Control of Robot Manipulators F. L. Lewis University of Texas at Arlington C. T. Abdallah University of New Mexico D. M. Dawson ‘Clemson University 1993 MACMILLAN PUBLISHING COMPANY, New York MAXWELL MACMILLAN CANADA Toronto MAXWELL MACMILLAN INTERNATIONAL New York Oxford Singapore SydneyLed Robot Dynamics This chapter provides the background required for the study of robot manipulator con- trol. The arm dynamical equations are derived both in the second-order differential equation formulation and several state-variable formulations, Some important prop- erties ofthe dynamics are introduced. We show how to include the dynamics of the arm actuators, which may be electric or hydraulic motors 2.1 Introduction Robotics is a complex field involving many diverse disciplines, such as physics, properties of materials, statics and dynamics, electronics, control theory, vision, signal processing, computer programming, and manufactur ing. In this book our main interest is control of robot manipulators. The purpose of this chapter is to study the dynamical equations needed for the study of robot control. For those desiring a background in control theory, Chapter 1 is provided. For those desiring a background in the basics of robot manipulators, in Ap- pendix A we examine the geometric structure of robot manipulators, cover- ing basic manipulator configurations, kinematics, and inverse kinematics. ‘There we review as well the manipulator Jacobian, which is essential for control in Cartesian or workspace coordinates, where the desired trajecto- ries of the arm are usually specified to begin with. The robot dynamics are derived in Section 2.2. Lagrangian mechanics are ‘used in this derivation. In Section 2.3 we review some fundamental proper- ties of the arm dynamical equation that are essential in subsequent chapters for the derivation of robot control schemes. These are summarized in Table 2.3+1, which is referred to throughout the text. The arm dynamics in Section 2.2 are in the form of a second-order vector differential equation. In Section 2.4 we show several ways to convert this formulation to a state-variable description. The state-variable description is a first-order vector differential equation that is extremely useful for devel- oping many arm control schemes. Feedback linearization techniques and Hamiltonian mechanics are used in this section. The robot arm dynamics in Section 2.2 are given in joint-space coordi- nates, In Section 2.5 we show a very general approach to obtaining the arm dynamical description in any desired coordinates, including Cartesian or workspace coordinates and the coordinates of a camera frame or reference. 6364 2 / Robot Dynamics In Section 2.6 we analyze the electrical or hydraulic actuators that per- form the work required to move the links of a robot arm. It is shown how to incorporate dynamical models for the actuators into the arm dynamics to provide a complete dynamical description of the arm-plus-actuator system. ‘This finally leaves us in a position to move on to the next chapters, where robot manipulator control design is discussed. 2.2 Lagrange-Euler Dynamics For control design purposes, it is necessary to have a mathematical model that reveals the dynamical behavior of a system. Therefore, in this section we derive the dynamical equations of motion for a robot manipulator. Our approach is to derive the kinetic and potential energy of the manipulator and then use Lagrange’s equations of motion. In this section we ignore the dynamics of the electric or hydraulic motors that drive the robot arm; actuator dynamics is covered in Section 2.6. Force, Inertia, and Energy Let us review some basic concepts from physics that will enable us to bet- ter understand the arm dynamics [Marion 1965]. In this subsection we use boldface to denote vectors and normal type to denote their magnitudes. The centripetal force of a mass m orbiting a point at a radius o and angu- lar velocity w is given by me = me?r = mir 2.21) See Fig. 2.2-1. The linear velocity is given by v=oXn (2.2.2) fo FIGURE 2.2-1 Centripetal force.22 | Lagrange-Euler Dynamics 65 which in this case means simply that v = or Imagine a sphere (ie., the earth) rotating about its center with an angular velocity of @,. See Fig. 2.2-2. The Coriolis force on a body of mass m moving with velocity v on the surface of the sphere is given by F,,, = —2mo, Xv. (2.2-3) Using the right-handed screw rule (i.e., if the fingers rotate @, into ¥, the thumb points in the direction of m, X ¥), we see that, in the figure, the Coriolis force acts to deflect m to the right. Ina low-pressure weather system, the air mass moves toward the center of the low. The Coriolis force is responsible for deflecting the air mass to the right and so causing a counterclockwise circulation known as cyclonic flow. The result is the swirling motion in a hurricane. A brief examination of Fig. 2.2.2 reveals that in the southern hemisphere F.,, deflects a moving mass to the left, so that a low-pressure system would have a clockwise wind motion. Since @, = 6 and v = R@, we may write Fg, = ~2mB9R sin(90" + ) = -2mR6G cos 9. (2.2-4) FIGURE 2.2-2 Coriolis force.66 2 / Robot Dynamics Itis important to note that the centripetal force involves the square of a sin- gle angular velocity, while the Coriolis force involves the product of two dis- tinet angular velocities. The kinetic energy of a mass moving with a linear velocity of v is K = hmv, (2.2-5) The rotational kinetic energy of the mass in Fig. 2.2-1 is given by Kou = Halo, (2.2-6) where the moment of inertia is T= f p(n Pdr, 22-7) with p(r) the mass distribution at radius r in a volume. In the simple case shown where m is a point mass, this becomes T= mr. (2.2-8) Therefore, Ko = emre2, (2.2-9) The potential energy of a mass mt at a height A in a gravitational field with constant g is given by P= mgh. (2.2-10) ‘The origin, corresponding to zero potential energy, may be selected arbitrar- ily since only differences in potential energy are meaningful in terms of physical forces. The momentum of a mass m moving with velocity v is given by pam. (2.2-11) The angular momentum of a mass m with respect to an origin from which the mass has distance r is Pang = ¥X De (2.2-12) The torque or moment of a force F with respect to the same origin is defined to be N=rXE (2.2-13) Lagrange’s Equations of Motion Lagrange’s equation of motion for a conservative system are given by [Marion 1965] dab ab_ a ag (2.2-14)22 | Lagrange-Euler Dynamics 67 where qis an n-vector of generalized coordinates g,, tis an n-vector of gener- alized forces t,, and the Lagrangian is the difference between the kinetic and potential energies sa cmeeeecaeee (2.2-15) In our usage, q will be the joint-variable vector, consisting of joint angles 0, (in degrees or radians) and joint offsets d, (in meters). Then tis a vector that has components n, of torque (newton-meters) corresponding to the joint an- ales, and f of force (newtons) corresponding to the joint offsets. Note that we denote the scalar components of t by lowercase ietters. ‘We shall use Lagrange’s equation to derive the general robot arm dynam- ics. Let us first get a feel for what is going on by considering some examples. EXAMPLE 2.2-1; Dynamics of a Two-Link Polar Arm ‘The kinematics for a two-link planar revolute/prismatic (RP) arm are given in Ex- ample A.2-3. To determine its dynamics examine Fig. 2.2-3, where the joint- variable and joint-velocity vectors are fh 0 The corresponding generalized force vector is i ° FIGURE 2.2-3. Two-link planar RP arm.68 2 / Robot Dynamics with a torque and fa force. The torque n and force fmay be provided by either motors or hydraulic actuators. We discuss the dynamics of actuators in Section 2.6. To determine the arm dynamics, we must now compute the quantities re- ‘quired for the Lagrange equation. a, Kinetic and Potential Energy ‘The total kinetic energy due to the angular motion @ and the linear motion ris K = amr + ami? @) and the potential energy is P= mgrsin a b, Lagrange’s Equation The Lagrangian is L=K~ P= ‘amr + vamp? — mgr sin 0. (5) Now we obtain aL 7 [mi] © an| 7 | mi a A ab _ | mrb+2mri6 at a5 ~ [ mi? | ® aL | ~mer cos 0 ® aq ~ | mr@—mg sin 0 Therefore, (2.2-14) shows that the arm dynamical equations are mr + 2mrié + mgr cos 0= n 0 mm? ~ mr® = mg sin 0 = f. (10) This is a set of coupled nonlinear differential equations which describe the mo- tion 4(0) = [6(0) {OF given the control input torque n(¢) and force /(), We shall show how to determine q(¢) given the control inputs n(t) and f(t) by computer sim- ulation in Chapter 3. Given our discussion on forces and inertias it is easy to identify the terms in the dynamical equations. The first terms in each equation are acceleration terms involving masses and inertias. The second term in (9) is a Coriolis term, while the second term in (10) is a centripetal term. The third terms are gravity terms.22 | Lagrange-Euler Dynamics 69 Manipulator Dynamics By using vectors, the arm equations may be written in a convenient form. Indeed, note that [e" s]UT-me)-fseare)-G]- ee ‘We symbolize this vector equation as Maid + Via.d) + Gq) = (2) Note that, indeed, the inertia matrix M(q) is a function of q (i¢., of @ and r), the Coriolis/centripetal vector V{q,d) is a function of g and 4, and the gravity vector G(q) is a function of ¢. EXAMPLE 2.2-2: Dynamics of a Two-Link Planar Elbow Arm In Example A.2-2 are given the kinematics for a two-link planar RR arm. To de- termine its dynamics, examine Fig. 2.2-4, where we have assumed that the link masses are concentrated at the ends of the links. The joint variable is a= (8, 0)" ay and the generalized force vector is sin ult @ with 1, and , torques supplied by the actuators. FIGURE 2.2-4 Two-link planar RR arm.70 2 | Robot Dynamics a, Kinetic and Potential Energy For link 1 the kinetic and potential energies are K, = Ym,a38 P, = mga, sin 6, For link 2 we have x, = a; 0s 8, + a, cos (0, + 0) J; =a, sin @, + a, sin (, + 04) ¥, = -a,6, sin 6, — a,(6, + 6,) sin @, + 8,) Jig = 4,8, cos 8, + 2,(8, + 6,) cos (8, + 8), so that the velocity squared is veut iy a6; + 0X(6, + 8,7 + 2a,0,(67 + 8,8) cos @,, ‘Therefore, the kinetic energy for link 2 is K, = Yam = Yam aii + am,a3@, + 6)? + m,a,a,(62 + 6,0.) cos ©, ‘The potential energy for link 2 is P, = magy, = magia, sin 8, + a sin(®, + ®,)). b. Lagrange’s Equation ‘The Lagrangian for the entire arm is L=K-P=K,+K,-P,-P, = % (m, + m,) ai6? + tam,a} (6, + 6,7 + m,a,a, (6 + 6,6,) cos 8, = (m, + m,) ga, sin 8, ~ m,ga, sin (8, + 8). ‘The terms needed for (2.2-14) are = (m+ mn) a, + mgs @, + 0) + maa, 28, + 8) 08 8, ge = (om, + my) a8, + mga (B, + 0,) + maaya, (26, + 6,) 0s — m,a,a, (26,6, + 6) sin @, EE = ~ (mn, + mga, cos 6, ~ mga, c0s (0, + ©) ma G6, + 6,) + m,,4,8, cos 0, mya2 (8, + 8,) + mya,a,8, cos8, ~ m_,0,6,0, sin 0, —m,a,a, (& + 6,6,) sin 8, — mga, cos(@, + 0,). QB) (4) () © aM (8) 9) 9) ay 2)22 | Lagrange-Euler Dynamics 71 Finally, according to Lagrange’s equation, the arm dynamics are given by the ‘two coupled nonlinear differential equations 4, = [Ge + ma} + mad + 2m,a,a, c0s 0,) 6, + [m,a3 + m,a,2, cos @,) 6, — m,a,a, (26,8, + 62) sin 0, (13) + (m, + m,) ga, cos 8, + m,ga, cos (8, + 8,) t= [m,q} + maa, cos 6,) 6, + m,a3b, + ma,a,82 sin 0, 14) + m,ga, cos (8, + @,). ied ¢. Manipulator Dynamics ‘Writing the arm dynamics in vector form yields m, + mm) a} + m3 + 2m,aa, cos 0, mya} + m,a,a, cos 0, | [¥, ‘m,a} + m,a,2 008 8 ma 8, (is) 4. | rie (28,6, + 69 sin 8], [(m, + m1) ga, 0050, + mzg, cos @, + 8,) m,a,a,9; sin 0, myga, cos (8, + 0,) -E] *. These manipulator dynamics are in the standard form M@a + Vad) + GQ) = (16) with M(q) the inertia matrix, V(g,4) the Coriolis/centripetal vector, and G(q) the gravity vector. Note that 4(q) is symmetric. EXERCISE 2.2-3: Dynamics of a Three-Link Cylindrical Arm We study the kinematics of a three-link cylindrical arm in Example A.2-1. In Fig. 2.2.5 the joint variable vector is q=[0 kon qa Show that the manipulator dynamics are given by J+ mp 8) [amyr6 0) m+m, lial] 0. le lomampehl=le |, @ m,||#| | -m62 0 ih with J the inertia of the base link and the force vector tein fA 3)72 2 | Robot Dynamics FIGURE 2.2-5 Three-link cylindrical arm. Derivation of Manipulator Dynamics We have shown in several examples how to apply Lagrange’s equation to compute the dynamical equations of any given robot manipulator. In the ex- amples the dynamics we found always had the special form Mad + Vad) + GQ) = %, (2.2-16) with q the joint-variable vector and + the generalized force/torque vector. In this subsection we derive the dynamics for a general robot manipulator. They will be of this same form. To obtain the general robot arm dynamical equation, we determine the arm kinetic and potential energies, then the Lagrangian, and then substitute into Lagrange’s equation (2.2-14) to obtain the final result [Paul 1981; Lee et al. 1983; Asada and Slotine 1986; Spong and Vidyasagar 1989].