You are on page 1of 16
INT, J, CONTROL, 1988, VOL. 47, NO. 6, 1697-1712 Globally asymptotically stable ‘PD +’ controller for robot manipulators BRAD PADENY} and RAVI PANJAt We describe a giobally stable tracking controller for robot manipulators. The controller is an extension of Takegaki and Arimoto’s position controller to the tracking case where a theorem of Matrosov is used to prove its stability. An attractive feature of this coniroller is its resemblance to the computed torque controller with the inertia matrix outside the position and velocity feedback loops. Thus, our controller is decomposed into an inner PD loop and an outer dynamic compensation loop. This structure allows the simple PD computations to be run at a higher speed than the dynamic compensation loop in digital implementations. 1. Introduction In this paper we present a simple globally stable tracking control scheme for robot manipulators. This scheme is a tracking controller which, from a block diagram point of view, is a variation of the popular computed torque scheme. The advantage of the scheme is that the manipulator inertia matrix is outside the position and velocity feedback loops. This allows the decomposition of the control scheme into decoupled PD joint control plus a higher level dynamics compensation loop. We refer to this controller as a ‘PD +’ controller. The higher level loop feeds forward nominal joint torques and compensates for gravity and forces which are Coriolis-like (see Fig. 1). Figure 1. ‘PD-+’ control scheme. Received 23 June 1987. + Department of Mechanical Engineering and the NSF Center for Robotics in Microelec- tronics, University of California, Santa Barbara, CA 93106, U.S.A. 1698 B. Paden and R. Panja The advantage of this structure is that the PD controllers can be run much faster in digital implementations. (It is generally believed that the position and velocity feedback loops must be updated at a higher rate than the dynamic compensation terms in digital robot controllers. This is motivated by the fact that the feedback terms have high gains.) To provide the background for our work, we review some of the presently available control schemes. The computed torque controller is the globally stable controller which is easiest to analyse as we can use simple linear theory. Robustness of this scheme in the presence of bounded disturbance torques, measurement noise and parameterized variations in the manipulator dynamics has also been shown by Spong and Vidyasagar (1985). Other schemes with stability proofs are Lyapunov designs. There are several of these. Variable structure controllers have been developed by Slotine (1985), Slotine and Sastry (1983), and Paden and Sastry (1987). Robust controllers which are similar to the variable structure controllers, but not having discontinuous control laws, have been proposed by Corless and Leitmann (1984) and Slotine (1985). Robust controllers typically have somewhat complicated stability proofs, but the controllers themselves tend to be rather simple and have compu- tational advantages when the computation of manipulator dynamics is expensive. A position control scheme, which is beautifully simple, is due to Takegaki and Arimoto (1981). This scheme is designed by deriving a Lyapunov function from the mechanical energy of the manipulator. It appears that this scheme had not been extended to the tracking control problem due to technical problems in proving stability. In the Takegaki and Arimoto scheme, stability is proved using LaSalie’s theorem for autonomous systems. However, when the desired position of the manipulator is time-varying, this theorem is no longer applicable, thus the technical difficulty. Recently, Slotine and Li (1987) cleverly overcame this difficulty by using some ideas from variable structure control. They introduced a complete adaptive control law which has a tracking control version as well. This scheme has the inertia matrix inside the ‘D’ part of the PD feedback loop (see Fig. 2). This is not necessary in quast-cortouis and \g GRAVITY TERMS Figure 2. Slotine and Li’s adaptive scheme. Globally stable controller for robot manipulators 1699 the tracking case, as we show here, and the resulting controller can be designed to have independent PD joint control with a high level outer control loop for dynamic compensation. Recently, Bayard and Wen (1987) (see also Wen and Bayard 1987) have studied both the adaptive and tracking control of robot manipulators. Their work is exceptional in that they propose some control schemes which are exponenti- ally stable. Their analysis technique differs from ours and that of Slotine and Li. We use a theorem of Matrosov to prove stability, whereas Slotine and Li integrate some ideas from variable structure control and Bayard and Wen use some technical lemmas and carefully chosen Lyapunov functions, These are all Lyapunov-based approaches, but differ in interesting ways. In this paper we develop our controller as follows. In § 2 we introduce our notation and derive the manipulator dynamics, paying careful attention to the structure of the fictitious forces derived from the inertia tensor. In §3 we re-derive Takegaki and Arimoto’s controller and in §4 we develop our controller. We demonstrate the stability of the control scheme in § 5 using a simulation of a two-link manipulator. Section 6 contains our conclusions, and the complete stability proof based on a theorem of Matrosov is contained in the Appendix. 2. Manipulator dynamics In this section we derive the dynamics of an m-joint rigid-link robot manipulator using tensor notation. This general construction is important as it preserves the structure of the dynamics very well; the fictitious forces are described explicitiy in terms of the inertia tensor and their multilinearity is clear. We use the following notation. real numbers R. non-negative real numbers C’ continuously differentiable n times f'%r)_ nth time derivative of f(r) d(x,Q) distance from the point x to the set Q [M]_ the matrix whose components are the same as the covariant tensor M of order 2 (in other words, #7[M]f # M(a, f)) D, derivative with respect to x Aj; the alternation mapping which skew-symmetrizes a tensor in its ith and jth arguments F(-)_ the map F(-):a > F(a) B, open ball of radius 2 centred at the origin 5 closure of the set § Let 0¢R* be the vector of manipulator joint displacements and denote the configuration-dependent inertia tensor of the robot manipulator by M(8). We make the Assumption 1 that M(0) is symmetric, positive definite, twice continuously differentiable, and its maximum and minimum singular values are bounded and bounded away from zero, respectively. Denote the components of M(@) by M(0),,- That is, M(8)(a, B) = M(0) ;2'B! (2.1) where a are the components of #€ R", and the summation over i and j on the right- hand side is implied. The derivative of M(8) with respect to 0 is a covariant tensor of 1700 B. Paden and R. Panja order 3 with components éM(O) DoMOiy = Mu (22) We define DyM (O(a, B73 & DoM Ou (23) and ; Dg M(6)(2, B)ON3) 4 > (2.4) Thus, D,M(0) is linear in each of the three arguments a, f and », as is D3M(0) in its four, In addition, let G(0) denote the potential energy due to gravity for the configuration 0. We make the Assumption 2 that G(@) is continuously differentiable. The lagrangian for a manipulator is therefore L=4M(8)(6,6) — G(6) (2.5) The joint forces are given by d P=, Dol — Dol (2.6) In this formalism F is a linear operator (covariant tensor of order 1) which acts on a displacement to do work. From (2.5) and (2.6) we have FC) =F MO(0,>) —4DgM(O}G, DK + DIG) 2 = M(6)(6, -) + DoM(0)(6,)(8) — 4 DpM(0)(6, 8)(-) + DpG(O)() = M(6)(9,) + 4 DpM(0)(6, )(8) — A23DyM(A)(0, 8)(-) + DeG(B)(-) (2.8) where Az3DyM(8)( 4% B)(7) © $[DaM(O)(x, B) (7) — DoM(@\(a, »)(B)] (29) 0G and D,G(0)(«) & R; three functions a, b, c of class K such that, for every (x, t) €[to, 00) x Q, (CL) a(x) < V(t, x) < BUI) Globally stable controller for robot manipulators 1709 (2) Vit, x) < V*(x) <0. Define E 4 {x ¢ Q| V*(x) =0}. (3) |W(t, x)| is bounded (4) max (d(x, £), |H(t, x)1) > e( Ix) (5) f(t, x)) is bounded. Choosing « > 0 such that B, < Q, define for all t € [t), 20) Vial = {x €Q: V(t, x) k(||x||)¥x © E and t2lo. Proof Since f(t) K, and W is continuous in f for all t>t) we have that ¥(x) & min |g(x, f)| exists and is continuous. Now consider BeKy e(a)£ min max(d(y, E),|¥(m))) (Al) neQ ||nl| 2a which is defined for all « = ||x|, x in @. The set over which the minimization occurs is decreasing with increasing , so c,(2) is nondecreasing with respect to #. Observe that 1710 B. Paden and R. Panja c,(a) > 0 and that c,(x) = 0 implies that the minimum in (A 1) is attained for 7 = 0 by condition 4’(b). Thus, in addition to being non-decreasing, c,(0) = 0, and c,(p} > 0 for all p> 0, To show condition (4) define the strictly increasing function ¢(-) by cin) [ clo) exo —(0 ok de (A2) o It follows from the properties of c, that c(x) eC Ix) (A3) s) This lemma and Theorem 1 are now applied to show stability. Theorem 2 Suppose that Assumptions 1-5 are satisfied. Then, for the dynamical system described by (2.8) and (4.4), (0, 0) is a globally asymptotically stable equilibrium point of the tracking error x£[0-0, 6—6,]", Proof We simply check the conditions of the theorem for V = 1/2M(0)(6—6,, 6 —6,) +1/2K, (6 — 04, 0 —0,) and X= f(x, t) defined implicitly by (4.6), where we are taking xtobe[0—6, 6—6,]7. Also, to show that the stability is global, we must show that we can choose « so that V;,, contains an arbitrary initial tracking error. (1) Clearly, by the assumptions (1 and 3) on M(@) and K,,, condition (1) is satisfied for (1, x) €[to, ©) x R2", In fact, we can choose a(-) such that a(p) + 00 as p+ 00. (Note that Q can be chosen arbitrarily large here. It will be restricted when condition (5) is checked.) (2) From (4.5), V= —K,(6 — 6,, 6 — 9,) <0 and depends only on x so condition, (2) holds for (1, x) € [to, 00) x R2™, Before we proceed, we establish the boundedness and continuity of several functions, Conditions (1) and (2) of Matrosov’s theorem are satisfied and therefore 0-6, and 6-6, are bounded. Moreover, 6, and 6, are bounded and so @ and 6 are bounded as well. Since M(@) is twice continuously differentiable and @ is bounded, the tensors D,M(8) and D2M(8) are bounded. Furthermore, 6 — 6, is continuous in the tracking error and depends continuously on time through @ and 6 which are bounded (this is important for verifying 4’(a)). With these bounds in mind, it is easy to see that the RHS of (4.6) is bounded. Recall that the minimum singular value of M(6) is bounded away from zero so we have (0— j,) is bounded. Finally, observe that the solution to the differential equation obtained by combining (2.8) and (4.4) is continuously differentiable and so 6, 6 and 0 are continuous. We proceed with checking condition (3). (3) Define W(x, 1) # V(x, 1). We have from (4.5) that V=2K\(6—6,,0-O)VxeEE (A4) Globally stable controller for robot manipulators 1711 It follows from the bounds above that this is bounded. (4) By the lemma it suffices to verify 4'(a) and 4'(b). Computing W = V we have W= 2K (0-6), 8-0) + 2x,(0-0,.40-09) (A5) dy x All arguments, except for 5-(0'—0,), in (A 5) are known (at this point in the proof) to be continuous in the tracking error and depend continuously on time through a bounded function. (We do not write 04? or 6) since they may not exist.) To establish aA 4((a) we show that 7-(0—0,) is continuous with respect to the tracking error and continuous with respect to time through a bounded function. Differentiating (4.6) with respect to time we have mon S08 a> 3} + DyM(0)(0— 9, (0) =~ K,(6— 64) —K,(G~6,,) 1 sg ae ~ FD3M(0) 0 ~ By, 9(8)(8) — 5 Do MCT — 6,10) 7 " ~ 5PoM(0)(0 ~ 64, )(0) + Aa3DoM(0)(0, 6 —O)(-) + AzsDyM(6)(6, 0 —0,)(-) + Az3D5M(6)(0, 6-0) (A 6) All terms in the RHS of (A 6) are continuous with respect to the tracking error and depend on time continuously through a bounded function. In addition, the minimum 7 d a singular value of M(8) is bounded away from zero. Hence, +- (0 — 0) is continuous in the tracking error and depends on time continuously through a bounded function and 4(a) is established for 9 = R?". Condition 4'(b) follows from (4.7) and the fact that the maximum singular value of M(8) is bounded. (5) Since 6(t) and 6(t) are bounded, and the RHS of (4.6) is continuous in these variables and the trajectory error, the RHS of (4.6) is bounded for all (x, 1) €Q x R, when Q is bounded. Moreover, [M(6) ] is bounded and so || f(t, x) |, defined implicitly by (4.6), is bounded on [0, oc) x © when Q is chosen bounded and the condition is satisfied. Thus the conditions of the theorem are satisfied. Moreover, for arbitrary initial conditions, we can use condition (1) to find an « and Q such that xo € V,,4. Thus, by Matrosov’s theorem, the origin is a globally asymptotically stable equilibrium point of the tracking error. Qo REFERENCES Bayakb, D. S., and Wen, J. T., 1987, Robust control for robotic manipulators. Part II: Adaptive case. JPL Engineering Memo, 347-817-204, Feb. 18, 1987. ConLess, M., and LeIrMaNn, G., 1984, Tracking in the presence of bounded uncertainties. Proc. 4th Int. Conf. on Control Theory, Cambridge University. Kopitscuek, D,, 1984, Natural motion for robot arms. Proc. 23rd Conf. Decision and Control. i712 Globally stable controller for robot manipulators Papen, B., and Sastry, S, S., 1987, [.E.E.E. Trans. Circuits Syst., 34, 73. Roucue, N., Hasers, P., and Laoy, M., 1977, Stability Theory by Liapunov's Direct Method (New York: Springer-Verlag) SLoTINE, J. J. E., 1985, Int. J. Robotics Res. 4, 49. Stominr, J. J. E., and Li, W., 1987, Int. J. Robotics Res., 6, 3. Stoting, J. J., and Sastry, S. S., 1983, Int. J. Control, 38, 465. SponG, M. W., and Vipyasacar, M., 1985, Robust nonlinear control of robot manipulators. Proc. 24th Conf. on Decision and Control, Ft. Lauderdale. TAKEGAKI, M., and ARIMOTO, S., 1981, JI. dynam. Syst. Meas. Control., 102, 119. VipyasaGar, M., 1978, Nonlinear Systems Analysis (Englewood Cliffs, NJ: Prentice-Hall). Wen, J. T, and Bayar, D. S., 1987, Robust control for robotic manipulators, Part I: Non- adaptive case. JPL Engineering Memor. 347-87-203, Feb. 18, 1987.

You might also like