Brigitte d ANDRA-NOVEL MINES ParisTech, PSL-Research University, - - PowerPoint PPT Presentation

1

Brigitte d’ ANDRÉA-NOVEL MINES ParisTech, PSL-Research University, Centre de robotique, 60 Bd St Michel 75006 Paris Sylvain THOREL SAGEM, 100 avenue de Paris, 91344 Massy Cedex

Systems not stabilizable by means of continuous state

feedback laws (Brockett condition)

Similarities concerning Controllability , Stabilizability,

Flatness properties

Two different approaches for trajectory tracking of

nonsingular reference trajectories and for fixed point stabilization

2

3

4

Idem for the slider :

The Unicycle : the TL system is controllable if

5

The slider : the TL system is controllable if

STLC is satisfied for the two systems at each trajectory The non controllability of the tangent linearized system at a fixed point does not necessarily imply that the original nonlinear system does not satisfy the STLC property. In fact, it can be shown that the LARC is satisfied for both systems, as well as the STLC property.

6

Let us consider a neighborhood of the origin It cannot be in the image of the unicycle dynamics at the origin. STLC is satisfied for the two systems at each trajectory respectively BUT The same holds for the slider. Therefore, due to Brockett’s theorem, these two systems cannot be stabilized at fixed equilibrium points by means of continuous state feedback laws.

7

[G. Kern, « Uniform controllability of a class of linear time-varying systems », IEEE TAC, 1982]

Definition found in [18] and [19]:

8

[18] Jean lévine, Analysis and Control of Nonlinear Systems : A Flatness-basedApproach, Springer 2009, pp 143. [19] M. Fliess, J. Levine, P. Martin, P. Rouchon, Flatness and defect of non-linear systems: introductory theory and examples, Int. J. Control, vol. 61, 1327-1361, 1995.

The unicycle robot is flat with flat outputs Y1 = x and Y2 = y :

9

The system is dynamic feedback linearizable, the decoupling matrix being singular when the longitudinal velocity is zero. The extended state χ1 is the longitudinal velocity v1 which has to be delayed.

The unicycle robot is flat with flat outputs Y1 = x and Y2 = y :

10

The system is dynamic feedback linearizable, the decoupling matrix being singular when the longitudinal velocity is zero.

11

Tracking non singular reference trajectories for the unicycle robot using dynamic feedback linearization

The slider is flat with flat outputs Y1 = x and Y2 = y :

12

The system is dynamic feedback linearizable, the decoupling matrix being singular when the longitudinal acceleration is zero, the extended state χ1 is the longitudinal acceleration which has to be delayed.

13

Autonomous Indoor exploration for wheeled

mobile robots

SLAM Trajectory generation and tracking control laws 3D reconstruction Object recognition…

Adapt these technologies to a hybrid terrestrial

and aerial quadrotor prototype

14

Slider dynamic behavior similar to hovercraft Tilting thrust Hovercraft model proposed in [8]

Simplified model derived from an underactuated surface

vessel modeling

Kinematic and dynamic equations

15

[8] I. Fantoni, R. Lozano, F. Mazenc, K. Y. Pettersen, Stabilization of a nonlinear underactuated hovercraft. Conference on Decision and Control (CDC), 1999

Hovercrafts belong to a more general class of

marine vehicles which are known to be not asymptotically stabilizable at equilibrium points by continuous state feedback laws ([9]). by continuous state feedback laws ([9]). See also [7], [8], [10]

16 [9] K. Y. Pettersen and O. Egeland, Exponential Stabilization of an Underactuated Surface Vessel, CDC, 1996. [10] R. W. Brockett, Asymptotic stability and feedback stabilization, Diff. Geometric Control Theory, Ed . Brockett, Millmann, Sussmann, Birkhauser, Boston, pp 181-191, 1983. [7] E. Sontag, H. Sussmann, Remarks on continous feedback, CDC, Albuquerque, 1980 . [8] C. Samson, Velocity and torque feedback control of a non holonomic cart, Advanced robot control, Springer, 1991.

Trajectory tracking

Non linear control laws based on a Lyapunov analysis

▪ Surface vessel/ Position tracking/ constraint : longitudinal speed ≠ 0 ([11]) ▪ Surface vessel/ Posture tracking/ Exciting reference trajectory ▪ Surface vessel/ Posture tracking/ Exciting reference trajectory

([12])

Flatness [13]

▪ Hovercraft stabilization/ Constraint on the reference trajectory

Sliding mode [14]

▪ Hovercraft stabilization

17 [11] J. M. Godhavn, Nonlinear Tracking of Underactuated Surface Vessel. Decision and Control conference, 1996. [12] Pettersen and Nijmeijer. Tracking control of an underactuated surface vessel. CDC, 1998 [13] H. Sira-Ramirez and C. A. Ibanez, On the control of the Hovercraft System. Decision and control conference, 2000. [14] H. Sira-Ramirez, Dynamic second order sliding mode control of the hovercraft vessel. Control Systems technology, IEEE Transactions on, 2002.

Point stabilization

Time-varying control laws

▪ Smooth feedback but slow convergence [11] ▪ Homogeneous continuous feedback, fast convergence, low robustness [9], [16] (Surface vessel stabilization)

Discontinuous control laws

Lyapunov based analysis ([8]) (Hovercraft Stabilization)

Practical stabilization

▪ Hovercraft Stabilization/ Position tracking/ C3 reference trajectory

([17])

▪ Transverse functions [18]

18

[11] J.M. Coron, Brigitte d’Andréa-Novel, Smooth Stabilizing time-varying control laws for a class of nonlinear systems. Application to mobile robots. Proceedings of Nolcos Conference, Bordeaux, June 1992, pp. 649-654. [16] K. Y. Pettersen and T. I. Fossen, Underactuated Ship Stabilization using Integral Control : Experimental Results with Cybership I. IFAC NOLCOS, 1998. [17] A. P. Aguiar, L. Cremean and J. P. Hespanha, Position Tracking for a Nonlinear Underactuated Hovercraft : Controller Design and Experimental Results, Decision and Control conference, 2003. [18] P. Morin and C. Samson, Practical stabilization of driftless systems on Lie groups, INRIA, Tech. Rep. 4294, 2001.

Model Commands

19

20

Outputs x, y & are flat outputs:

21

Control law: Stability and convergence are assured for the closed

loop system with:

22

Experimental platform Motion Capture system Drone Identification process Identification process Aerodynamic forces and moments Friction effects (static and kinetic) Grey box identification Trajectory tracking results

23

• software MOTIVE
• Remote control

Streaming VRPN Communication ZigBee

Infrared cameras s250e Optitrack

24

• Embedded computer
• Microcontroller MikroKopter

Serial port Drone

25

Trajectory tracking realized by the flatness control law

Reference trajectory constraints: Derivatives until the second order for the state and third order

for the reference for the reference

Experimental conditions

Circular trajectory tracking with radius 1.1m Initial position :50cm from the reference trajectory Ground : parquet

26

27

28

29

[11] J.M. Coron, Brigitte d’Andréa-Novel, Smooth Stabilizing time-varying control laws for a class

• f nonlinear systems. Application to mobile robots. Nolcos Conference, Bordeaux, June 1992

The theorem applies for the unicycle robot with

30

31

32

[J.B. Pomet, B. Thuilot, G. Bastin, G; Campion, « A hybrid strategy for the feedback stabilization of nonholonomic mechanical systems », Proc. IEEE CDC, 1992]

The theorem applies for the slider with

33

34

We can consider the error tracking system at an equilibrium point with

35

36

[G. Kern, « Uniform controllability of a class of linear time-varying systems », IEEE TAC, 1982]

37

A change of coordinates

38

A Lyapunov function candidate : Its time-derivative :

39

By duality, OK if : [H.K. Khalil, « Nonlinear systems », Prentice Hall, 1995]

For the complete system, the previous law u1 and the following yaw rate viewed as virtual Input (completed by backstepping technique) ensure the fixed point stabilization:

40

Proof : Lyapunov function candidate : Time-derivative of the Lyapunov function :

LaSalle arguments :

41

On this invariant subset we can prove that e2 remains constant and that the yaw rate satisfies the assumptions of proposition 3.1 if ω is sufficiently small:

42

43

• A hybrid control strategy for unicycle and slider type robots.

44

• How to extend these results in the context of finite-time stabilization ?

We collaborate in a recent research project on FTS funded by ANR with colleagues from INRIA Lille (W. Perruquetti, A. Poliakov, D. Efimov, J.P. Richard …), UPMC (J.M. Coron, E. Trélat …) and L. Rosier

45

• B. d’A-N, J-M. Coron, W. Perruquetti, « FTS of nonholonomic or underactuated

mechanical systems: the examples of the unicycle and the slider », in preparation [37] P. Morin, C. Samson, «Time-varying exponential stabilization of a rigid spacecraft with two control torques », IEEE TAC, 1997

The case of the double integrator

46

• E. Bernuau, W. Perruquetti, D. Efimov, E. Moulay, « Robust FT ouput feedback stabilization
• f the double intergrator »,
• Y. Hong, « FT stabilization and stabilizability of a class of controllable systems », SCL, 2002
• L. Rosier, « Etude de qqs problèmes de stabilisation », PhD thesis, 1991

The case of the slider

• s. t.

2) to stabilize the double integrator

47

1) We conclude by time-periodicity.

Stabilization of dynamical systems modeled by hyperbolic PDEs coupled with Boundary Conditions modeled by non linear ODEs. Some examples :

48

Overhead crane with flexible cable :

• B. d’A-N, J.-M. C. : “Exponential stabilization of an overhead

crane with flexible cable via a back-stepping approach”, Automatica, (26), 2000.

• B. d’A-N, J.-M. C : “Stabilization of an overhead crane with

a variable length flexible cable”, Computational and Applied Mathematics, Vol. 21 No. 1, 2002.

Irrigation canals :

49

J.-M. C., B. d’A-N, G. B. : “Les équations de Saint-Venant pour le contrôle des canaux et des voies navigables. Penser globalement, observer et agir localement”, article pour la Recherche, Wxyz, No. 417, Mars 2008, pp. 82-83. J.-M. C., B. d’A-N, G. B. : “A Strict Lyapunov Function for Boundary Control of Hyperbolic Systems of Conservation Laws”, in IEEE Transaction on Automatic Control,

• Vol. 52, No. 1, January 2007.

Wind Instruments:

• B. d’ANDRÉA-NOVEL, B. FABRE, J.-M. CORON : “An

acoustic model for automatic control of a slide flute”, in Acta Acustica, Journal of the European Acoustics Association (EEA), Volume 96, 2010, pp. 713-721.

50

• B. d’ANDRÉA-NOVEL, J.-M. CORON, T. HÉLIE:

“Asymptotic observers for a simplified brass instrument model”, in Acta Acustica, Journal

• f the European Acoustics Association

(EEA), Volume 96, 2010, pp. 733-742.

51

• We have considered the simpler problem of controlling a slide flute : a cylindrical recorder

without finger holes but with a piston mechanism to modify the length.

• From a mathematical point of view : a system of conservation laws leading to hyperbolic

PDEs coupled with a nonlinear ODE with delay.

52

PDEs coupled with a nonlinear ODE with delay.

flue channel labium labium Bernoulli equations Bernoulli equations W h

Model of the jet channel and the mouth

The ideal boundary condition at x=0 si too simple !

53

mouth mouth pipe pipe

• Interaction jet/labium, jet/pipe, jet drive mechanism [Cremer, Ising 1968; Coltman 1976].

Interaction jet/labium, jet/pipe, jet drive mechanism [Cremer, Ising 1968; Coltman 1976].

• Vortex shedding at the labium [Verge, Fabre et al. 1994; Verge, Hirschberg, Caussé 1996].

Vortex shedding at the labium [Verge, Fabre et al. 1994; Verge, Hirschberg, Caussé 1996].

• is the jet velocity and the pressure jump responsible of the sound production,

is the jet velocity and the pressure jump responsible of the sound production, mainly composed of the pressure jump due to the jet drive mechanism, and due mainly composed of the pressure jump due to the jet drive mechanism, and due to the vortex shedding. to the vortex shedding.

h

The pressure in the mouth is related to the output flow by the radiation impedance

• The pressure term due to the dipolar jet-drive :

Jet position

Uj c~0.3Uj

54

• The dissipative pressure term due to vortex shedding :

Mouth cross section at the flue exit Jet position Spatial amplification of the jet Vena-contracta factor of the flow

W

Labium delay

pf = 300 Pa pf = 55 Pa

55

pf = 5 Pa f1s = 310 Hz (close to 290.8 Hz) f2s = 925 Hz (close to 903.7 Hz)

56

• The note being chosen, we solve the linearized B.C. w.r.t. L and U0. For

example, to obtain Fe ideal (c/4L) = 324 Hz, this leads to :

• This is our feedforward control algorithm.

57

• Feedback :

Slide

feedforward

58

Slide + Artificial mouth

+ PID modal analysis

feedback

Target pitch

59

60

61

HAPPY BIRTHDAY

62

JEAN-MICHEL !!!