adaptive consensus of discrete-time heterogeneous multi ... · multi-agent dynamical systems...

Adaptive Consensus of Discrete-time Heterogeneous

Multi-agent Systems

Takashi Okajima (The University of Tokyo)Koji Tsumura (The University of Tokyo)Tomohisa Hayakawa (Tokyo Institute of Technology)Hideaki Ishii (Tokyo Institute of Technology)

○

Multi-Agent Dynamical Systems (MADSs)

2011/9/16SICE Annual Conference 20112

Active research field [Fax & Murray 2004] [Olfati-Saber et al. 2007]

Application for engineering Cooperation control of robots Vehicle formation

Unmanned Aerial Vehicles (UAVs) Automated Highway Systems (AHSs)

Control

J.A. Fax and R.M. Murray, “Information flow and cooperative control of vehicle formations”, 2004.

R. Olfati- Saber et al. “Consensus and cooperation in networked multi-agent systems”, 2007.

Multi-Agent Dynamical Systems (MADSs)

2011/9/16SICE Annual Conference 20113

Agent = Dynamical system (with controller) communicates with neighbor agents. autonomously acts on local information.

The system as a whole achieves some objective.

Local information

Communication

Global objectiveControl

Consensus problem of MADSs

2011/9/16SICE Annual Conference 20114

Consensus problem To reach an agreement on a certain quantity

In this presentation,we consider state consensus problem.

: state vector of ’th agent (at time step )

Research objective

2011/9/16SICE Annual Conference 20115

Homogeneous MADSs have been mainly researched.

Large number of the agentsThe dynamics of them are

heterogeneous & uncertain.

?

?

?

?

state consensus control of uncertain heterogeneous MADSs

[Wieland et al. 2011] :heterogeneous

[Fax & Murray 2004] [Olfati-Saber et al. 2007]

P. Wieland et al. “An internal model principle is necessary and sufficient for linear output synchronization,” 2011.

H. Kim el al. “Output consensus of heterogeneous uncertain linear multi-agent systems”, 2011.

[Kim et al. 2011] :heterogeneous & uncertain

Contents

2011/9/16SICE Annual Conference 20116

Introduction Adaptive control approach Framework Numerical example Conclusion

Adaptive control approach (1)

2011/9/16SICE Annual Conference 20117

A standard adaptive control can be regarded as a consensus problem of 2 agents.

Extension to more general communication topology.

Adjustmentmechanism

Feedback Controller Plant

Model

Agent 1

Agent 2

2

1

uncertain dynamics

known (desired) response

A standard adaptive control can be regarded as a consensus problem of 2 agents.

Extension to more general communication topology.

Adaptive control approach (2)

2011/9/16SICE Annual Conference 20118

4

12

1

3

62

5 7

Previous study (Adaptive control approach) Kaizuka & Tsumura, 2010. continuous-time systems

Corresponding result for discrete-time systems is not straightforward.

Adaptive control approach (3)

2011/9/16SICE Annual Conference 20119

4

12

1

3

62

5 7

Contents

2011/9/16SICE Annual Conference 201110

Introduction Adaptive control approach Framework Problem formulation Distributed adaptive controller Main result

Numerical example Conclusion

Communication topology

2011/9/16SICE Annual Conference 201111

Edge⇔ Agent receives information of Agent .

A special agent is called leader. (⇔known model )

The other agent are called followers.(⇔uncertain plants)

4

l=1 3

2

N

There are no edges to the leader. Edges between followers are bidirectional. There are directed spanning trees.

Assumption about Communication topology

Heterogeneous dynamics of agents

2011/9/16SICE Annual Conference 201112

Leader :Follower :

Agent Dynamics state: input:

:known, asymptotically stable

:unknown, heterogeneous, (possibly unstable):known, full column rank

:reference input, bounded:control input to Agent

Assumption (Matching condition)

:unknown, heterogeneous‘the true value of gain’

++

=

State consensus problem

2011/9/16SICE Annual Conference 201113

Design control inputs . Achieve state consensus.

Use only the states of the neighbors and its own.

Leader :Follower :

Agent Dynamics state: input:

3

6

5 7

4

1

2

Adaptive Controller of Follower

2011/9/16SICE Annual Conference 201114

++

:estimation (at time ) of .Feedback control law:

Gain update law:

Controller of Follower

:Neighbors of agent

3

6

5 7

4

1

2

( )

Gain update law

2011/9/16SICE Annual Conference 201115

Continuous-Time

[Hayakawa et al. 2004]

Discrete-Time

2l

[Kaizuka & Tsumura 2010]

summation

standard

multi-agent summation

Main result

2011/9/16SICE Annual Conference 201116

Theorem

is non-increasing.

The system of {state errors & gain errors} is globally stable.

(state consensus)

⇒ The gain update law is a globally stable estimator.

The controlled MADSs satisfies following.

(Details are given in the proceedings.)

( : Gain error at agent )

Contents

2011/9/16SICE Annual Conference 201117

Introduction Adaptive control approach Framework Numerical example Conclusion

Numerical example

2011/9/16SICE Annual Conference 201118

l=12 3

45

Leader :Follower :

Agent Dynamics

Communication topology

Poles(stable)

(unstable)

Consensus achieved

2 3

45

Conclusions

2011/9/16SICE Annual Conference 201119

Conclusions We proposed a state consensus control framework

for discrete-time uncertain heterogeneous MADSs. We showed the adaptive control based framework

is also valid for discrete-time systems.

Future works Output feedback case The case that none of the agent dynamics is known.

2011/9/16SICE Annual Conference 201120

Thank you for your attention!

2011/9/16SICE Annual Conference 201121

Continuous-time vs. discrete-time

2011/9/16SICE Annual Conference 201122

Continuous-time case

Discrete-time case

Lyapunov function:

2

l=1

Update law:Linear!

⇒

(Positive definite) cross term!Lyapunov function:

Update law:

: large enough.⇒