## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 46, NO. 9, SEPTEMBER 2001

**Passive Velocity Field Control (PVFC): Part II—Application to Contour Following
**

Perry Y. Li, Member, IEEE and Roberto Horowitz, Member, IEEE

Abstract—In contour following applications, the various degrees of freedom of a mechanical system have to be well coordinated, but very often, the speed at which the contour is followed is not critical. Moreover, in the context of machining, the system has to interact closely with its physical environment. When the contour following task is represented by a velocity field on the configuration manifold of the system, the coordination aspect of the problem is made explicit. The passive velocity field control (PVFC) scheme developed in the Part I companion paper [7] can then be applied to track the defined velocity field so that the desired contour is followed, and to ensure that the interaction of the closed-loop system with the physical environment is passive to enhance safety and stability. Unfortunately, for some contours, an encoding velocity field on the configuration manifold does not exist or is difficult to define and, as a consequence, the PVFC cannot be directly applied. For systems whose configuration manifolds are compact Lie groups and the desired contour is represented by a parameterized trajectory, a general methodology is developed, using a suspension technique, to define a velocity field on a manifold related to the configuration manifold of the system for which PVFC can be applied. With this strategy, timing along the contour can be naturally varied on-line by a self-pacing scheme so that the contour tracking performance can be improved. The experimental results for a 2 degree of freedom robot following a Lissajous contour illustrates and verifies the convergence and robustness properties of the PVFC methodology. Index Terms—Coordination, lie groups, passivity, self-pacing, suspension, velocity field.

Fig. 1. In a timed trajectory based controller, the system may leave the desired contour to keep pace with the moving trajectory.

I. INTRODUCTION HE manipulation task of a mechanical system is traditionally specified by means of a desired timed trajectory in the workspace. The control objective is to track this trajectory at every instant of time. In many contour following applications, the actual timing in the desired trajectory is unimportant compared to the coordination and synchronization requirement between the various degrees of freedom. For example, in machine deburring, the machine is required to traverse the contour at the maximum speed without exceeding the limits on the force experienced by the deburring tool. The requirement that the machine be at a position specified by a predetermined trajectory at each time may be an overly stringent imposition. Indeed, it has been demonstrated in [2], [11] that in the presence of uncertainties, trajectory based contour tracking algorithms tend to exhibit

Manuscript received July 11, 1997; revised October 22, 2000 and June 23, 2001. Recommended by Associate Editor O. Egeland. P. Y. Li is with the Department of Mechanical Engineering, University of Minnesota, Minneapolis MN 55455 USA (e-mail: pli@me.umn.edu). R. Horowitz is with the Department of Mechanical Engineering, University of California at Berkeley, Berkeley CA 94720-1740 USA (e-mail: horowitz@me.berkeley.edu). Publisher Item Identifier S 0018-9286(01)08826-2.

T

the radial reduction phenomenon where the radius of the actual contour is smaller than the desired one. Fig. 1 illustrates the situation. As an alternative, it is proposed in Part I of this paper [7] and in [6] that coordination critical tasks, such as contour following, can be encoded by a time invariant velocity field on the configuration space . An appropriately designed velocity field, which defines a desired velocity for each configuration of the mechanical system, guides the system to approach the contour in a well behaved manner. To encode a contour following task, the velocity field should have the following properties: i) its value at each point of the contour must be tangent to the contour; ii) the flow of the field has a limit set which is contained in the contour. Fig. 2 shows a velocity field for the task of tracing a circle on a rectangular configuration space. Notice that the flow of the field, which is determined by tracing the arrows, converges to the circle, and that the arrows are tangent to the circle. be the trajectory of the mechanical system. The Let velocity field tracking control objective is to cause the -veto vanish locity field tracking error, , thus enabling the contour to be followed for some asymptotically. Notice that the mechanical system is not required to be at a particular position at each time. Instead, the velocity field guides the robot to approach the contour in a well behaved manner. The passive velocity field control (PVFC) algorithm proposed in [5]–[7] ensures that a scaled multiple of the specified velocity field is asymptotically followed, i.e., for some . In addition, when the feedback system is treated as an input–output system with the environment force as its input, the velocity as its output, and a supply rate

0018–9286/01$10.00 © 2001 IEEE

The rest of this paper is organized as follows: in Section II. For these reasons. deburring). For example. Since the physical environments with which the machine interacts are often passive themselves. which are developed in detail in [6]. stability and safety can be enhanced. and underwater vehicles.g. The general configuration from the desired location procedure for the design of such a velocity field will be developed for mechanical systems with compact Lie groups as their configuration spaces. In Sections III–IV. a contour encoding velocity field must first be defined. 3 to validate the proposed method and to illustrate the convergence and robustness properties of the PVFC as presented in [7]. for many desired contours. determined by the scalar . If the closed-loop control system is passive with respect to the environment mechanical power. . this difficulty is overcome by using a parameterized trajectory description of the contour. we constructively show that a contour following problem in which the contour is described by a parameterized curve reduces to the velocity field following problem. 2. In the context of machining (e. such as the magnitude of the interaction force. then the control scheme also maintains the passivity of the closed-loop system with respect to this supply rate in the sense that: for all and (1) One advantage of maintaining passivity is that the coordination requirements of the contour following task becomes naturally embedded in the feedback system and will be decoupled from the speed at which the contour is followed. In Section III. This is achieved by defining a time invariant velocity using the field on the suspended configuration space concepts of gradient vector fields [4] and self-pacing. the method. In Section IV. satellites.. In this paper. such a velocity field is either difficult to define or may not even exist. [7]. 3 does not have a unique nonzero tangent and any velocity field must command the mechanical system to stop at the intersection. which is the environment mechanical power input to the system. the energy in the system is limited to the initial energy of the system and the additional energy input from the environment. Unfortunately. this procedure is performed for a 2 DOF SCARA . Fig. 3. based on the deviation of the . the center point of the Lissajous contour in Fig. so that PVFC can be applied.LI AND HOROWITZ: PASSIVE VELOCITY FIELD CONTROL: PART II 1361 Fig. cutting. A parameterized where and is the trajectory is a curve configuration space of the mechanical system. The speed. In order to apply PVFC to contour following. Extensive experiments have been conducted for a 2-link revolute jointed serial robot following the Lissajous figure in Fig. the PVFC algorithm in [6]. Under the same assumption. It reduces to the traditional desired trajectory if is time. The basic idea behind self-pacing is to increase or decrease the desired rate of the progression of the parameter relative to the desired speed of the mechanical system. where contour following problems are frequently encountered. Contour of the desired parameterized trajectory. The latter can be controlled by a variety of additional control loops taking into account other environment factors. [7] is a good candidate for contour following applications. we briefly review the PVFC algorithm and its properties. A velocity field for the task of tracing a circle on a rectangular configuration space. we show that the mechanical system converges to the parameterized trajectory. A suspended system of the mechanical is then defined so that the configuration of the contour are cosystem and the parameterization ordinated. the robot or the machine tool has to interact closely with its workpiece. defined to be the action of on . to illustrate robot whose configuration space is the 2-torus. The latter can happen when the tangent to the contour is not unique. These constitute a wide class of important mechanical control systems including revolute jointed manipulators. is a function of the energy storage present in the feedback system. the procedure is generalized to mechanical systems with compact Lie groups as their configuration .

Coriolis matrix in coordinates. The readers are referred to [6]. such that the kinetic energy of the augmented system is a constant (6) globally and exponentially. defines the kinetic energy of the augmented system via: (5) denotes the inner products defined by the where .e. and skew-symmetric matrices. 9. VOL. as long as where is the feedback gain in (7). Then. 1) Defining an augmented system as a product system with . inertia metric 2) Defining an augmented desired velocity field of the form: . [7] for details. except from a set of initial conditions of measure 0. Remark 1: The exponential velocity field convergence property is established using the following Lyapunov function. we also have and when (11) and are the configuration and where for the augmented the inertia Riemannian metric on system. i. denotes the action of the co-vector (or 1-form) on the denotes the inner tangent vector (or a vector field) . Section VI contains concluding remarks. we briefly review and summarize the design procedure for and properties of PVFC. and depend on the augmented . Experimental results are given in Section V. [7]: The closed-loop system (8) consisting of the augmented system (4) and the coupling control (7) has the following properties. respective bold letters and symbols denote their coordinate representations. is given by (8) The key passivity and convergence properties of PVFC are summarized below: Theorem 1[6]. . in the absence of environment solution if forces. Throughout the paper. is a Lyapunov stable . time varying two-form. NO. satisfies (13) where if and only if is a nondecreasing function. Its dynamics can be expressed geometrically in terms of a Levi–Civita connection [7] (2) or in terms of coordinates as (3) where is the Levi–Civita connection associated with the inand are the inertia matrix and ertia metric . the latter is given by . the design methodology presented in [7] consists in the following three steps. . Therefore. is compatible 1) The closed-loop affine connection so that the kinetic with the augmented inertia metric in (5) satisfies energy of the augmented system (9) 2) The controlled system is passive with respect to the . supply rate . In coordinates. the desired velocity field that the system Given is supposed to track. II. SEPTEMBER 2001 spaces. and the main convergence results are also presented. where are the coordinate basis vectors.1362 IEEE TRANSACTIONS ON AUTOMATIC CONTROL. REVIEW OF PASSIVE VELOCITY FIELD CONTROL (PVFC) In this section. unbold letters or symbols denote elements in a manifold. We consider a -degree of freedom fully actuated mechanical system with configuration space subject to both control forces and environment forces . because of (9). product defined by some Riemannian metric (sometimes with subscripts to signify the inertia metric that defines it). between the plant (2) configuration space and a fictitious flywheel (4) 3) Defining a coupling control in (4) of the form (7) where gain constant. .. the . Thus. which will be useful for establishing the contour following convergence result in Section IV: (12) where . contraction operator. be related to the kinetic energy of the system via: 4) Let (10) Therefore. Their exact definitions can be found in velocity field [6] and [7]The closed-loop system consisting of the augmented system (4) and the coupling control (7). define the -velocity field error to be 3) For each . the coordinate repis so that resentation of . 46. In this case. and .

then the parameterized trajectory reduces to the traditional desired trajectory. Suspension First. and 2) to apply PVFC to ensure that this velocity field is tracked. iii) apply the PVFC methodology to the suspended system. we first consider the two-link mais the torus . To illustrate the procedure. A. In addition. which shall be verified from the experimental results in Section V. we form a suspended system by forming a product system from the original dynamical system and the dynamics : of the evolution of the parameterization Fig. being the inertia matrix. in3) As the energy of the system increases (i. nent of . Thus. Following [4]. 1) If the environment force is in the direction is the desired momentum and where . ii) design a velocity field on the suspended system configuration space in order to require that the mechanical system and the parameter are coordinated. The more general case where the configuration manifold is a compact Lie group will be presented in Section IV.LI AND HOROWITZ: PASSIVE VELOCITY FIELD CONTROL: PART II 1363 rate of convergence is given by in the neighborhood of . forming a suspended system. Thus. Velocity Field Design on It is now necessary to design a velocity field on . can by be eliminated if the feedback gain is sufficiently large. let eterized trajectory in these coordinates. identified with nipulator in Fig. 2) If the environment force is in the direction that annihilates . (14) where the first row in (14) corresponds to the dynamics of the being the coordinates. the configuration space of the suspended system in (15).. Using the joint angles as parameterized trajectory be the desired paramcoordinates. then the velocity tracking error will be ultimately bounded. then its effects on the velocity field tracking error given with defined in (10). represents the environment forces. Since the design of a contour encoding velocity field may be difficult and for some contours impossible. the lower bound for the gain needed in 1) as well as the ultimate bound in 2) both decrease. . being the Corirepresents the control forces. The suspended system (14) corresponds to a mechanical system with configuration matrices and inertia and Coriolis under the influence of the environment force control torque and . 4. The readers are referred to the companion paper [7] for details. III. 4 will be the coordinates of the system. and olis matrix. . Notice that attains its minimum when and increases as deviates more from the desired . we define a potential function by (17) are positive constants. 4 where . A parameterized where and is the trajectory is a curve configuration space of the manipulator. creases). we develop a general methodology to achieve contour following by representing the contour as a parameterized trajectory. If is time. i. given a . A two link direct drive robot..e. APPLYING PVFC TO CONTOUR FOLLOWING: AN ILLUSTRATIVE EXAMPLE IN In this and the next sections. two link robot with . is a measure of the tracking perforlocation mance. instead of just a set of points. PVFC will control the dynamics of both the mechanical system itself and of the parameterization . the velocity field will be designed so as to be related to a gradient vector field of a potential function. we discuss the application of PVFC to contour following problems. the PVFC has the following robustness properties. Define the angle error to be function (16) Next.e. Equation (14) can be written as (15) The next steps will be 1) to define a suitable velocity field for the suspended mechanical system that encodes the contour following task. we expect that the performance of the system to improve as the speed of the system is increased. The key ideas are to i) dynamically extend the mechanical system to include the parameter of the parameterized trajectory as another coordinate. Thus. and is the th compowhere . the joint angles in Fig. B. Hence.

Self-Pacing . denoted by Let maps a tangent vector at to a tangent such that if . so that the velocity field emphasizes the action that decreases the tracking error is large. The experimental results of this control scheme are presented in Section V. is uniformly positive. For . of . which measures the deviation of function the configuration from the desired location. One possibility is to and depend on the value of the potential make in (17). However. The potential function in (17) ensures that. Before we proceed. NO. is not needed (see [6].1364 IEEE TRANSACTIONS ON AUTOMATIC CONTROL. given in Section III to the generic setting when the configuration space is a compact Lie Group. . . 9. If the speed of the progression parameter is large. when the tracking error when would progress at a slower is large. then (19) so that IS A COMPACT LIE GROUP (20) is lower bounded. Hence. (20) implies that Because converges. and the right translation by is de. The design procedure for the velocity field is motivated by the navigation function approach to real-time path planning introduced in [4]. Consequently. the left translation by is denoted by tion operators. the push forward of . SEPTEMBER 2001 The velocity field on is defined to be a linear combination of two terms as follows: controller will now have 3 internal states: . . [7] for details). However. . implies that from almost anywhere. 2) an identity element such that . noted by . if the direction of is appropriate metric on opposite to it is the self-pacing parameter such that as inwhere creases. IV. it is a precompensation term coordinating the dynamics of and so that if of the suspended system will remain unchanged. it would Notice that (19) is independent of take some time for the angle error to decrease to an acceptable level. there exist 1) a smooth group operation . the configuration space of such an aug. The overall mented system becomes In this section. must converge to as long as or . In general. Since the passive velocity field controller involves augmenting the system to be controlled by the dynamics of a fictitious flywheel with configuration space . . denoted by The group operator defines a family of left and right transla. a critical point of . C. a large portion of the contour will be tracked poorly. We can decrease this portion by decreasing . and 3) a smooth inverse operator such that . • The first term in (18) satisfies Therefore. the action of on is the 1-form such that for all . A smooth manifold is a Lie Group if it has a group structure. 46. to denote Let . the emphasis on eliminating contour following error is also increased. Specifically. we shall use the notation. notice that the components are in fact (21) which can be shown to be the gradient vector field for an . itself. Notice that the flywheel configuration flywheel velocity. the PVFC maOnce a velocity field on chinery summarized in Section II can be applied to the suspended system in (15) to track the velocity field (18). vector at . (15) is in this direction. The idea of self-pacing is to exploit this relationship to improve contour following performance. the relative magnitudes and control the speed of progression of the deof sired parameterized contour relative to the rate of convergence. (18) are positive bounded functions. the desired trajectory can be: speed. The where rationales for this definition of in (18) are as follows. Indeed. is designed. which are characterized by . VOL. GENERAL CASE WHERE • The combined effect of the two terms in (18) is that if for some (as expected from (11) in Theorem 1). and • For the second term. and the fictitious . where denotes the Lie (directional) derivative in the direction . since the only stable critical point is . we generalize the procedure of applying PVFC to contour following problem. For example. let us recall some differential geometric concepts and notations.

one can choose to define ever. The error function can be defined to be (25) is the desired parameterized trawhere jectory.. where . The navigation function is defined to be (26) where point of .. are satisfied. bounded when embedded in The assumption for to be compact may be relaxed in many situations. and only the global minimum is stable. In coordinates. invariant.. and are the control and environment forces for the suspended system. a minimum). Since is a Lie group. as long as some of the boundedness conditions. a natural definition of the group error function is (24) . .e. The group action is the matrix multiplication and the identity element is the identity matrix. • a vector field on which leaves the error corresponding to the first term in (18). Remark 2: The navigation function is defined so that all the critical points are isolated. Next. . since is assumed to be compact. • a group error function with the appropriate • a potential function properties. 2) For revolute jointed robots. The desired contour is defined where is the doby a parameterized curve main of the parameterization. can be identified with where is the space of 3 3 skew symmetric matrices. this is automatic. it can be shown [1] that is the only critical point at which has a positive definite Hessian (i. 2) it has a unique minimum at the identity with value 0. The gradient vector field is defined by A. with each element space of the satellite is identified with a 3 3 real orthogonal matrix with determinant 1. Definition 1: The twice differentiable function is a navigation function if it has the following properties: are 1) all the critical points nondegenerate. is given by connection associated with . The geometries of many robotic systems (see. To generalize the velocity field design procedure in Section III-B. Each element of can be identified with where is the angle of the th joint and . and the gradient vector field of . Velocity Field Design Assumption 1: The configuration space of the mechanical system is a compact Lie group. . [8]) can be naturally described in terms of Lie Groups. and the derivative is uniformly bounded. This together with the compactness of ensures that the converges to a critical point. Thus . and let be the inner products that define a Riemannian metric on . the formulae given below will have to be properly adjusted. be the parameterized trajectory on a Lie group . HowSimilarly. being compact can be considered equivalent to being closed and for some sufficiently large . i. Let be a symmetric matrix with distinct eigenvalues and .e. . . The is and the group action identity element of is given by . 1) Group Error Function and Navigation Function: Let . the gradient vector field of the function must converge to identity from almost everywhere. a navigation function as defined below possesses the sufficient conditions. where is the circle.e. and is the Levi–Civita pended system. is the -torus ( times). for example. . Following [4]. corresponding to the second term in (18). also requires that However. a navigation function over can be defined as follows. iff is a critical This is so iff is symmetric for which there are exactly four instances. Following [4].LI AND HOROWITZ: PASSIVE VELOCITY FIELD CONTROL: PART II 1365 Given . the Hessian (which is the matrix of second derivatives) in any coordinate system. i. A navigation function is in fact a Morse-Bott function [3] whose definition have compact sublevel sets. A must be defined so that the flow of the function . the potential function in (17) must be generalized. at each critical point has nonzero eigenvalues. the configuration . Of these instances. . Suspended System The suspended system is formed in exactly the same way as in Section III. B. and negative gradient flow of to the global minimum from almost everywhere. . which will become apparent. (22) which can be rewritten as (23) is the configuration of the suswhere is its inertia metric. . . . Examples: 1) For the control of satellite orientation. we need to define . .

and be the desired parameterized trajectory. which may not be related to the Riemannian metric that defines the inertia of the suspended system. 9. . respecward of the maps tively. and . The Hessians are given by the so that it is positive definite matrix are multiple of . where is a 3 3 skew-sym(26). SEPTEMBER 2001 . be the velocity field deProposition 1: Let and being the components in and fined in (30). we obtain expression for is periodic. a Riemannian metric is necessary. Then To enable self-pacing. corresponding to the iff . denote the pushfor. A navigation function can be tion in (16) when defined to be (28) (29) which is the required result. let us define a Riemannian metric on as fol. then for each in (31) . we have Therefore. . . identity element which is also the global minimum. it is equivalent to the error funcSince . Let where denotes a unit velocity vector and are gain functions. the left and right translation operators. . Examples: 1) Returning to the satellite orientation example where . this is the familiar expression for . is 0 and it corresponds to The global minimum of . Denote the inverse map by . where and for the parameterby ized trajectory . In the above derivation. Define the desired velocity field on . using the product rule on Let the error and navigation functions be defined as in (25). NO. . and . if negative gradient field of evolves in the direction of the . Expanding the . be a vector field on Lemma 1: Let where . and ’s are positive conwhere stants. and . VOL. (27) Let with where denotes the pushforward of the map . for each . and can be defined in the same way as in (21). we used the . where are 3 3 skew lows: let symmetric matrices The push forward operator is essentially a derivative operator. and . defined so that and be such a metric. and are. Notice that if where In matrix Lie groups. Other critical points are characterized by . with respectively. i. . fact that . the formula for and . 46.e. be where Proof: Let .. Let us rewrite the error function in (24) as Proof: This is a direct consequence of Lemma 1 applied to the velocity field in (30). .1366 IEEE TRANSACTIONS ON AUTOMATIC CONTROL. Hence. To define a gradient vector field. The error function can be defined to be . 2) Gradient Vector Field on : We need to define a vector such that field (30) where denotes the pushforward of the map . denotes the pushforward of the error function . and .

in (24) converges to eterized trajectory tracking error . For any skew symmetric 3 3 matrix and are positive scalar bounded functions. . . 2) Consider again the revolute jointed manipulator example where . . The DOF closed-loop system that results has configuration space of after augmenting the DOF suspended system (23) by the fictitious flywheel. the speed of the progression parameter is . . is the in constant used to construct the augmented velocity field is the gain parameter in (30). . which is the Euclidean metric. . and with the degiven by (30). In this case. and the second in term is merely the component of Now consider the Lyapunov function (32) . where . Second.e. . . . .. or the configuration space be punctured at those locations. . . in fact conalmost every initial condition. so that The desired velocity field in (30) is then given by: since . given by is the kinetic energy of the augmented system in (5). and incorporating the coupling in (30)). and define two types of norms on First. we obtain Remark 3: Obstacle avoidance can also be incorporated by defining suitable navigation functions. Let the Riemannian metric on be . is the projection. identify an element of by . . Its dycontrol (7) (which depends on namics will be given by (8) with the augmented inertia metric given in coordinates by where is the inertia matrix for the suspended system. (28). Hence verges to from almost everywhere. and be the Riemannian metric on inertia metric used to compute the gradient vector field . Let . . the identity element. Readers are referred to [9] and [10] for details of this approach. Further. the navigation function should be defined such that its value is large at the locations of the obstacles. be the inner product defined by the augmented Let . and Proof: We prove this theorem in a coordinate independent manner. be the desired parameterized trajectory to be followed. is a constant. PVFC Applied to the Suspended System The PVFC summarized in Section II can now be applied to degree of freedom (DOF) suspended system in (23) the to track a multiple of the desired velocity field (30). From one of the critical points of the navigation function . From this and the fact that . the -torus. where where is given by: . notice from the definition of the desired in (30) that the velocity of the error velocity field is given by: or Let There exists be the desired parameterized trajectory.LI AND HOROWITZ: PASSIVE VELOCITY FIELD CONTROL: PART II 1367 metric matrix to be determined. For each . Define and the navigation function the error function as in (27). Asymptotically. . . i. Theorem 2: Consider the closed-loop system (8) that results from the suspended system (23) under the control of the passive velocity field controller defined in Section II. recall from Theorem 1 that when . C. . The contour following result is given in the following theorem. where . (6). sired velocity field . and is the inertia of the fictitious flywheel. . . The velocity field on . the paramIn the absence of environment force.

a very simple friction compensation scheme was used in some of the experiments . we have Remark 4: Since is a navigation function. i. is the navigation function. uniformly bounded. Using stanconverges and unidard Barbalat’s arguments (i. in (5). Specifically.. from (13) (33) Let be the maximum of the induced norm of .e. NO.1368 IEEE TRANSACTIONS ON AUTOMATIC CONTROL. should remain the total energy level in the system. dissipative forces and imperfect friction compensation can cause the energy level to vary. If implies that converges to one of then the isolated critical point of . since all the critical . which is determined by the kinetic energy of the augmented system. Boundedness of is necessary for the constant in (6) to exist. 2) It is used to make sure that velocity field defined in (30) and are is bounded when the gain functions where Nm. Nm. In deburring applications.. on a compact manifold). . the requirement for compactness can be replaced by other technical assumptions in certain cases. A. 1) Compactness is used in conjunction with the definition of navigation functions to ensure that implies that converges to a critical point of . imply that . and formly continuous. by Hence. and is a positive number to be determined. the assumption that has a group property may also be replaced if a meaningful alternate definition of group error function in (24) can be made. However. that is able to distinguish two different points in . from (30).e. it is necessary that the induced is bounded. has compact level sets (e. compactness will not be necessary if these conditions can be independently guaranteed by other means. (33) can be written as is finite. V. The passive velocity field controller was applied to the 2 link direct drive SCARA manipulator in Fig. 46. Although we have restricted our attention to mechanical systems whose configuration spaces are compact Lie groups. there is a such that Remark 1. constant and the rate at which the parameterized trajectory progresses would be determined by the initial conditions. e. Thus. from (32) and (34). in addition to the friction compensation in in (35). the identity element of . as of the velocity field. and from almost all initial stable. a velocity field on was designed using (18) and (21). Many norm of . we have used compactness in the following. VOL. ). 3) In the proof of Theorem 2.. There exists a neighborhood of at and Thus. the component Finally. conditions. is given by: The image of this parameterized trajectory is the Lissajous figure shown in Fig. . as long as . These terms were in (7). EXPERIMENTAL RESULTS Notice that because is assumed to be compact. . are unpoints except for . the matrix in (34) is positive–definite for all times. Therefore. is nonzero. when the passive velocity field controller is used. the following exogenous signal was also added to (7) (36) where is a damping coefficient. SEPTEMBER 2001 where is defined in Remark 1. such as by the careful design of the navigation function. 3. the nominal speed can be made to depend on the measured force on the tool so that the maximum tool force is not exceeded. satisfy this condition. Similarly. in actual implementations. added to the In theory. given by parameterized trajectory (34) and (therefore. Since PVFC (Theorem 1) ensures that . with the fictitious flywheel velocity was applied to the suspended system to track the extended velocity field as described previously. It can be shown that to stabilize at .. Notice that this trajectory is not one to one so that a time invariant velocity field on that traces out the same contour does not exist. For this parameterized trajectory. convergence of to is locally exponential.Thus. is defined to have This condition can be satisfied if compact sublevel sets. and is the desired momentum of the augmented-suspended tends to cause the velocity system. noncompact spaces. .g. as long as ). However. Then. its Hessian is positive definite. Section II. and the PVFC presented in in (4). Experimental Setup Because of the existence of significant friction in the joints. 4 to track a desired . in order to control the nominal rate at which the parameterized trajectory progresses. 9.g.

. D. position A controller that utilizes a time parameterized trajectory will also exhibit this behavior. This is consistent with the robustness result in [7] and summarized in Section II. the fictitious flywheel velocity) was set to 10 and the manipulator was initially at rest. Effect of Self-Pacing The effect of self-pacing on the response of the manipulator when the initial position of the manipulator is not on the desired contour is investigated next. parameter values.e. The contour following performance at different self-pacing are shown in Fig. In Fig. 8. 5 B. until the manipulator the parameter progressed slowed . When self-pacing was used . since the initial position error. nominal speed control and the passive velocity field controller is shown in Fig. These correspond to the regions when the joint velocities were required to reverse direction. as evidenced by the fact that at each value of . 6. The tracking performances in this case at the different values are shown in Fig. the initial total energy was 50. Bottom: Kinetic energy. . and hence . the friction compensation scheme in (35) was in effect and in (36) was kept at the nominal value of 1. Notice also that as in (21) decreased. Block diagram of control scheme. which measures the tracking error. we investigate the effect of nominal speed on tracking performance. Notice that as the total energy set to the nominal value of system dissipated. 6). E. The initial value of the internal by setting state of the controller (i. tour. the manipulator was still able to follow the contour although the rate of traversal decreased (Fig. Effect of Nominal Speed: Finally. 9. Thus. Friction compensation was not used in these experiments. As shown in Fig. In these experiments. Hence. Robustness to Friction Since stiction/friction is suspected to be the main cause of the tracking error. However. As expected. C. as in the friction compensated case. the manipulator approached from the initial to the desired contour by following a curved path. the tracking performance as measured by the potenin (17) became progressively worse as the tial function energy level decreased. tracking performance improved when was increased. 5. Fig. formance was obtained at the expense of decreasing the rate at which the trajectory progressed at these critical regions along the contour. Unforced Response We first investigate the unforced response of the manipulator in (36). A comparison between Figs. the value of . when (no self-pacing). The desired contour repeated every Notice that the error was large at four distinct positions. 8 and 9 shows that tracking performance deteriorated when friction compensation was reis moved. The self-pacing parameter in (21) was . 8. The relationship between the robustness bounds and the energy level in the system controlled Fig. Top: Contour traced. stiction and friction are suspected to be the main cause of these errors. 7. was large. However. and 3 while was kept at 100. Thus. Middle: Value of potential function U (E (t)).LI AND HOROWITZ: PASSIVE VELOCITY FIELD CONTROL: PART II 1369 The overall control scheme that combines the friction compensation. the improved perwas increased. 2. The nominal speed was varied by setting in (36) to 1. the potential function is plotted against the parameter during one pass of the concoordinates units. Unforced response. it is interesting to investigate the performance of the control scheme when the friction compensation in (35) was removed. the robot became close to the desired position approached the desired contour nearly in a straight line before moving along the contour. approximately doubled when friction compensation is removed.

Top: Contour traced. however. Middle: Value of the potential function U (E ) during one pass of the contour. The improvement. 46. and the environment power as the supply rate. 8. Top: Contour traced. VOL. 50. 10 shows that was significantly lower when indeed the error measure was increased from 1 to 2. _ by the PVFC suggests that robustness to disturbances should improve at higher energy levels. In order to apply this control technique to contour following tasks. it is expected that the tracking performance should improve as increases. 7. is not as notable when was increased from 2 to 3. Effects of different self-pacing parameter R = 0. The controller mimics the dynamics of an energy storage device and preserves the total energy of the system. 200. VI.1370 IEEE TRANSACTIONS ON AUTOMATIC CONTROL. . Bottom: Rate of trajectory progression during one pass of the _ contour. SEPTEMBER 2001 Fig. Fig. a velocity field must be designed which encodes the contour. the velocity as the output. 9. Middle: Value of potential function during one pass of the contour. rendering the closed-loop system passive when the environment force is considered as the input. 9. the application of the PVFC methodology to contour following tasks for mechanical systems is discussed. A methodology for designing such a velocity field is presented when the configuration space Fig. CONCLUSION In this paper. NO. 200 when friction compensation was removed. Since the nominal speed determines the total energy level. 100. Contour traced out by the manipulator with and without self-pacing when the initial position was not on the desired contour. Effects of different self-pacing parameter R = 0. Bottom: Rate of trajectory progression during one pass of the contour. 100. Fig.

. S. Horowitz. S. Berkeley. His research interests are in dynamic systems and control as applied to fluid power systems. “Total energy for mechanical control systems. 71–116. no. vol. of Mechanical Engineering.” Ph. Perry Y. degree in mechanical engineering from the University of California. 1993. vol. 1. [11] E. REFERENCES [1] D. A simple self-pacing scheme which takes advantage of the flexibility of the timing along the contour to improve tracking performance is also proposed. [9] E. degree (with highest honors) and the Ph. Dept. Dept.” IEEE Trans. and optimal control. CA. . Univ. [5] P. Y. in 1987. 10. pp.. Webster.” in Dynamics and Control of Multibody Systems.D. 1999. H. C. Rational Mech.” Ph. manufacturing. respectively. Automat. J. Sept. The procedure involves suspending the contouring parameter by considering it as an extra coordinate. Venezuela.” Trans. Chiu. ser. J. . 1994. Math. Eds. In 1982. _ of the system is a compact Lie group. 1982. mechatronics. Berkeley. A Mathematical Introduction to Robotic Manipulation. transportation systems. 46. Prior to joining the University of Minnesota in 1997. Dr. Marsden. 1990. and J.” IEEE [10] Trans. 1991.. and imaging and printing. part I. He is a member of ASME. X. “Passive velocity field control (PVFC) Part 1—Geometry and ro[7] bustness. Boston. pp. 1995. Roberto Horowitz (M’89) was born in Caracas. Li (S’87–M’96) received the B. of California. vol. “Symmetry and bifurcation in three-dimensional elasticity. Providence. “Identification and Control of High-Speed Machine Tools. U. Marsden. RI: American Mathematical Society. Automat. 295–331.(Hons) and M. Horowitz received the 1984 IBM Young Faculty Development Award and the 1987 National Science Foundation Presidential Young Investigator Award. University of California.” Ph. pp. Fig. “Exact robot navigation using artifical potential functions. Robot. Tung.” Arch. he was a Researcher at Xerox Corporation. Univ.. [3] M. The effectiveness of the contour following scheme is experimentally demonstrated on a two link direct drive robot following a Lissajous figure. Hirsch. 1346–1359. Automat. Dept. . 131–157. Murray. Top: Contour traced.K. St. Amer.0 (without friction compensation). where he is currently a Professor. Li. AMS series on Contemporary Mathematics. “Passive velocity field control of mechanical manipulators. and 1995. E. Berkeley. with applications to microelectromechanical systems (MEMS).D. [4] D. Sept. degree in mechanical engineering from the University of California. 15. 3. Analy. degrees in electrical and information sciences from Cambridge University. Krishnaprasad. Contr. dissertation. and the Achievement and Special Recognition Awards from Xerox. “The construction of analytic diffeomorphisms for exact robot navigation on star worlds. Li. Robot. degree in biomedical engineering from Boston University. Rimon and D. Wan. “Coordination Control of Multiple Axes Mechanical System: Theory and Experiments. in 1955. pp.S. J. E. He is currently Nelson Assistant Professor in Mechanical Engineering at the University of Minnesota. Dr. robotics. Berkeley. vol. Chillingworth. of Mechanical Engineering.LI AND HOROWITZ: PASSIVE VELOCITY FIELD CONTROL: PART II 1371 [2] T. computer disk file systems. and intelligent vehicle and highway systems (IVHS). learning. 1989. nonlinear. Li was awarded the 2000–2002 Japan/USA Symposium on Flexible Automation Young Investigator Award. Sastry. Simo. University of California. Li and R. man–machine systems. no. Oct. He teaches and conducts research in the areas of adaptive.A. he joined the Department of Mechanical Engineering. Cambridge. robotics. Z.A. vol. P. E. no. Bottom: Rate of trajectory progression . R. 1990. and then by designing a velocity field based on a gradient vector field of a navigation function. He received the B. 501–518. 751–763. Differential Topology: Graduate Text in Math. CA. and the Ph. and the contour is specified by a parameterized trajectory.. vol.” IEEE Trans.. Paul. the M.S. [8] R. 5. 237. 2. 8. Berkeley. 80. CA. NY. 2001. pp. 4. vol. 33.0. Koditschek. Effects of different energy levels: r = 1:0. 1992. 1994. respectively. of. New York: CRC Press. 1976. [6] P. New York: Springer-Verlag.. Koditschek. and Y. Y.D. 97.D. and S. Berkeley.. of Mechanical Engineering. Middle: Value of potential function. MA. in 1978 and 1983. “Self-Optimizing Control and Passive Velocity Field Control of Intelligent Machines. Soc.D. pp. Apr.

- Wald General Relativity Solutionsuploaded byGerardPc
- MATLAB Expressionsuploaded byAhmed Ahmed
- MAA101 Chapter 2uploaded byHoàng Đình Cường
- contraction paperuploaded byvargil
- MPK - 04 Struktur Organisasiuploaded byAnonymous 5To5jx82
- 10.1.1.8.1137uploaded byajitsatapthy
- Chapter1.pdfuploaded bySahaya Grinspan
- Cvf-6.6 Extended Math Libraryuploaded byitetejac
- math123uploaded byShailendraPatel
- Gravastarsuploaded bygatitahanyou
- Ch-3uploaded byKumar Gaurav
- Matrix Methods,Applied Linear Algebra,Third Editionuploaded byk_ij9658
- 6.4Lectuploaded bydvdiyen
- FEM_Appendix_Auploaded byjtallapragada
- Matrices.pdfuploaded bySudheer Kothamasu
- 3uploaded bythanhthoakt
- MIT18_022F10_l_5uploaded byAtudh
- R-Programming Online Training in Hyderabaduploaded byMohan R
- A Bernstein Property of Affine Maximal Hypersurfacesuploaded bymerchinets
- Generaluploaded byNima Dharmapal
- Matlab Introuploaded byFachransjah Aliunir
- Locality of Gravitational Systems from Entanglement of Conformal Field Theories.pdfuploaded byAronOliveira
- EURMT-202.rtfuploaded byJohn Cena
- Octave Tutorialuploaded byPeter Kolossa
- lec34uploaded byEdwardklaus
- Script File to Calculate the Fugacity of a Mixture Based on Theuploaded byHoàng Minh Ái
- Civiluploaded byme22244
- Palm M3Chapter2uploaded byramu reddy
- Practical Exam Question Papers - XII -Externalsuploaded bySudhakar R
- 2037977530maths Value Based Question Bank for Class Xii 2013-14uploaded byArjun Prasad

Close Dialog## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

Loading