixed graph G, or a collection of graphs G 1, G 2,…G m

SWARMS 7 Vicsek’s model switching signal, adjacency matrix Valence matrix finite set of indices corresponding to all graphs over n vertices.

SWARMS 8 Conditions for reaching consensus Theorem (Jadbabaie et al. 2003): If there is a sequence of bounded, non-overlapping time intervals T k, such that over any interval of length T k, the network of agents is “jointly connected ”, then all agents move in a formation. This happens to be both necessary and sufficient for exponential coordination, boundedness of intervals not required for asymptotic coordination. (Moreau ’04, Ren & Beard ‘05)

SWARMS 9 Extensions Asynchronous update (Tsitsiklis et al. ‘84,Cao and Morse ‘04) Switching, directed graphs (Moreau ’04, Ren & Beard’04, Zhu, Francis ’04) Gossip in networks ( Boyd et a. ’04) Balanced, directed graphs, no switching. Olfati & Murray 04 Consensus +quantization (Savkin ’04). Consensus on random graphs (Hatano and Mesbahi ’04) No quadratic Lyapunov function exists, but max j  i – min j  j is a valid Lyapunov function, if connectivity holds. (products of length n-1 of F i s are pseudo-contractive with respect to a subspace norm. ) Analysis extends to Choose x = tan(  )

SWARMS 10 Motion Coordination with Dynamic Models Double integrator model Neighbors of i:

SWARMS 11 The Laplacian of the graph The graph Laplacian encodes structural properties of the graph Some properties of the Laplacian: It is positive semi-definite The multiplicity of the zero eigenvalue is the number of connected components One corresponding eigenvector is the vector of ones, 1. The second smallest eigenvalue quantifies connectivity.

SWARMS 12 Dynamic Topology Local sensing/communication Graph changes with time Control is discontinuous Non smooth Lyapunov theory Topology dictates analysis Fixed Topology Fixed (logical) network Graph is constant Control is smooth Classic Lyapunov theory

SWARMS 13 For both fixed and dynamic topology: If the neighboring graph stays connected, all agent velocity vectors become asymptotically the same, collisions between interconnected agents are avoided and the system approaches a configuration that minimizes all agent potentials. Conditions for coordination We could shape potentials for any desired configuration, and also update it as the objective changes.

SWARMS 14 Synchronization of coupled oscillators Consider a group of N oscillators coupled nonlinearly as It is the simplest model of coupled oscillators, simple enough for analysis, but complicated enough to have interesting non-trivial behavior. The degree of synchronization, is measured with the magnitude of the average phasor: r(t) close to 1 means synchronization, and r(t) close to zero means asynchrony.

SWARMS 15 History of the coupled oscillators History of the coupled oscillators Study of Mutual synchronization of biological oscillators goes back to Weiner in 1950s. Examples: pacemaker cells in the heart and nervous system, collective synchronization of pancreatic beta cells, synchronously flashing fire flies. Synchronization of oscillators has also been studied in the context of injection locking in RF circuits Good abstraction for studying networks of loads and generators in the power grid. All-to-all case with infinite oscillators characterized, finite case and arbitrary topologies open … See the book by Steven Strogatz

SWARMS 16 Laplacian & Incidence Matrix Weighted Laplacian Some properties of the Laplacian: The e dimensional vector space of edges can be decomposed to an n-1 dimensional cut space (span of columns of B T ) and m-n+1 dimensional cycle space (Kernel of B). W is diagonal B is the (n x e) incidence matrix of graph G

SWARMS 17 B is the incidence matrix of the graph representing the interconnection of oscillators Simple case: all oscillators are identical Theorem: Consider the unperturbed Kuramoto Model defined over an arbitrary connected graph with incidence matrix B. For any given  0 and any positive value of the coupling, the vector is a locally asymptotically stable equilibrium solution. Furthermore, the rate of approach to equilibrium is no worse than Kuramoto model with incidence matrices

SWARMS 18 For |  i|<  /2 for a connected graph, all trajectories will converge to S Therefore, all velocity vectors will synchronize. A Special Case Fixed points: But, this stability result is not global. In the case of the ring topology is not the only equilibrium. This is due to the fact that B and B T have the same null space! is also stable: Thus

SWARMS 19 When  =0, is an asymptotically stable fixed point. The order parameter can be written as Where e is the number of edges in the graph is a Lyapunov function, measuring velocity misalignment. Using LaSalle, all trajectories converge to invariant sets. Can extend to the case of changing topology, if the graph is “jointly connected”. The speed of synchronization depends on the algebraic connectivity of the graph (2 nd smallest eigenvalue of the Laplacian). Properties of the model

SWARMS 20 Onset of Synchronization When the frequencies are non zero, there is no fixed point for small values of coupling. Theorem: Bounds on the critical value of the coupling can be determined by maximum deviation of frequencies from the mean, and algebraic connectivity of the graph. When  is random, Can develop a mean field model for general topologies by

SWARMS 21 Want to globally minimize 1-r 2 over the whole network Let z= sin(B T  ) Subject to: Dual decomposition and nonlinear network flow Supply at each node Sum of pair-wise potentials Kuramoto model is the Subgradient algorithm for solving the dual Subgradient algorithm Lagrangian Shor 87, Tsitsiklis ’86

SWARMS 22 Kuramoto model with non- homogeneous delays Phase information from neighbors arrive with arbitrary time delay  ij < 1. [A ij ] is the adjacency matrix. In case of degree regular graphs, linearized model with  ij =  was studies by Earl and Strogatz Using Lyapunov-Krasovskii functionals, can analyze the case of arbitrary connected graphs and non- homogeneous delays. We assume K is large enough that the oscillators are synchronized, and linearize the dynamics around the synchronized state,  i (t) = , where  i =  t +  i (t) with G ij = A ij cos(   ij )

SWARMS 23 Theorem: Consider a network of N identical oscillators with linearized dynamics G ij = A ij cos(   ij ) and G ij >0 when i and j are neighbors. Synchronized state is stable independent of delay. Proof sketch: Use the following V(  ) as a Lyapunov/Krasovski functional : Delay Independent Stability Corollary: If [G ij ] is the adjacency matrix of a connected graph, then the continuous time, consensus problem is asymptotically stable with arbitrary time delays

SWARMS 24 Biologically plausible coordination for kinematic robots The input Minimizes the misalignment potential The control law minimizes the potential by following its gradient. But we can’t measure the headings of neighbors W/O communication

SWARMS 25 Biologically plausible sensing Knowing relative heading would mean having binocular vision or solving structure from motion. This would require multiple visible features on each agent. Measured Quantities the projection of an agent (bearing) β ij the speed of the projection (optical flow) The time-to-collision or Expansion rate (rate of approaching or receding of an object), measured as the relative rate of change of the projection area Pigeons and flies are capable of all 3 measurements!. Wang & Frost, Nature, 1992, Fry and Dickinson, Science 2003  ij l ij ii jj XwXw YwYw j i

SWARMS 26 Theorem: with the distributed controller joint connectivity in time flocking A distributed control law Proof based on construction of Lyapunov function whose derivative is the quadratic form of a state dependent Laplacian, and the following lemma

SWARMS 27 Simulations for 2d and 3d kinematic models

SWARMS 28 Coordination in 3D Consider a group of N agents with different velocity vectors and constant, unit speed (extension to dynamic case possible). Agent i’s neighborhood θ is the heading and φ is the attitude.

SWARMS 29 Geodesic Control Law XwXw YwYw ZwZw vivi vjvj  ij Bullo, Murray and Sarti, “Control on the Sphere and Reduced Attitude Stabilization”, 1995 TiSTiS X i  X i 

SWARMS 30 Geodesic Versor Geodesic versor Y ij shows the geodesic direction from v i to v j is the component of vj orthogonal to vi. XwXw YwYw ZwZw vjvj TiSTiS XX XX vivi Y ij  ij For any two agents iand j :

SWARMS 31 Theorem [ Moshtagh, Jadbabaie and Daniilidis, CDC’05]: Consider the system of N equations If the proximity graph of the agents is fixed and connected, then the control laws result in flocking. Furthermore the consensus state is locally asymptotically stable. A similar result holds in the case of switching graphs, if the union graph is connected in time. YwYw ZwZw vjvj TiSTiS XX XX vivi Y ij XwXw  ij

SWARMS 32 Lyapunov-based proof Lyapunov function: measure of discrepancy between velocity vectors Using LaSalle’s invariance principle, all trajectories converge to the largest invariant set within the set: velocity vectors will synchronize. In 2-d this is v T L v It Can be shown that For 0<|  i|< , could also use ||  ||^2 as a Lyap. Function. For a sublevel set inside |  i |<  /2, all trajectories converge to a set where  i =  j Use ||  ||^2 as Lyapunov function on this set, then all trajectories will converge to

SWARMS 33 Vision-based control law generalizes to 3D Visual Servoing Approach Equation of Motion We can construct distributed control laws for flocking, based on visual sensing and measurement of bearing, time to collision and optical flow. No communication or relative distance or heading information is needed We can solve for the input in terms of the measurements. Q ij vjvj Agent i

SWARMS 34 Current Research Simulations

SWARMS 35 Research Issues Implementation on ER robots (underway!) Measurement of OF and TtoC is noisy!! (How do flies do it?) How to optimize connectivity in a distributed way? Use 2 (L(x)) =0 as an obstacle 2 (L(x)) is matrix concave!!, the corresponding eigenvector gives a subgradient direction 2 (L+  L) ≥ 2 (L)+Trace(G  L), G=v 2 v 2 *, Lv 2 = 2 v 2 Can find v 2 in a distributed way!! How does the graph evolve as a function of the positions? Potential-based forces can be used for collision avoidance, but how can we avoid local minima in graphs with cycles? Determine which edges have the most impact on 2 (L)