Sie sind auf Seite 1von 23

Controllability notions in nonlinear systems

Technical Report Computation and Control Laboratory COMPCON 2001/1

Arun D. Mahindrakar & Ravi N. Banavar

Systems and Control Engineering Indian Institute of Technology, Bombay Mumbai


January 23, 2002

Abstract Controllability notions have been entrenched in control theory for the past four decades. While linear controllability has clear denitions/interpretations, notions of nonlinear controllability are not so well established. This report presents a brief survey of the different types of controllability that are relevant to mechanical systems, in particular underactuated mechanical systems.

1 Introduction
Time-dependent events

and

are said to commute if the occurrence of

during a time interval followed by the occurrence of

during an interval leads

to an outcome that does not change when the order of events is reversed

Most events do not commute. For example, an automobile driver knows that the rotations of the steering wheel do not commute with either forward or backward motions of the automobile. Control of many dynamic systems, like the control of an automobile, consists of a time-sequence application of noncommuting events. The recognition of noncommutativity as a fundamental issue of control is a starting point for geometric control [1]. The report is organised as follows. In section 2, we state the different types of controllability using topological concepts. There exist sufcient conditions to test some of the controllability properties. The results which are in the form of Lie algebraic conditions are presented in section 2.1. A few simple examples are illustrated in section 3. In section 4 we state the controllability rank condition for a kinematic system with nonintegrable constraints . In section 5 the controllability properties for some underactuated mechanical manipulators are presented. A few concluding remarks are made in section 6. The general problem of controllability (or reachability) is Given any point in a state space, what are the set of points which can be reached from the initial point in nite time by a suitable choice of the input functions. Consider a nonlinear system of the form

(1)

where

are local coordinates for a smooth manifold , and

with some subset of .

are smooth vector elds on M. The control vector

is locally integrable

Denition 1.1 An afne control system is a triple

We now state certain denitions pertaining to nonlinear controllability. The denitions are taken from [2, 3].

2 Accessibility and controllability


Denition 2.1 The nonlinear system (1) is called controllable if and only if for each and for each , there exists an admissible input with the property that if is the solution to the initial value problem (1), then . Let denote the set of reachable points in from at time of for all
,

using

admissible controls, , and such that the trajectories remain in the neighborhood

. Furthermore, dene

The various types of controllability are dened as follows: Denition 2.2 Let 1.

.
. If this holds for any for each

is locally accessible from if int( then the system is called locally accessible.

2. 3. 4.

is locally strongly accessible from if is locally controllable from if

is small time locally controllable (STLC) from if there exists for each . is globally controllable from if

so that

5.

2.1 Lie algebraic conditions


Lie algebraic conditions in the form of sufciency conditions exist for the types of controllability presented in the previous subsection . We rst dene a few denitions which involve some basic concepts from Lie algebra 1 . Denition 2.3 It is the linear space of vector elds on with the bilinear operation being the Lie Bracket of vector elds. In fact with the Lie Bracket is an innite dimensional Lie algebra. Example 2.4 Consider the vector elds on given by

Then,

It can be shown by induction that the vector eld is contained in . Therefore, is am innite dimensional algebra of vector elds.
Denition 2.5 Consider the system (1). The accessibility algebra subalgebra of that contains . Denition 2.6 Accessibility distribution It is the distribution generated by the accessibility algebra .

is the smallest

span

X is vector eld in

Theorem 2.7 If

then the system is locally accessible. The above criteria is termed accessibility rank condition. Denition 2.8 Let be the accessibility algebra of (1). Dene as the smallest . Dene the subalgebra which contains and satises distribution

span

X is vector eld in .

A quick tour of Lie algebraic ideas is given in the appendix.

and are called the strong accessibility algebra and strong accessibility distribution
respectively. We now state the theorem concerning the strong accessibility. Theorem 2.9 Consider the system (1). Suppose that

(2)

then the system is locally strong accessible from . Condition (2) is called the strong accessibility rank condition at .

2.2 Small time local controllability


Sussmann [3] showed a general sufcient condition for STLC of systems with drift. A particular case of this result known as the Bianchini and Stefani condition has been used by Reyhanoglu et al. [4] to prove the STLC property for underactuated systems. We now present their results which centers around identifying the good and bad brackets and their relationship. Let and let

denote the smallest Lie algebra of vector elds containing

number of times

denote any bracket in

occur in the bracket

. Let

denote the

respectively. Dene

and let an admissible weight vector be


is dened as

such that

. The l-degree of

l-degree(B)

The so-called bad bracket is dened as that bracket which has even for each

odd and

is 1 and

. For example the bracket is a bad bracket since . The above concepts are made clear by considering a two

input system. For admissible weights

the good brackets along with their

l-degree are tabulated as shown in table (1). The corresponding table for some of the bad brackets is as shown in table (2). From the tables it clearly follows that for a bad bracket is odd and the l-degree for the spanning good brackets is For bad brackets with
.

, the l-degree is 4

The Bianchini and Stefani

Vector

Table 1: Good brackets Vector

Table 2: Bad brackets condition for STLC for a strongly accessible system is that the bad brackets must be a linear combination of good brackets of lower l-degree at the equilibrium. Then the bad brackets are said to be l-neutralized. We now state the general theorem by Sussmann for STLC of an afne system. Theorem 2.10 Sussmann [3]: The sufcient condition for the STLC at the equilibrium of an afne system (1) is 1. The system satises the Lie algebra rank condition at the equilibrium. 2. Every bad bracket is a linear combination of good brackets of lower degree at the equilibrium.

3 Examples
To bring out the distinguishing properties of the denitions of controllability dened earlier, we present a few simple examples [5]. Throughout we assume the controls to be piecewise constants functions which are piecewise continuous from the right.

Example 3.1 Consider the system

(3)

with and . Let the origin be the point of interest and further, let us assume over some time interval of interest . The solution of (3) over is given by

The solution corresponding to the boundary values of is given by

for

for

Combining the above two equations, the left and right boundaries for is given by the graph of the function (see gure 1)


In order to picturize the set to (3), Switching time

, we apply an arbitrary switching mechanism


Graph of

Table 3: Switching law to generate the boundary of set

for for 6

Figure 1: The reachable set is the complete portion between the left and right boundary while the dark shaded portion is the set . where

. The solution at time is given by


The graph of the function for typical values of is as shown in table (3). The top two rows correspond to bang-bang controls. In the above, each control value determines a vector eld, and the corresponding trajectory is a solution curve of this eld. As the control switches to a new value so does the vector eld and the trajectory begins to follow the ow associated with the new vector eld. The trajectories generated in this manner result in a continuous curve in with discontinuous derivatives. These discontinuities occur at the switching instants. The resulting trajectories are sometimes called as continuously broken trajectories. 3.0.1 Lie algebraic test Let , it is easy to verify

We note that the bracket is identically equal to zero at . But the higher along with the control vector eld gives a constant dimension order bracket 7

distribution for all . We note that is the strong accessibility distribution spanning a two-dimensional space, thus the system is strongly accessible. Remark 3.2

The system is accessible from

.
.

The system is strongly accessible

The system is not controllable in any of the three senses dened earlier.

Example 3.3


with

(4)

and . Let the origin be the point of interest and . The solution of (4) then is given by

The set is given by the graph of the function (see gure 2)


3.0.2 Lie algebraic test

The drift and control vector eld are given by


It is easy to verify that

Hence the system is locally accessible. The Lie bracket of and turns out to be the zero vector eld. Hence no conclusions can be made regarding strong accessibility based on the strong accessibility distribution . Remark 3.4

The system is accessible from

. 8

Figure 2: The shaded region denotes the set and the thick line denotes the set for some .

The system is not strongly accessible. The system is not controllable in any of the three senses dened earlier.

Example 3.5

(5)

and . Here, the system evolves on a cylinder. Let the with be the point of interest and . The origin solution of (5) is given by

For small times the set is as shown in gure (3). We note that the open semicircle centered at the origin is an open set of . For small times the reachable set does not contain the origin, but for large time the origin is contained in the 9

interior of the reachable set (see gure 4). Hence the system is not STLC .

Open set of

Figure 3: (a): The reachable set for small time is represented by the shaded area , (b): The set of points that can be reached in some time is represented by the horizontal line.

Figure 4: Reachable set for large time

3.0.3

Lie algebra test

The drift and control vector eld are given by


It is easy to verify that

10

Hence the system is locally accessible. The vector elds and commute since the Lie bracket of and turns out to be the zero vector eld. Hence the sufcient conditions for strong accessibility fails. Remark 3.6

The system is accessible . The system is not strongly accessible. The system is not STLC . The system is globally controllable.

This example brings out the importance of the underlying manifold in determining the controllability properties. Example 3.7 Consider the system


(6)

with . Here and in calculations are straightforward:

The following Lie bracket

The brackets span a three dimensional space at the origin which is an equilibrium point. The system is locally accessible. The highest degree of the good bracket in the accessibility distribution is six. In order to satisfy the second condition of theorem (2.10) , we need to express the bad brackets of degree not greater than six as a linear combination of good brackets of lower degree ( ). In fact, the bad brackets vanish at the equilibrium. Hence the system is STLC at the equilibrium. 11

In all the examples considered , the assumption the assumption the design.

was valid since we used open

loop controls. When using feedback control law, becomes a function of the states and for initial conditions in the region of interest will have to be a posteriori guarntedd by

cannot be a priori veried. The satisfaction of this assumption

4 Controllability of Kinematic Systems with nonholonomic constraints


Kinematic systems are systems with velocities as inputs rather than external forces. Constraints on the system could of the following types: one in which the constraint condition can be expressed as an algebraic equation involving the conguration variables and time, and the other in which it is not possible to do so. The former is called holonomic and the latter nonholonomic (nonintegrable) . Kinematic systems with nonholonomic constraints can be expressed in state-space as a drift-free system,

(7)

with the usual assumptions made earlier. For systems that are drift-free, the controllability test is quite simple. Dene by

. Let and dene the controllability distribution as


to be the Lie algebra generated

X is a vector eld in

The controllability test is stated in the following theorem which was originally presented and proved by W.L. Chow . Theorem 4.1 (Chow) The control system (7) is locally controllable if

for all x in the neighborhood of

The following example [6], which models a car, illustrates the above notion of controllability. Consider a car moving on a plane as shown in gure (5). The front and the rear pairs of wheels are modeled as single wheels at the midpoints of the axles. The conguration of the system is given by 12

and is parametrized by

where are the Cartesian coordinates of the rear wheel(s), the angle is the angle made by the

measures the direction in which the car is heading, and

front wheels with the car. The constraints on the system arise by allowing the wheels to roll and spin, but not slip. The constraints for the front and rear wheels are formed by writing the sideways velocity of the wheels:

With the inputs chosen as the driving velocity and the steering velocity , the control system can be written as

Figure 5: Kinematic model of a car

where

13

The control vector eld is called drive and is called the steer. The Lie bracket of and , called as wriggle is given by

A similar computation yields the Lie bracket of and , called as slide and is given by

It is interesting to compute the higher order Lie brackets which together with previously calculated brackets is summarized below.

It is clear that the vector elds

span a

dimensional subspace for all

Hence by Chows theorem, the system is locally controllable.

In fact the system is STLC.

5 Controllability of mechanical systems with second order nonholonomic constraints


Control of underactuated mechanical systems has seen a surge of interest in recent years due to the challenges involved. The challenges stem out from the fact that such systems have complex nonlinear dynamics and loss of full state feedback linearizing property. The case for systems that are not linearly controllable imposes additional constraints such as the nonexistence of a smooth static state feedback law for the asymptotic stabilization [7] of an equilibrium point. We present a few examples of 14

underactuated manipulators, some of which have appeared in the literature [8, 9, 4] and for which controllability properties have been analyzed. 1. Consider a three link manipulator moving in a vertical plane under the inuence of gravity. Let the dimension of the control space be two. The placement of the two actuators does not alter the controllability properties; the system is still linearly controllable. The same holds for manipulators of higher degrees of freedom with a single unactuated joint. 2. Consider a two degrees of freedom manipulator with revolute joints moving in the horizontal plane so that gravity can be ignored. We assume that there are no frictional forces and the potential energy is solely contributed by gravity. An immediate consequence of the absence of gravity is the loss of linear controllability. Moreover, the controllability properties alter with the location of the actuator. For example, the system with the actuator at the second joint is not controllable. In this case, the torque applied is like an internal torque and hence the torque felt by the rst link is equal and opposite to the torque applied to the second link. In the other case, where the actuator is placed at the rst joint, the reaction torque is felt at the base where the rst link is mounted. This system is locally accessible [10], but the sufcient conditions for STLC are not satised. Luca et al [10] have used a nilpotent approximations [11] to control the 2R manipulator. Nilpotent approximation of control systems are higher-order approximations that prove useful when linearization does not preserve the original controllability properties. 3. Consider a

link planar manipulator moving in a horizontal plane as shown in

gure 6. The manipulator has two prismatic joints which are actuated, while the third joint is a revolute passive joint. The two prismatic joints are orthogonal. This system is STLC at the equilibrium. The same results hold for a three degrees-of-freedom planar manipulator with revolute joints (3R) moving in a horizontal plane (see gure 7). It can be veried that for a 3R manipulator, the controllability properties do not alter with the placement of the actuators. It can be shown [4] that for underactuated mechanical systems without the potential terms , the equilibrium conguration cannot be asymptotically stabilized using timeinvariant continuous state law. But if the system satises the sufcient conditions for STLC and coupled with the fact that the dimension of an underactuated mechanical 15

I 3 = inertia of the third joint m1 , m , m3 = link masses 2

joint-3

joint-1 F1 F2 joint-2

Figure 6: PPR planar manipulator

m 1, m2 ,m3 = link masses I1, I2, I3 l1, l2, l3 r1, r2, r3 = link moments of inertia = link lengths = centers of mass
l3

q3

l2

q2
l1

q1 Actuator 1

Actuator 2 x

Figure 7: 3R planar manipulator

16

system is atleast four (

) , the existence of asymptotically stabilizing con-

tinuous time-periodic feedback laws [12] is guaranteed.

6 Conclusions
There are many notions of controllability in the literature. We have illustrated the various types of controllability using both topological and Lie algebraic conditions. A few examples of underactauted mechanical systems bring out the importance of controllability properties in determining the stabilizability properties. Recently some new notions of controllability called conguration controllability [13] have been dened for simple mechanical systems2

A simple mechanical system is triple (

1. 2. 3.

is the conguration space,

) where

is the Riemannian metric on , and

is the potential energy function on .

17

Appendix

A Mathematical preliminaries
In this section we briey discuss some of the mathematical concepts from Differential Geometry [14] and Lie Groups which we will be subsequently using to analyze the various notions of controllability. All the machinery will be dened on a smooth manifold .

A.1

Vector elds and Tangent bundle


on a manifold is a map

at the point , that is, , where is the projection map and , the identity map. The set of tangent vectors at
is the tangent space at , denoted by . The tangent bundle of M, denoted by , is a differentiable manifold whose underlying set is the disjoint union of the tangent spaces to at the points

A vector eld

that assigns a vector

, that is

A.2

Lie groups and Lie algebra

Lie groups and Lie algebras arise in robot kinematics and also while studying nonlinear controllability. We provide some examples of Lie groups and Lie algebras that one usually encounters in robot kinematics. We begin by the denition of a topological group which is a space endowed with a group structure such that the following mappings are continuous.

from from

to to

Since we are dealing with manifolds, we talk of Lie groups. A -parameter Lie group which carries the structure of an -dimensional smooth manifold in such a way that the following operations

are

from

from

to to

18

Denition A.1 In general, a Lie algebra is a nite dimensional vector space , with a bilinear operation called the Lie bracket, satisfying 1. Skew-symmetry

2. Jacobi Identity

for all

Denition A.2 Let denote the algebra of a Lie group . A Lie subalgebra of is a such that implies . It can be shown that if is Lie subspace subgroup of with Lie algebra , then is a Lie subalgebra of . Examples of Lie groups in robot kinematics are the rotation group and the special Euclidean group

respectively. The set of smooth vector elds on with the Lie bracket is also

with the corresponding Lie algebras being and

Lie algebra. In local coordinates one calculates the Lie bracket of two vector elds as

where

and

are the Jacobians of

and

respectively. Successive

Lie brackets are computed in a similar way.

A.3

Distributions and Frobenius theorem

A distribution assigns a subspace of the tangent space to each point in in a smooth way. A -dimensional distribution on is a function a -dimensional subspace of . For any vector elds

such that distribution if it is possible to choose vector elds One calls a

there is a neighborhood and are a basis for , for each .

, where

is

with this property, in a neighborhood of each point . A distribution is said to be regular if the dimension of the subspace does not vary with . A distribution is said to be involutive if for any two vectors elds be integrable if for every point on with values in
,

is

also a vector eld with values in . A distribution (also called subbundle) is said to

, there is a local submanifold of containing

such that its tangent bundle equals restricted to this submanifold. We can now state a very important theorem relating the involutivity and integrability of the distribution. 19

Theorem A.3 Frobenius A distribution is integrable if and only if it is involutive. The above theorem is of paramount importance in the analysis and design of nonlinear control system. One such applicability is in the proof of the nonintegrability of the constraints in underactuated mechanical systems.

20

References
[1] V. Jurdjevic, H. K. Khalil, and F. L. Lagarrigue, Analysis and Design of Nonlinear Systems. CRC press, Inc., pp. 861-872, 1996. [2] H. Nijmeijer and A. J. van der Schaft, Nonlinear Dynamical Control Systems. Berlin: Springer-Verlag, 1990. [3] H. J. Sussmann, A General Theorem on Local Controllability, SIAM Journal on Control and Optimization, vol. 25, pp. 158194, Jan. 1987. [4] M. Reyhanoglu, A. J. van der Schaft, N. H. McClamroch, and I. Kolmanovsky, Dynamics and Control of a Class of Underactuated Mechanical Systems, IEEE Trans. on Automatic Control, vol. 44, pp. 16631671, Sept. 1999. [5] A. D. Lewis, A brief on controllability of nonlinear systems, (Dept. of Mathematics and Statistics, Queens University, Kingston, Canada, http://www.mast.queensu.ca/ andrew/). Technical report. [6] R. Murray, Z. Li, and S. S. Sastry, A Mathematical Introduction to Robotic Manipulation. NY: CRC Press, 1994. [7] R. W. Brockett, Asymptotic Stability and Feedback Stabilization, in Differential Geometric Control Theory, (R. W. Brockett , R. S. Millman and H. J. Sussmann, Eds. Boston, MA: Birkhauser), pp. 181191, 1983. [8] A. Luca, R. Mattone, and G. Oriolo, Steering a Class of Redundant Mechanisms Through End-Effector Generalized Forces, IEEE Trans. on Robotics & Automation, vol. 14, pp. 329335, April 1998. [9] H. Arai, K. Tanie, and N. Shiroma, Nonholonomic Control of a Three-DOF Planar Underactuated Manipulator, IEEE Trans. on Robotics & Automation, vol. 14, pp. 681695, Oct. 1998. [10] A. D. Luca, R. Mattone, and G. Oriolo, Stabilization of an Underactuated Planar 2R Manipulator, International Journal of Robust and Nonlinear Control, vol. 24, pp. 181198, July 2000. [11] H. Hermes, Nilpotent and High-order Approximations of Vector Field Systems, SIAM Review, vol. 33, no. 2, pp. 238264, 1991. 21

[12] J. M. Coron, On the stabilization in nite time of locally controllable systems by means of continuous time-varying feedback law, SIAM Journal on Control and Optimization, vol. 33, no. 3, pp. 804833, 1995. [13] A. D. Lewis and R. M. Murray, Conguration Controllability of Simple Mechanical Control Systems, SIAM Review, vol. 41, no. 3, pp. 555574, 1999. [14] M. Spivak, A Comprehensive Introduction to Differential Geometry, volume-I. Houston: Publish or Perish Inc., 1979.

22

Das könnte Ihnen auch gefallen