Proceedings of the 2001 IEEE International Conference on Robotics & Automation Seoul, Korea • May 21-26, 2001 On the Stiness Control and Congruence Transformation Using the Conservative Congruence Transformation (CCT) Yanmei Li y and Imin Kao Department of Mechanical Engineering SUNY at Stony Brook, Stony Brook, NY 11794-2300 Abstract The conservative congruence transformation (CCT), K Kg = J T Kp J , was proposed by Chen and Kao as the correct congruence transformation to replace the conventional mapping, K = J T Kp J , proposed by Salisbury in 1980. The conventional mapping was shown to lead to physically inconsistent results when external force is present in stiness control. Theoretical proofs are also provided to show the conservative nature of the CCT, and the non-conservative property of the conventional mapping. The CCT is established as the general and valid mapping of the stiness matrices between the joint and Cartesian spaces of robotic manipulators. In this paper, the work of CCT is extended to a redundant planar manipulator. Numerical simulations are presented to illustrate issues related to the application of generalized inverse in the analysis of redundant manipulators. 1 Introduction The new theory of the conservative congruence transformation (CCT) between the joint space and the R3x3 Cartesian space for stiness control was presented by Chen and Kao [1, 3, 4]. The CCT relates the mapping between the stiness matrices of the joint and Cartesian spaces, and can be expressed by the following equation J T Kp J = K Kg (1) where J is the Jacobian matrix, Kp and K are the Cartesian and joint stiness matrices, respectively, and Kg is the eective stiness matrix due to the change in geometry of grasping and manipulation under the presence of external force. In their paper [1], the authors showed that the conventional formulation for the mapping between the Cartesian and joint stiness matrices, given by the following equation [6] J Kp J = K Research Assistant; Email: [email protected] y Associate Professor of Mechanical Engineering; T is in general an incorrect mapping that results in physical inconsistency in conservative properties of stiness control, even though equation (2) has generally been accepted and widely used. Chen and Kao [1] showed that equation (2) is only valid when robotic manipulators are at their unloaded equilibrium conguration, i.e., without external force. In this paper, we will prove theoretically that the CCT can preserve the conservative properties of stiness matrices between the joint and Cartesian spaces, while the conventional formulation can only preserve the conservative properties under special condition. In addition, we extend the CCT work in [1, 3, 2] and the analysis of conservative properties of stiness matrix in [5] to further illustrate, with mathematical formulation, that the conventional formulation in equation (2) is a non-conservative mapping. In addition, we provide a theoretical proof to show that the CCT is the correct and general relationship for congruence mapping with stiness control in robotics. The simulation of a three-link, redundant planar manipulator is used to illustrate the CCT as the correct transformation. 2 The conventional mapping of stiness matrices between the joint space and the R33 Cartesian space is given in equation (2). In this section, we will show with mathematical formulation that the conventional mapping proposed by Salisbury is a nonconservative mapping. There are two typical stiness control schemes by which we can show this conclusion: (i) the Cartesian-based stiness control, and (ii) the joint-based stiness control. (i) the Cartesian-based stiness control: solv- ing for the joint stiness matrix K with specied Cartesian stiness matrix Kp , via equation (2) For the Cartesian-based strategy, K is shown to be generally non-conservative, even with a conservative Kp . (2) Email: (ii) the joint-based stiness control: solving for the Cartesian stiness matrix Kp with specied [email protected] 0-7803-6475-9/01/$10.00 © 2001 IEEE The Conventional Formulation 3937 joint stiness matrix K , via an inverted relationship given by equation (2) For the scheme of joint-based stiness control, we will show in the following that Kp is generally non-conservative even when K is conservative. In other words, the conventional formulation can not preserve the conservative properties between the joint stiness matrix K and the Cartesian stiness matrix Kp , even though the symmetry of stiness matrix is preserved under equation (2). We will rst focus on the joint-based control scheme in the following. Jij 1 @i . Substituting these terms into equation = @x j (8), we have @K @K = l=1 m=1 n = @Kik @xj 1 i; j 3 (3) 1 i; j; k 3 (4) where xi denotes the coordinates. Let Jij 1 denotes the (i; j )th element of J 1 , K;ij denotes the (i; j )th element of K , and Jij denotes the (i; j )th element of J . Then the Cartesian stiness matrix Kp based on Salisbury's congruence formulation can be written as Kp = J T K J 1 Kp;ij = n X n X l=1 m=1 Jmi 1 K;ml Jlj 1 Kp;ik = l=1 m=1 Jmi 1 K;ml Jlk 1 (7) p;ik 6= @K ; 1 i; j; k 3 @x j (8) Following the denition of J , the (i; j )th element of th element of J 1 is i J is Jij = @x @j , and the (i; j ) @ m K;ml l @xi @xj = @ 2 m @l K @xk @xi ;ml @xj @ @xj = {z | + @l m K;ml @xi @xk @ @ @xj @m @K;ml @l @xi @xk @xj } | @ term1 @ {z @l m K;ml @xi @xk + @l m K @xj @xi ;ml @xk | {z term1 + {z } term3 (10) @m @K;ml @l @xi @xj @xk } | 0 o (9) m l K @xi ;ml @xk @2 @m @ 2 l K @xi ;ml @xk @xj } | term2 {z + @2 @m l K @xi ;ml @xj @xk } | 0 term2 {z } 0 term3 It is obvious that the third term (term3) in equation (9) is equal to (term3') in equation (10) due to commutative property. Next, we show that the second terms (term2) and (term2') in equations (9) and (10) are equal to each other. First, the term2 in equation (9) is ! n @m X @K;ml @p @l @m @K;ml @l @xi @xk = @xj = @xi @p p=1 @xl @xj n X @m @K;ml @p @xi p=1 @l @xk @xj @p (11) Similarly, term20 in equation (10) can be written as ! n @m @K;ml @l @m X @K;ml @p @l @xi @xj = @xk = @xi @p p=1 @xj @xi @p @l @xj @xk From this we have n X n n X n X n X X @m @K;ml 0 (term2 ) = l=1 m=1 @xk n X @m @K;ml @p p=1 In order to prove that the conventional formulation is non-conservative, we need to show that @Kp;ij @xk l=1 m=1 @ @ @ @xj m l K @xi ;ml @xj @xk (6) The (i; k )th element of Kp is n X n X @ @xk n X n X p;ik @xj Next, we calculate the two terms in the braces: @ @ @ (5) If Kp is non-conservative, it will violate either one of the two conservative criteria in equations (3) and (4). Since equation (2) preserves symmetry, Kp will also be symmetric as long as K is symmetric. What is left is to examine the exactness property of Kp . The (i; j )th element of Kp is @ m K;ml l @xi @xj l=1 m=1 According to the two criteria for a conservative matrix in [5], a stiness matrix K is conservative if it satises the following two criteria: = Kji @ @ @xk n n XX = 2.1 Joint-based stiness control using the conventional formulation Kij @Kij @xk n X n X p;ij @xk l=1 m=1 p=1 @xi @p @l @p @xk @xj (12) (13) When K is conservative, we require that @K;mp @l So we have n X n X (term2) l=1 m=1 3938 = = @K;ml @p n X n X n X @m @K;ml @p l=1 m=1 p=1 @xi @p @l @xk @xj = n X n X n X @m @K;mp @p @xi l=1 m=1 p=1 = n X n X n X @m @K;ml @xi p=1 m=1 l=1 = n X n X @l @xk @xj @l @p @l @p @xk @xj 0 (term2 ) = K d Kg d = (dJ T )f d (14) l=1 m=1 In the derivation of equation (14), exchange of the dummy indices is used. It is clear from equations (13) and (14) that the sums of term2 and term20 are the same. Next, let us examine the rst terms. From the denition and properties of the Jacobian matrix, we nd that in general 2 2 @ m @l @l 6= @ m (15) @xk @xi @xj @xj @xi known relationships dened in the joint space as follows, Equation (19) was derived in [1]. Our task is to show that the formulation of CCT, J T Kp J = (K Kg ), always yields the consistent Kp in the Cartesian space as dened by Kp = ddxf . Since = J T f , we can dierentiate this equation to render d 6= (16) Hence, the rst term in equation (9) is not equal to the rst term in equation (10), in general. Thus, n X n n X n X X @l @ 2 m @l @ 2 m l=1 m=1 @xk @xi K;ml @xj 6= l=1 m=1 @xj @xi K;ml @xk 2.2 Discussions Based on the above derivation, we can identify special cases for which Kp is conservative. One such special case is when the Jacobian matrix J satises the equality in equation (17), i:e:, the Jacobian matrix is conguration-independent and constant, as described in [1]. We can show similar results using the Cartesianbased control scheme. That is, when Kp is conservative, we can prove that K is in general nonconservative unless a special condition similar to that in equation (17) is met. 3 = The Conservative Property of CCT The theory of the conservative congruence transformation (CCT) between joint space and the linear R33 Cartesian space is given by equation (1). Our objective is to show that equation (1) always preserves the conservative property of stiness control, and is consistent with the denition of stiness, namely, K = ddxf , where f represents the bias force (external force) acting on the end-eector and dx denotes the displacement of the end-eector. Let us proceed with the df df dx d = J T d d dx d df J T J d dx (21) Substituting equation (18), (19) and (21) into equation (20), we have K d = Kg d + J T df df = Kg d + J T J d dx (17) Therefore, Kp is non-conservative, even with conservative K , using the conventional mapping. (20) J T df = J T @xk @ 2 m @l K;ml @xj @xi @xk = (dJ T )f + J T (df ) The second term in equation (20) can be rewritten in alternate mathematical expression as This results in @ 2 m @l K;ml @xk @xi @xj (18) (19) (22) Thus, we can write K Kg = J T df J dx (23) Comparing equation (23) with the denition of CCT in equation (1), it is clear that Kp = df dx (24) Therefore, with the combination of both K and Kg , the Cartesian stiness matrix is shown to be the standard denition. Thus, the Kp obtained from the CCT is consistent with the fundamental property of a conservative system, and that the CCT establishes equivalent stiness matrices in both the joint space and Cartesian space. 4 Simulation of a Three-link Planar Manipulator In this section, we will consider a three-link redundant planar manipulator to illustrate the conservative property of the stiness control, using the Cartesianbased stiness control strategy via the CCT in equation (1). The incorrect results of the conventional formulation in equation (2) will also be computed and compared with the results of the CCT equation. 3939 Parameters: L1= 0.29 m L2 = 0.23 m L3 = 0.05 m L3 trajectory 2.5 fx fy 2 P(x,y) L2 Initial states: o 1= 21.8 Y 2 L1 3= 60.9o f 0 = [20 x0 = [0.45 X O = 39.1o 25]T N 0.33]T m Work done over circular path (Joules) f= Figure 1: A three-link redundant planar manipulator for simulation. For a two-link planar manipulator, the orientation of the end-eector is dened by the orientation of the second link, and can not be controlled; whereas, we can control the orientation of the end-eector of a three-link planar manipulator through the extra degree of freedom of the redundant joint. Thus, this three-link manipulator is called a \redundant manipulator" if we are concerned with only the linear fx; y g coordinates with three joint parameters, f1 ; 2 ; 3 g. The geometry of the redundant manipulator is shown in Figure 1, where three serial links are congured with the reference point at the end-eector at P (x; y ), and the robot base frame is at O-XY . We denote the position of the end-eector as x = [x; y ]T in the Cartesian space and = [1 ; 2 ; ; 3 ]T are the joint parameters. From Figure 1, the Cartesian position of the end-eector can be written as = L1 c1 + L2 c12 + L3 c123 y = L1 s1 + L2 s12 + L3 s123 x J22 1 0.5 0 work in Cartesian space -0.5 J23 where J11 = (L1 s1 + L2 s12 + L3 s123 ), J12 = (L2 s12 + L3 s123 ), J13 = L3 s123 , J21 = L1 c1 + L2 c12 + L3 c123 , J22 = L2 c12 + L3 c123 , and J23 = L3 c123 . 4.1 Cartesian-based stiness control The initial orientation of the third link is maintained in horizontal orientation, as shown in Figure 1. After that, the orientation of the third link is specied to align with the radial direction of the trajectory, as shown in the shaded third link in Figure 1, so as to align the end-eector with the orientation of the grasp object for grasping tasks and stable prehension. This work in joint space -1 -1.5 -2 0 100 200 300 360 degrees Figure 2: Results of simulation, using the CCT, for the Cartesian-based stiness control. gives rise to the advantage of the three-link redundant manipulator over the two-link manipulator. Since we are considering the Cartesian-based stiness control scheme, the stiness matrix, Kp , and the bias force, f , in Cartesian space should be specied rst. In this simulation, the prescribed conservative Cartesian stiness matrix and the bias force are 300 0 Kp = 0 250 N=m f = [20 25]T N The matrix Kp satises the two conservative criteria in equations (3) and (4). The initial joint torque based on f is calculated to be = [ 17:25 8:66 1:25]T N m. The stiness matrix Kg dened in equation (1) is 2 3 Kg = 4 (25) (26) where c1 = cos 1 , s1 = sin 1 , c12 = cos (1 + 2 ), etc. Thus, the Jacobian matrix becomes J11 J12 J13 J = (27) J21 1.5 Kg;11 Kg;21 Kg;31 Kg;12 Kg;22 Kg;32 Kg;13 Kg;23 Kg;33 5 (28) where Kg;11 = (L1 c1 + L2 c12 + L3 c123 )fx (L1 s1 + L2 s12 + L3 s123 )fy , Kg;12 = (L2 c12 + L3 c123 )fx (L2 s12 + L3 s123 )fy , Kg;13 = L3 c123 fx L3 s123 fy , Kg;21 = (L2 c12 + L3 c123 )fx (L2 s12 + L3 s123 )fy , Kg;22 = (L2 c12 + L3 c123 )fx (L2 s12 + L3 s123 )fy , Kg;23 = (L2 c12 + L3 c123 )fx (L2 s12 + L3 s123 )fy , Kg;31 = L3 c123 fx L3 s123 fy , Kg;32 = L3 c123 fx L3 s123 fy , and Kg;33 = L3 c123 fx L3 s123 fy . The trajectory of the end-eector is a circle with a radius of R = 0:08m, centered at (x; y ) = (0:37; 0:33)m. The point P of the end-eector moves along the circular trajectory in counterclockwise sense, starting from the initial position at (x0 ; y0 ) = (0:45; 0:33)m and nally returning to the same initial position. The force and net work done in the Cartesian space are calculated using the following equations for numerical integration fi = fi 1 + Kp dx Wc;i = Wc;i 1 + f T dx 3940 (29) (30) Work done over circular path (Joules) 14 120 12 100 Work done over circular path (Joules) 10 work in joint space 8 6 4 80 60 40 20 2 0 0 -2 work in Cartesian space −20 0 100 200 300 360 degrees 0 50 100 150 200 degree 250 300 350 400 Figure 3: Results of simulation, using the conventional mapping, for the Cartesian-based stiness control. Figure 4: Results of simulation, using the CCT, for the joint-based stiness control. After Kg matrix from equation (28) and K matrix from equation (1) are computed, the torque and work done in joint space can be calculated using the following equations of numerical integration in Cartesian space and in joint space are not identical. The discrepancy reveals the inconsistency of the conventional mapping. i Wc;i = i 1 + K d = Wc;i 1 + T d (31) (32) In equations (29) to (32), the second-order and higher terms are neglected since they have little contribution in numerical integration. Furthermore, the integrations of both the net work and force over the circular path should be zero if the system is conservative. The results of numerical simulation are shown in Figures 2 and 3. 4.2 Discussions From Figures 2 and 3, the following observations are in order: When the CCT is employed, the net works done in both the Cartesian and joint spaces of this redundant manipulator are identical. Since we are considering a conservative system (Kp is conservative), so we expect that the work calculated in both Cartesian space and joint space should be the same. Form Figure 2, we can see that the net work done in both Cartesian space and joint space are equal to zero when returned to the starting point after a full circle. This indicates that energy is conservative when CCT is used. When the incorrect conventional formulation is used, the net work done in the joint space in Figure 3 is not zero, although the net work done in the Cartesian space remains zero. That indicates that energy is not conservative using the conventional formulation. Furthermore, the work done The two work proles of the conservative stiness control should be identical to each other, regardless of which coordinate system the work is integrated with respect to. Therefore, only the CCT should be used. The work prole depends on the initial bias force and the Kp matrix. However, the conservative characteristics of the system does not depend on the initial bias force, nor the trajectory as long as it is a continuous closed path. It is important to note that equation (1) used in this simulation does not involve the generalized inverse of J { an important subject to be discussed in Section 4.3. 4.3 Joint-based stiness control The geometry of this simulation is similar to that in Section 4.1. The initial orientation of the third link is also maintained in horizontal orientation. After that, the orientation of the third link is specied to align with the radial direction of the trajectory for the upper half circle (0 ! ). When the end-eector moves to the lower half of the circle, the orientation of the third link changes from ! 0. With the joint-based control scheme, the joint stiness matrix K and the joint torque are specied rst: " 300 0 0 # K = 0 250 0 N m = [10 5 5]T N m 0 0 5 The matrix K is conservative. The joint stiness of link three is kept low (K;33 = 5) due to its larger 3941 matrix is currently being explored. 200 5 Work done over circular path (Joules) 150 100 50 0 −50 0 50 100 150 200 degree 250 300 350 400 Figure 5: Results of simulation, using the conventional mapping, for the joint-based stiness control. movement than the other two links. The initial bias force is calculated to be f = [45:1 53:1]T N . The Kg matrix is calculated using equation (28). The center of the trajectory circle is at (x; y ) = (0:30; 0:30)m, with a radius of 0:15m. The initial position of the end-eector is at (x0 ; y0 ) = (0:45; 0:30)m. The algorithm to calculate the work is the same as that in the previous sections. The simulation results are plotted in Figure 4 (using the CCT) and Figure 5 (using conventional formulation). Unlike the Cartesian-based control scheme, the joint-based control strategy in such redundant manipulator involves the Penrose-Moore generalized inverse of the non-square Jacobian matrix in order to obtain the Cartesian stiness matrix, i.e., Kp = (JT ) (K Kg )(J ) where `*' denotes the generalized inverse. Consequently, the inverse CCT equation will utilize the generalized inverse that masks part of the Kp components that belongs to the null space of the generalized inverse of the Jacobian matrix. Due to the generalized inverse of the non-square Jacobian matrix, part of the information for stiness control is lost; thus, from Figure 4 (CCT is used) we can see that there is a discrepancy between the work done in the Cartesian space and joint space. But the tendency of these two curves are similar, and the net work done in the joint space is zero and is almost zero in the Cartesian space. In Figure 5 (conventional mapping), it is obvious that even though the work done in joint space returns to zero, as expected, there is a large dierence between these two curves and the net work done in Cartesian space is far from zero. Simulation and derivation of a methodology to recover the lost information for stiness control due to the application of generalized inverse of the Jacobian Conclusions In this paper, we prove theoretically that the conventional formulation of stiness mapping between the Cartesian and joint spaces is generally incorrect, only valid under special conditions. The conservative nature of the CCT is proven theoretically, and illustrated with simulation results of a redundant planar manipulator. The results suggest that the stiness matrix obtained through the CCT is consistent with the fundamental property of a conservative system. Results of numerical simulation for the work done over a circular path using both equations (1) and (2) are presented and compared. The detrimental eect in the application of the generalized inverse of the Jacobian matrix for redundant manipulator is reported. Future work is needed to resolve the issue of using the generalized inverse, in order to deal with the inevitable loss of part of stiness control information due to the masking of the null space of generalized inverse of the Jacobian. 6 Acknowledgment This research has been supported by the National Science Foundation under grant number IIS9906890. References [1] S.-F. Chen and I. Kao. Conservative congruence transformation for joint and cartesian stiness matrices of robotic hands and ngers. the Int'l J. of Robotics Research, 19(9):835{847, September 2000. [2] S.-F. Chen and I. Kao. Geometrical methods for modeling the asymmetric 66 cartesian stiness matrix. In Proc. of IROS'00, pp. 1217-1222, Japan, Nov. 2000. [3] S.-F. Chen and I. Kao. Simulation of conservative congruence transformation: Conservative properties in the joint and cartesian spaces. In IEEE Int'l Conf. on Robotics and Automation, pp 1283{1288, April 2000. [4] Shih-Feng Chen. Fundamental Properties Of Stiness Control In Grasping And Dextrous Manipulation In Robotics. PhD thesis, State University of New York at Stony Brook, May 2000. [5] I. Kao and C. Ngo. Properties of grasp stiness matrix and conservative control strategy. the Int'l J. of Robotics Research, 18(2):159{167, Febrary 1999. [6] J. K. Salisbury. Active stiness control of a manipulator in cartesian coordinates. In Proc. 19th IEEE Conf. on Decision and Control, pp 87{97, NM, Dec. 1980. 3942
© Copyright 2026 Paperzz