• Aucun résultat trouvé

Consensus protocol

N/A
N/A
Protected

Academic year: 2022

Partager "Consensus protocol "

Copied!
28
0
0

Texte intégral

(1)

Institute for Design and Control of Mechatronical Systems

Introduction aux Systèmes Collaboratifs Multi-Agents

UPJV, Département EEA

Fabio MORBIDI

Laboratoire MIS

Équipe Perception Robotique E-mail: [email protected]

M1 EEAII - Découverte de la Recherche (ViRob)

(2)

Consensus protocol

Directed weighted networks

(3)

  Consider the weighted digraph in the figure, which corresponds to the first-order dynamics :

1

2

Reaching consensus: directed weighted networks

(4)

We can compactly rewrite the previous system as:

1

2

3 4

where

Recalling the definition of in-degree Laplacian for digraphs, we can rewrite the dynamics of the networked system as:

where is the underlying directed interconnection between the vertices

Reaching consensus: directed weighted networks

(5)

Another example:

•  Three robots coordinate their speeds

according to the chain of command in the figure

1

3 2

1 1/2

1/2 1

Reaching consensus: directed weighted networks

•  The dynamics of the resulting system can be written as:

(6)

The previous equations can be rewritten as:

where

1

3 2

1 1/2

1/2 1

The matrix in the system above corresponds to the negated in-degree Laplacian of the network, thus:

where is the weighted digraph of the network

Reaching consensus: directed weighted networks

(7)

Reaching consensus: directed weighted networks

Are there necessary and sufficient conditions on the graph that lead to the convergence of system to the consensus set ?

As in the case of undirected networks, the rank of the Laplacian matrix and how this relates to the structure of the graph, plays a critical role

The following notion parallels that of spanning tree for undirected graphs

A digraph is a rooted out-branching if:

a) 

It does not contain a directed cycle

b) 

It has a vertex (root) such that for every other vertex there is a directed path from to

Definition (Rooted out-branching or directed rooted tree)

(8)

Reaching consensus: directed weighted networks

Example of rooted out-branching

root

(9)

Reaching consensus: directed weighted networks

A digraph on vertices contains a rooted out-branching as a subgraph if and only if

Proposition

Theorem (Main result)

For a digraph containing a rooted out-branching, the state trajectory generated by initialized with , satisfies:

where and , are respectively, the right and left eigenvectors associated with the zero eigenvalue of , normalized such that

As a result, one has for all initial conditions if and only if

contains a rooted out-branching

(10)

Reaching consensus: directed weighted networks

Example:

root

Blue: a rooted out-branching of (not unique)

(11)

Reaching consensus: remark I

Let be the left eigenvector of the digraph in-degree Laplacian associated with its zero eigenvalue. Then the quantity:

remains invariant under the consensus dynamics,

Proposition (Constant of motion)

Remark:

The vector is a left eigenvector of the squre matrix with associated

eigenvalue if:

(12)

Reaching consensus: remark II, balanced digraphs

A digraph is called balanced if, for every vertex, the in-degree and out-degree are equal

Definition (Balanced digraph)

Balanced digraph (unitary weights) Unbalanced digraph (unitary weights)

in-degree = out-degree in-degree ≠ out-degree

(13)

Reaching consensus: remark II, balanced digraphs

Balanced digraph Unbalanced digraph

in-degree = out-degree in-degree ≠ out-degree

2

3 5

2

5

5

1

2 4

3

1/2

2

(14)

Reaching consensus: remark II, balanced digraphs

When the digraph is balanced, in addition to having , one has

The consensus protocol over a digraph reaches the average consensus for every initial condition if and only if the digraph is weakly connected and balanced

Theorem

Thus, if the digraph contains a rooted out-branching and is balanced, the common value reached by the consensus protocol is the average value of the initial states, i.e. the average consensus, since:

i.e.

Launch the Matlab file:

“Rendezvous_directed.m”

(15)

Consensus protocol

Some extensions for undirected networks

(16)

1.  Consensus protocol with uniform communication time delays

“Consensus problems in networks of agents with switching topology and time-delays”, R. Olfati-Saber, R.M. Murray, IEEE Trans. Automat. Contr., vol. 49, n. 9, pp. 1520-1533, 2004

Consider the uniformly delayed consensus dynamics over a connected, weighted undirected graph, specified by:

for some This delayed protocol achieves average consensus if and only if

where is the largest eigenvalue of the weighted Laplacian

• 

Trade-off between faster convergence rate and tolerance to uniform delays on the information exchange links

Extensions of the consensus protocol

τ < π

2 λ

n

(L(G ))

λ

n

(L(G ))

(17)

2.  Consensus protocol for double-integrator agents

Consider the following second-order dynamics for agent :

where and are, respectively, the position and velocity of agent w.r.t. an inertial frame, and is the control input (acceleration), with

Inspired by the consensus protocol, we can define the control for agent as:

where is a positive gain

Extensions of the consensus protocol

or equivalently

(18)

3.  Discrete-time consensus protocol

An iterative form of the consensus protocol can be stated as follows in discrete time ( ):

where is the step size, and is the maximum degree of the graph

The collective dynamics of the network can be written in compact form as

where and

Extensions of the consensus protocol

(19)

•  is referred to as the Perron matrix of the graph with parameter

•  is a stochastic matrix, i.e. the row-sum is equal to 1:

•  The conditions for achieving consensus in discrete-time are the

same as in continuous-time

•  The convergence speed to the consensus set is dictated by:

the second largest eigenvalue of

For more details on discrete-time consensus protocols and their connection to the theory of Markov chains, see “Consensus and Cooperation in Networked Multi-Agent Systems”, R. Olfati-Saber, J. A. Fax, R. M. Murray, in Proc. IEEE, vol. 95, n. 1, pp. 215-233, 2007

Extensions of the consensus protocol

3.  Discrete-time consensus protocol

(20)

•  Antagonistic interactions (e.g. friends vs. adversaries in a social network), can be modeled as negative

weights on the communication graph

•  On signed networks all agents can converge to a consensus value which is the same for all agents except for the sign (“bipartite consensus”)

•  Bipartite consensus can be achieved if the graph is structurally balanced (this means that all cycles of the graph are positive, i.e. they contain an even

number of negative edge weights)

“Consensus Problems on Networks With Antagonistic Interactions”, C. Altafini, in IEEE Trans. Automat. Contr., vol. 58, n. 4, pp. 935-946, 2013

"Predictable Dynamics of Opinion Forming for Networks with Antagonistic Interactions", C. Altafini, G. Lini, in IEEE Trans. Automat. Contr., vol. 60, n. 2, pp. 342-357, 2015

Extensions of the consensus protocol

4.  Consensus protocol with antagonistic interactions

Bipartite consensus

time [s]

Group 1

Group 2

(21)

Extensions of the consensus protocol

Example (signed undirected graphs, ):

-2 -4

1

-2 4

1

2 4

1

Structurally balanced Structurally unbalanced Structurally balanced 4.  Consensus protocol with antagonistic interactions

(22)

Sujets de projet

(23)

Types de publications scientifiques

On peut distinguer les publications scientifiques selon leur support de parution:

1.  Les revues à comité de lecture ou ‘peer-reviewed’ (jusqu‘à 20 pages) 2.  Les comptes-rendus (‘Proceedings’) de congrès scientifique à comité de

lecture (typiquement 6 pages)

3.  Les ouvrages collectifs rassemblant des articles de revue ou de recherche autour d'un thème donné, coordonnés par un ou plusieurs éditeurs

Structure standard d‘un article scientifique   Titre (10-15 mots)

  Résumé ou abstract (10-20 lignes)

  Introduction (état de l’art et contributions par rapport à la littérature)   Formulation et résolution du problème

  Validation (simulations numériques et/ou expérimentations)

(24)

Nom de la revue

Titre

Auteurs et affiliations

Résumé

Introduction Date de

publication

Page

Volume

(25)

Maisons d‘édition

Maisons d‘édition les plus importantes pour l‘ingénierie:

• 

IEEE (Institute of Electrical and Electronics Engineers)

• 

Elsevier

• 

Springer

• 

John Wiley & Sons

• 

Taylor & Francis

• 

Sage

• 

ACM (Association for Computing Machinery)

Visitez, par ex. ieeexplore.ieee.org et sciencedirect.com

(26)

Sujets de projet (trinômes)

1.  Contrôle de formations de robots mobiles 2.  « Flocking » d’un groupe de robots

3.  « The Flying Machine Arena »

4.  Controllabilité de réseaux complexes

5.  Transport et construction par drones coopérants 6.  Écran robotique

7.  Surveillance aérienne coopérative et décentralisée

8.  Contrôle de couverture avec un réseau de caméras volantes 9.  Le protocole de consensus et les réseaux sociaux

10.  Le protocole de consensus et l‘algorithme PageRank de Google

(27)

Sujets de projet (trinômes)

11.  Propagation des épidémies dans un réseau complexe

12.  « Robotarium »: une plateforme multi-robots accessible à distance 13.  Graphes de proximité et cartes spatialement distribuées

14.  Le problème de la poursuite cyclique (« n-bug problem ») 15.  Formation d’opinion dans les réseaux sociaux

structurellement balancés

16.  Synchronisation d‘un réseau de générateurs électriques

17.  Sécurité des systèmes cyber-physiques

(28)

À propos de la recherche scientifique ...

« Doing research is a hard process, involving many erroneous assumptions, disappointments, and dead ends. Being positive, imaginative, and keeping good judgment are always helpful. Research is all about generating many, many ideas - and just making sure to keep the good ones. »

Panagiotis Tsiotras (IEEE Control Systems Mag., April 2016)

« A researcher must accept to slog away at a problem for a hour, a day, or all his life.

Rather, he uses up his energy excessively with respect to the results, he asks himself several questions, he gropes in the dark, he moves forward step by step.

This is a hard task; then, at a certain point, the illumination comes. It is often unexpected but it is the result of a huge amount of unsuccessful reflections. »

Laurent Schwartz (1915-2002)

« Genius is an illusion: all great human accomplishments are actually just the result of hard work and high tolerance to failure. »

Simon Ramo (IEEE Spectrum, August 2016)

Références

Documents relatifs

They show that regardless of the implicit temporal nature of the query, web snippets contain a broad range of temporal information that can be used as a

An upper triangular matrix is almost in a row echelon form. If all the elements on its diagonal are non-zero then they are the pivots and the matrix is regular. If one of the

To properly plan an outdoor activity, it’s essential to set a clear educational intention for each outing (e.g. have students explore the diversity of arthropods that live

Game-Based Learning involves learning situations where children play or design games – whether digital, physical, or table-top games – in which they solve problems and

The ANSI IS-IS routing protocol specifies a two level hierarchical routing where Level 1 routing deals with routing within an area, while Level 2 routing deals with

On page 29, the pseudo-code for the OPEN state suggests that if a data packet is received in sequence, the acknowledgement packet should not contain EACKs. This

On receipt of either ESROS-RESULT.request or ESROS-ERROR.request from the ESROS performer user, the ESROS provider initiates the inactivity timer for the ESRO- RESULT-PDU

In the given paper we have proposed some approaches for the solution of the problem of random re-ordering for parties which will participate in the imple- mentation of the next epoch