= Control of Manipulato, ut the specified task by moving its end-effegg, accurately and repeatedly. In the previous chapter, techniques to determing the set of joint-location time histories for a desired end-effector motion through space have’ been developed. In this chapter, the analysis and design of manipulator control system that accepts the joint-location time history as inpu, and causes the manipulator to track the commanded trajectory, is discussed. iis assumed that the reader has the basic knowledge of control systems. The control requires the knowledge of the mathematical model and some sr of intelligence to act on the model. Whereas the required mathematical modelis obtained form basic physical laws governing robot dynamics and associated devices, intelligence requires sensory capabilities and means for acting and reacting to sensed variables. A robot performs the specified tasks in is environment, which can be divided into two classes: contact type tasks and ena type tasks. The noncontact type tasks involve manipulation of the ein ee cae the desired work, while in the contact type a whichis not only able 7 the environment. In general, a perfect robo! ail applies on the avionmen Fe ee eas ecnirols tT ie nis manipulators are also classified rea eof the. type of task pet manipulator. The first half of th Scontact type manipulator and noncon™ ig this chapter is devoted to the control of nom?” and the cor : a iter arts “Ontrol of contact type manipulators involviN® A robot is required to carry 01 a i ators | are powered and driven by act “st apply a force or a ton qu ° Ton of the links. The actuator comm’ il '€ to cause mot; i santa derivatives can be fed back to in and/or end-effector positions ion le Control s rate Y stem to get acct!287; Control of Manipulators 287) ystem is shown in Fig. 8.1. The eram of a manipulator control fic the feedback indicate that the control system may or may not wines ’ .dback of the actual joint locations and velocities. The parameters g, g, joy fe A : j « : oon. in the figure, have the usual meaning discussed in previous chapters. j esired task son and orientation *yend-ffector) Actuator “7 commands i to, Manipulator | r maa Actual joiny | ,_”| Manipulator end-effector control system Vy pg eee velocity, etc. Control qa 4 set-points cl | Feedback = + ~-| (Sensor/observer) | Fig. 81. Block diagram of a manipulator control system me links and the end-effector from the Present location to the desi tae With oF without the Constraints of a particular planned end-effector deseo thor planned end-effector force/torque. In both situations of the ini end-effector location and the desired end-effector force, the control of Ach pint (oint > location is important. Hence, itis an obvious requirement that ‘Pint is controlled by a position servo. bee manipulator is to move very slowly or move one joint at a time, then the € cong amPle because coupled dynamic forces are negligible, Each joint can ing tolled independently by using a simple control system, which produces a Vaigb nO! forcehtorque proportional to the required change in the joint le. This is the Proportional control algorithm. If the motions are fast, which ions, all joints must move simultaneously. In are is Mt for effective robot appli situation, the coupled dynamic forces are significant and it is known that contttics is nonlinear and complex. Joints cannot move independently and "™Plex control algorithm will be required. The typical robot control architecture manipulator consists of a master control system to control and = a n-DOR canst hize joints, This master control is responsible for sending “set point fonds to each ofthe joint controllers. The njoint controllers use the setpoint Mor i troller mation to Command the joint actuator to move the joint. The joint con ma cally giv fee) MPIOY a Feedback of current join position. It may also perioticaly sire fon sbaek to the master controller, The schematic of typical control architect Ta manipulator control system is illustrated in Fig. 8.2.Joint? | | Manipulator 9; 4, “ conacllel joint 1 mode) + ~ ] a Joint 2 »| Manipulator | 92 ¢; cohol joint 2 model | ~*-—_. : a “! Jointn | +| Manipulator | % 4. | controller | joint n model Satine Be 1 Fig.82 Robot control architecture for an -DOF manipulator 8.1 OPEN- AND CLOSE-LOOP CONTROL FL Mya + DD dias +6, eo inn this model, accurate computation Of the parameters My yy. and Gi p aeull. Moreover, this model does nvr include effects of friction, backlash and Ses tthe joins and other extern disturbances or noise, These effects a to diffi ; | i pete fais highly nonlinear Because of all these problems. t® ing rOnprech nn has limited trajectory. tency a inven : ° -tracking capa and is use a few nonprecision applications, ‘¥-tracking capabilities and : i The limitations ic i ct : are overcome be Fe pettc inaccuracies and unmodeled nonlinear ef wn: : closed. sta joint rote 2sed-loop control scheme Here, at every inst’ any oe m 1ONS and velocities are measured directly by = 40 i tachometers Mounted at the joints, These are we 82 domme ae the desireg and actual positions and veloc et q ot force Feedback is use for error in the desired end" ques nek dasa finer rau, A control law then computes the joint ' cacif i onetes FY? €t10rs using the dynamic model al ope * i gravila Ing forces vary conti ith int! Coriolis, centrifugal, and grav ai ty Niro, eaautinuously with time, Th, dy ween i ‘ fi the dy” od Model is ¢ * COmputation of these from t cont 7 S complex and time Consuming Thi ion iy reat-time extremely difficult This makes the actualControl of Manipulators sail) \* a - Tot} pirect and | Trajector | 44 Control inverse | erator | r 1; tnt Homans | | ners jay mem Manipulator | |g tt Set ey ie ait onein i? 4 2 Fig. 83. Operations for a point-to-point motion control of a manipulator ‘The schematic description of the point-to-point motion control of an n-DOF inanipulator arm is shown in Fig. 8.3. For this operation, the initial and final ipctions of the end-effector in Cartesian space, specified by the transformation inatrices Tiniar 290 Tyngy, Serve as the input. The inverse kinematics model computes the desired end-effector locations in joint space. Then, a trajectory generator computes the joint-position time histories, based on the joint-space ‘gorithms discussed in Chapter 7. Depending on the servo error computed from the base reference values and the sensor measurements, the control system commands the individual actuators to achieve the desired motion. (ae. (power ll, i 2 coma » Controller |) ampfer [>| Actuator [>| Manipulator |} Sensors Use se LS LS | = Fig. 8.4 General block diagram for manipulator control system The block diagram in Fig. 8.4 illustrates the components of the control system ‘© control the manipulator joints, consisting of actuators and the mechanical kinematics chain, A typical control system has a comparator (a small computer); ® digital 10 analog converter (DAC); a controller to implement the control Stategy; an amplifier to boost and transform the small command signals so that it ,, m5 oe 0q Control of Manipulators 383) istics equation, ry, r +1» Py Which cha given input, 2 hich char of the eharact " acterize the system's : AE be ‘sponse of the system. If roots a i x s ryand f the system is sluggish and nonoscillatory. If r, (e, then the response of isoseillatory. The Ne, then the response ofthe systems oscillatory. The fh \ ps in fixing the controller parameters that ensure | oat ible trajectory tracking performance of a joint. | asset te mature OF response of a Second-order system, unit-step input i on nse 1002 sp ied. For a unit-step input, R(s) = I/s, the output of the system, from is given by : cL (Sei bow ¥(s) = G(s) R(s) = —~—Po@n | s(s* +260, + 02) @.10) | spiscan be rearranged as 2 ¥(¢) = Po 1) s(s +H )(s +) sere, rj and r are the roots of the characteristic equation, and are given by (89). Depending on the value of the damping ratio ¢, these roots can be real and vnequal or real and equal or complex conjugate. These three classes of roots characterize the three types of time response of the system. These three classes of roots arise with value of damping ratio greater than unity, equal to unity, and less than unity, respectively and are discussed below. Case i) Damping ratio greater than unity A damping ratio greater than unity (¢> 1) gives the roots r, and r, as real and tmequal. The roots r, and r, are = 0, + 0,10? =I = $0, - 0,10? -1 The time response is obtained by taking the inverse Laplace transform of 4. (8.12), which gives (ev 2 et [am ene WO =1- Sry [er Je + (=n +) | (8.13) itlatory behavior (time (8.12) ral this case, the system has a sluggish and nonose Na and system is overdamped. (i) Damping ratio equal to unitySe —, 1264) Robotics and Control For a damping ratio equal to unity (C1), both 7; and rare reat with ry, ="2=— Om and the time response of the system is and al yo= [Ie a! — Wy re" Big Under this condition, the system exhibits the fastest possible non overshoot-free response and is critically damped. Scillatory (iii) Damping ratio less than unity roots rand rare complex conjugates of each other when the ¢ le dat ratio ig less than unity (¢ < 1). The time response, y(1) is 7 fo, ¢ 2 ron sin. NTB . In this case, the system has an oscill i : scillatory behavior and is ‘under: Bie se bore a it is observed that the linea ond mre values of damping ratio, iy ie sc caer a ping ratio, although the behaviour may be oscilla The sys og seine) responses for unitate input Eas (813), (14 G5) 2 ig. 8.6 for these typical values of damping ae iet bing (8.15) wo) Yr : | Legend e tors 1] >> 1), itis quite reasonable to veking performer tt interacting term d and sill guarantee 4 ata tedhaenanee by the controller. Neglecting the nonlinear jor torque reduces to the simple form case of highly geared manipul contribution of the nontine t, Top Ou + Bay ,, (6.27) tive inertia and B. J re gis the mand Ba . effective damping with | eg SL + M/ 1? | Beg = By + Bl? (3:28) J pecause the configuration dependent term M is reduced by a factor 72, with poklapis almost constant for highly geared manipulators, iia be useful to summarize the assumptions made in obtaining the model sendy Bq. (8.27): © (a) The links are rigid bodies. (p) Link inertia is constant, (©) Nonlinear interacting terms such as the interactive and centrifugal torques and gravity torque are ignored. (@) Nonlinear effects such as coulomb friction and external disturbance are ignored. (e) Backlash is ignored. Thus, Eq. (8.27) is a SSO model of a rotary joint of a manipulator as a second- vat linear system with actuator torque 1, as the input and angular di ‘eat @, as the output. To construct a controller for an n-DOF manipulator, ritdependent S/SO linear control systems may be designed, lnthe above discussion, a rotary joint model has been considered. The model ‘‘nvational joint holds analogously for prismatic joint also. Before proceeding to “nine the design of controllers based on this approximate joint model, the ‘acteristics of actuators employed to drive the robotic links are discussed “Suse the actuator also influences the controller design. ‘6 JOINT ACTUATORS "actuators used in robotic systems to drive the links are broadly classified as: i : 1) Hydrautic actuators, e Peumatic actuators, and pele actuators ing tlic actuators can produce large force "Ss without i ne : h the use of reduction gearing ind require @ “pcOatol. But hydraulic systems are cumbersome and messy We NTE Of equipment such as pumps, actuators, hoses, and seve MT ons where position and/or torque must be accunee® Oa! He actuators prove disadvantageous due to friction “ity “WY of ol, and its complex temperature dependence. torque to drive the manipulator 'd are easily applied for robotic4 Control Jay Robotics am eas all the disadvantages or hydraulic Alay Paaaie aH oT cane Pcie attr a otro accurate senna cme) jee for actuators For robotics stems the elim The most popular ¢ mated the power-to-weight ratio 0) : Ny traulic orpneuna, ‘etuators, these are €2 ed manipulators: a vonerally require reduction gears of high ratio, alors «the system dynamics and reduces the coupling ete f the electric actuators but at the cost Of increase ratiity, and backlash. On the other hand, use of hydraulic vrs to directly drive the joint minimizes the drawbacks due jy h, Electric actuators are subclassified into foy small-to-medium sI The electric high-gear ratio linearizes ‘This is an added advantage of joint friction, el pneumatic actuators friction, elasticity, and backlas categories. (a) AC motor, (b) DC motor, (c) Stepper motor, and (a) Other devices such as solenoid. Amongst electric motors, DC motors are most straightforward to interface and control and, hence, are the most commonly used. AC motors and stepper motors are less utilized because AC motors are difficult to control and stepper motors have low torque capability. Moreover, stepper motors cannot be used for Cartesian space trajectory tracking operations (like plotting), The model of a typical DC motor is discussed in the following section keeping in view its applications in developing control strategies. Models for other types of motors and other electrical actuators can be easily found in other books. 8.6.1 Model of a DC Motor The schematic di shown in Fig. 8 iagram of an armature-contr : olled hae 8. Motors with field rovded by pe anent are preferred for robot DC (PMDC) motors, excitation provided by permanent magnets ic applications. Such motors are labelled permanent magnett, = Kyi, Aida (8.29) current directly induced in the where Kis the motor torque constant. Observe, that the armature controls the torque generated by the actuator. The back emf e, mature winding is given by ? Kieu (8.30) where Ki, is the back emf constant. Applying the Kirchoff’s voltage law for the uit of Fig. 8.8, gives ss 2, =e, + igRy (8.31) wheree, is the emf across armature and R, is the armature resistance. Substituting gqfrom Eq. (8.30) gives ea = KyB, +i,R, (8.32) Thus, the armature voltage is adjusted, depending on the commands generated by the manipulator control system, which specify the torque required from the actuator. The continuous adjustment of the armature voltage e, is carried out by the motor driver circuitry so that a desired current i, (corresponding to required torque f,) flows through the armature windings. The schematic of a motor driver circuitry is shown in Fig. 8.9. Joint control” 2, | Motor driver | system | Oesireqy | ofoulty TT \6 le ————— Fig. 8.9 Control of a manipulator joint driven by a DC motor 87 PARTITIONED PD CONTROL SCHEME With the development of linear second-order S/SO model of a manipulator joint and modelling of the actuator, it is now possible to study various linear control strategies for independent joint control. The first scheme discussed is a modified PD control. A linear controller based on a Partitioned Proportional Derivative (PPD) control strategy is slightly different from a simple PD controller. It is useful for Systems that are more complex. The controller is partitioned into two parts: a Model-based portion and a servo portion, such that the joint parameters (J,and rin this case) appear only in the model-based portion. The approximate model tthe manipulator joint developed in previous section is used to implement this ontroller. The block diagram of this controller is shown in Fig. 8.10. Note that bas prion to the right of the partition line is the physical system, the model- wale and to the left of the Partition line is the servo-based control system, 'y implemented in a computer or microcontroller.———— Bi) Ros ‘Thus, the re: and Control isa “discrete Staircase’ function ang tanning ofeach cycle. ThE controller will ra servo error is 1 ve guring, each cycle. Hence, the actual trajega this noua “Jose to, but not exactly the same as the desired trajectory, ‘ tracked will be close 10, Dit ry, another Factor that constrains the se Apart from a damping which are assumed to be rigid bo, f, e flexibil yf links i of control gains is the flexibility o! sumed (0 be rigid bodies in the development of the joint model. The unmodelled structural exibiy oy link and other mechanical elements produces resonance at frequencies other thay natural frequency, @,- Because these structural flexibilities have been ignored the controller must be designed so as not to excite these unmodeled resonances The lowest unmodelled resonance, which corresponds to the maximum Value of the effective inertia seen by the actuator, Jax has a resonance frequency I es = O Pa (8.44) where a is the structural frequency when the effective inertia is Jy. To prevent exciting these structural oscillations and also to ensure structural stability, the controller natural frequency @, must be limited to 0.5 @,,,. That is 0, $0.5 Oe I or @, $0.50 J (8.45) From Egs (8.41) and (8.45), siting command torque Fa yonzero at the rvo error t0 2 lection K, $0.5)" (8.46) Thus, the control gains K, and K, fe eee the PD controller are selected in 8.7.1 Effect of External Disturbance a ‘omain, given by Eq, (8.40), is modified as : ce KES) + K,E(3) = tag(s) ef ist i i Ae is bounded. Inthe ettPace input will be stable if and only if the disturbane words, for the system sable, Assume tt a Pounded disturbances, the controller will EP at the disturbance is as a step function of magni” Tas) = K 648) 8Control of Manipulators 808) sgervo-based —— Model-based - +| tas (8) oes |} —4 + Dy Fal) | s ian 1 | [et "| Tes? + Bors | 7 | | E(s) Jee yr + Fig. 8.12 PPD controller with external disturbance s becomes From Eqs (8.47) and (8.48), the error dynami (S4 Ky + K,)E(s) = K 5 K is) = ——_* ___ (8.49) # Ke) S(5° + K,s+K,) Thesteady-state error is defined as eq¢= lim| s E(s)] (8.50) 590 Substituting Eq, (8.49) in Eq, (8.50), and applying the limit gives the steady se error in actuator shaft position as Gyan (8.51) Ky Thus, the steady-state error is inversely proportional to the proportional gain 5, ofthe PD controller. Because one of the purposes of a controller is to provide ‘ordisturbance rejection, that is, to maintain good performance, K, is chosen at themaximum possible value that does not excite the structural oscillations. This stimum value of K,, which minimizes the steady-state error, is obtained from #4. (8.46) a - 46) as I K puae = (0.5409)? 2 (8.52) The reader must understand that the above discussion is for the simplest case ‘constant disturbance and is intended to show good disturbance rejection e™Mance of a PD controller. sill that in the development of the linear S/SO model in Section 8.5, the in can now be considered as a disturbance, Ihe bay f steady-state error means that the end-effector will not reach “8l point and will always be slightly away from it. i.StH Robotics and Conte! : 88 PID CONTROL SCHEME he steady-state error caused by a disturbaey : minimized but not eliminated. The control ma ware by 3 : ig . = f a alaw al JJiminate the steady-state er ms PID controller. The design of a partitioned PID (PPID) conte te apeble of eliminating te steady-state error caused BY @ constant disturyn : ostpe block -PPID control scheme is shown in Fig. 8.13. Tp, k diagram of t een pee of the PPID controller is identical to that of a PPD con Fig. 8.12, Eq. (8.34), that is G5) = Leg Tas) + Bop 88a(5) (853 The servo portion includes, in addition to the position and deriva components, an integral of the servo error E(s). The PID servo law is, thy obtained from Eq. (8.38) by adding the integral term as i In the PD control scheme th © Mody. lle ig 8, £,(3) = 5784(3) + Ks, (3) + K,0,(s) + AE (ass where K; is the integral constant. From Eqs (8.35), (8.53), and (8.54), the enor dynamics for the PPD controller is written as K, 8°E(s) + K,sE(s) + K,E(s) +" E(s) =0 (8.55) s ~~ Servorbased rx Modetbased ——____> i gf tanta) Joint ; +s mls) 1 ats) ln eee ns? + Boys | | ————j Bars }-+—+ In the presen : Presence of a constant disturbance as a step function of constatl Iapiide,a0) = Ks the eo dynamic gee by PE (+ Kuso) + Kye(6) 4 Ei peg) K a0 or Es ce iControl of Maniputators 08) ines BQ ti \K -_ OOS Ras Ret K, (8.98) (gives ead (8.59) Hor (o reach the desired target esenee of a constant disturbanee. , g.the PID controller eatnses he manip even in the p opserved that the P2ZD controller isa third order system, while PPD lew soit itis uate is 0 second order, Ln a third-order system, stability eriterion must be votstat the nal design stage and the serve pains are modified to ensure a pood Seaiigy MAT 89 COMPUTED TORQUE CONTROL. jnite previous sections, each joint of a manipulator was v an individual javeem (0 be controlled and linear controllers based on PD and PAD control ‘ehemes were examined for the simplified S150 model. More realistic problem is: ihe control of an 1-DOF manipulator as a single system. A control scheme that inakesdirect use of the complete dynamic model of the manipulator to cancel the fioct of gravity, Coriolis and centrifugal force, friction, and the manipulator inertia tensor, is discussed in this section. ‘The dynamic model of the n-DOF manipulator based on rigid body dynamics ives the n X 1 vector of joint torques , Eq. (D.16), Appendix D, as t= M(g)gt+ HG 9+ GQ (8.60) where M(q) is the n xn inertia matrix, H(g, 4) isthen x1 vector of centrifugal and Coriolis torques, G(q) is the n x 1 vector of gravity torques and q(), 4()y dW) denote the nx 1 vectors of joint positions, velocities and accelerations, respectively, Recall that q is the generalized joint variable The overall contribution of friction and other non-rigid-body effects can be included in dynamic model as a function, F(g,q) of joint positions and velocities. ‘Thus, the Multi-Input-Multi-Output (MIMO), nonlinear dynamic "Model of an n-DOF manipulator becomes t= MQg+HaDthaDte@ (8.61) The design of a nonlinear control system based on the complicated model given by Eq, (8.61), is considered now. Because this controller is based on a more curate (hough not the exact) dynamic model of a manipulator i provides better ‘ajectory-iracking performance than linear controllers do. The controller iscussed here employs the computed torque control law to modify the system and effectively decouple and linearize it. Similar to the PPD and PPID laws, the computed torque control scheme also Comprises two portions—a model-based and a servo portion “The model-based I torques Tusing a structure identical to tron defines the nx 1 veetor of contro of Eq. (8.61) as306] Robotics and Control im MQ)’ +H(G@ + FG 4) +G(q) : (86 1 is the nx 1 torque vector specified by the Servo portion, Com’ here 7 is the n < ; Ege (8.61) and (8.62), it is observed that id ively linearizes as well as deo.) |-based portion effectively Jeoup ‘Thus, the ars yr 4 nonlinear feedback of the actual Postan Sei otie "Thc schematic representation ofthis nonlinear contro] sch veloc . is shown in Fig. 8.14. +—§ Sere parton —+he— i q - r(s) SQ Manipulator [~*~ + Qa) Sf tema J oN i? eS al i ! ‘HG. 9) + FG. a) + G(q) | Model-based portion Je—p——_f Fig. 814 A nonlinear controller based on the computed torque control lax The control law of the servo Portion is based on then x 1 vectors E and E of dete in joint positions and velocities, Tespectively. These errors are defined as EW = gat) - (0) (8.64) and EO= 4d) 4p) (8.65) where q and q, denote the nx I vectors of actual and desired joint positions, respectively. The servo portion of the computed torque control scheme is, therefore, defined as U=Gi+K B+ KE (8.66) Ua tHe then xn matrices 0 Position and velocity gains, respective K, are ch Usually, and ° . ae. ‘osen as diagonal matri it stant gains. This serves decouple the ener ‘Sonal matrices with constant g E+ Kj by KE=0 0 This shows that the erro, dj i ified bY ® Second:orde near en cnet OF a closed-loop system is specifiedControl of Manipulators 307] mais vector equation is decoupled if K, and K, 2 i ris vector equation is decoup p and K, are diagonal matrices with ins. Hence, the error equation could oastant & be written on a joint-by-joi as. For joint ithe error equation is, ot dont by:solnt B+ Kus + Kye, =0 (8.68) with Kyi and K,; are the position and velocity gains, res ectively, for joint i. Forcritically damped performance of joint i, the relationship between K,, and x, is,a8 obtained in Eq. (8.43). That is 7 a Ka Ky (8.69) Observe that the computed torque controller employs nonlinear feedback to linearize error dynamics and provides a better trajectory tracking performance than linear controllers, since it is based on a more accurate dynamic model. But, the biggest limitation of this approach is that it is computationally more intensive and, hence, highly expensive when compared to linear controllers. Inaccuracies inthe parameters of the manipulator in the dynamic model are other factors, which limit the manipulator’s performance. Discrete time control, robust control, impedance control, force control and so on, are the methods employed to overcome these disadvantages and tackle more sophisticated control schemes. Some of these are discussed in next section. 810 FORCE CONTROL OF ROBOTIC MANIPULATORS The control systems for non-contact type of tasks have been dealt with in previous sections, where the manipulator moves in space and its position is controlled, using various approaches. A contact task can be used to perform an assembly, do drilling, wiping a surface and so on. In these applications, the interacting force(s)/ torque(s) must be regulated such that force(s)/torque(s) remain within specific range(s). A simple position-controlled manipulator is not suitable for these applications. Take the example of a high stiffness end-effector in contact with a fragile material such as glass. Say, itis required to pick a glass full of water. Ifthe environment is not known with exactness, or there is an uncertainty in stiffness of ‘he contact model, then even for small variation in these parameters, a large force Tay be applied on the environment. This may damage the environment (glass ‘umbler in this case) and no useful work may be done, Another example of limitation of position control is that it is impossible to scratch paint froma glass Pane with a high stiffness end-effector, using position regulation. If this is tempted, either the glass pane will break or the manipulator would wave the “taping tool with no contact with the glass pane. An alternative is to sense and control the force/torque that is exerted on the ‘vironment instead of position control of manipulator. The ability to measure ®d control contact force(s) generated at the end-effector offers a possible “Nemative for extending the effective precision of a manipulator rare force control is entirely dependent on the force sensing. Force sensing is le by locating sensors at the actuator, or at the tip of the end-effector, orppotics and Control he manipulator wrist. The last one ig m F getual force/torque May TEqUiTe Zeon, a tie location. ‘TROL TASKS BL from the en commonly Use transformations. If, or at th mination o! he sensor nvironment itse xd, The deter depending on OF FORCE-CON ntrol approach is discussed for a situation ily constrained to move due to contact to ion between the end-effector ang i rerstand the partially constrained tasks. Every arte tasks. A subtask is defined by a particular conta constraints in some directions and freedom ‘ 8.11 DESCRIPTION k, first force-c or is partial .d model of interact In this framewor which the manipulal surface, A simplifies environment is used to und task may be divided into su situation, which may have specific others srt action task is described using the natural constraints and/or artifiit constraints. These are + The natural constraints are the constraints imposed by the environment and are defined by the geometry of the task to be performed, either position constraint or as force constraint on the end-effector of the manipulator. ¢ * The sategy for executing a task imposes artificial constrains, The eine Hx ee, ch et ie by the manipulator are termed oe of the variables controlled These constraints are for each de; annals two sets are complimentary to each other, of freedom of the end-effector and the two classes of constr er. The following example illustrates these Consider the situat jon where a rob i : lass pane and (ii) wipe it clean, For ne is required to (i) scrap the paint fromaControl of Manipulators in the cave of wiping. of ghiws pane with soap, surface is frictionless, anol face and 0 pply an arbite ine jurce constraint cists, This is, the tangential force along, and & i ‘constraints arine from the fact sin of frame {c}, and hence the y force tangent to th Fons aust be Zero, Three additional natura jue ae action COrque is available at O, the « ape wo real " , about thet, he zero. int from the window pane, ipplied by the ‘owt For seraping Of P sired and a force has to be pecific motion of end-effector J effector in the di ction in is ew sic the end-effector hand is not free to moye, that is, perpendicular 10 # rice, The motion is specified by the user as the trajectory for faite surface of the glass. The artificial position constraints on the trajectory are gpevitied in terms of desired position and orientation trajectory of the end-effector its velocity. Similarly, if excessive force is applied it can break or damage the mall force is applied, paint will not be scrapped. Hence, this force: trust be controlled by the manipulator and it is the artificial constraint associated with the task. Notice that this force is not a variable that is subjected to natural constraints. Ingeneral, for each subtask configuration a generalized surface can be defined with position constraint along the normal to the surface and force constraint along the tangent. Ifa position constraint is naturally defined normal to the surface, then an artificial force constraint exists normal to the surface and is user defined, Similarly, in the direction in which natural force constraints exist, a user defined attfcial position constraint exists and is to be specified. It is worth noting that iheposition constraint is a vector comprising of position as well as orientation of the end-effector and the force constraint is a vector of forces in three principal directions and their respective torque components. Inother words, position and force along each degree of freedom are determined by either a natural or an artificial constraint. Thus, the number of natural constraints and the number of artificial constraints are together equal to the umber of degrees of freedom of the constrained (task) space, which is six in eneral, The natural constraint may be expressed in terms of artificial constraints ‘r vice-versa, in certain situations. For instance, in the above example of paint ating the friction between the scrapping tool and the glass surface relates the ential force along y- and z-axes with the normal force along x-axis, locity ne are usually indicated by giving values for components of ang, othe end-effector, V, which is a vector, comprising of linear as well as crs ities Because in many cases i is simpler to specify a position 2 a 88 a“velocty equals zero” constraint. istates a example, consider the peg-in-hole assembly operation as rnd pee re: 8.16. The figure shows an assembly sequence used to put a Steho sone hole. The peg is brought down onto the surface to the side sted uae lid along the surface until it reaches above the hole. Itis then Compe qa Pee Teaches the bottom ofthe hole, at which time the assembly is ical and difficult to meet requirement in this assembly is that the surface and ifsRobotics and Control 310} Robotics and Coniro alignment with the axes of the hole. This i &Ssemby e peg are always in ; ear is hee nb axes of the peg are alw “four subtask domains First the arm is fre iD the space task can be divided into four s >and ses vad the hl \e surf Ee thet a and atria taoving down the hole, and lasiy s serted into the hole a yi the peg is just inser touches the bottom of the hole. Gripper Assembly complete Peg ! inserted Costrained: frame (c) @ Fig. 8.16 Peg-in-hole assembly operation tasks ‘These four subtasks are illustrated in Fig. 8.16. For each of the subtast natural and artificial constraints keep on changing as the a: sembly is performed, The sequence of the planned artificial constraints desired manner. This requires substantial task planning and the controller has to switch from one control strategy to another. This is explained with the above CLIMPIC of peg-in-hole assembly. A frame (c} is attached to the end of peg in which the constraints are defined for the assembly. For the first subtask, the arm is free to move in space and the natural constraint, become that of force because the end-effector will not be able to apply arbitrar direction. Therefore, the natural constraints for the the frame (c} attached to the tip of the eg are y force in free space in any first subtask, with respect to ‘Omes the artificial constraint in which the task i rection with user-defined velocity v,. Hence, the artificial constraints are d velocity 2, *=1 0 v.00 of (8.7!) ‘are tabulated in Table |._— $$ Control of Manipulators 311} Table 8.1 Natural and artificial constraints for second subtask Natural Constraints =0 Py = Pvtite Z=0 A velocity in z-direction (crossing the threshold of zero) detects the third , subtask as the peg slightly falls into the hole, as soon as it reaches completely above the hole. This signals a change of natural constraints and anificial constraints. These new constraints are tabulated in Table 8.2 with Vinsers AS the «dlocity with which peg is inserted into hole. Table 8.2 Natural and artificial constraints for third subtask Natural Constraints Artificial Constrainis A force in z-direction, when the assembly is complete, detects the final stage. Apoint to note is that sensed variable is one, which is not being controlled. For txample, in case of the third subtask, the velocity in z-direction is monitore (sensed), while the force in z-direction is controlled. ; The above example is a simple one. In real situations, determining i Strategies may be quite complex. The planning including uncertainty, 7 Neh is aways present in real systems, in practical situations and automation of this is Silla research topic. &p FORCE-CONTROL STRATEGIES Teduaity between position and force control requires special bas ae mth various force control methods, two broad categories emerge for foree “nol: pure force control and impedance control. In fare fore saute Aime foree is directly tracked to the desired force, as seat ythenvionrtont. Inq’ ©an be applied only when the manipulator is in conta ition to regulate the “OMe, the impedance control methodology controls crepe impenae, the dynamic position-force relation, known astarget peda. paolled 1 the end-effector has been compliant enough, ee he malta 4 glass tumbler filled with wet i ton control itself ° Pane, the job could even have been Neae Robotics and Control __ olosies to achieve this compliant motion nang jologies et impedance control. " ‘e control and impe dance £0 Thatta ce hybrid positonfor oe on the observa hat when th sori, The first approac ronnie the workspace of t pe end ctor can aturally ye 2 envil Is ; -e subspace. Thesi contact with the en ion subspa and force nt v : {Wo suby S decomposed into iectiett in which the end-effector it Tespectively, ire 7 correspond to ihe oT py the environment, The compliance is achieved by move and consitiine’ te in force subspace and position in position subspay controlling for at pee ons explicitly con ficant task planning and control law Switching jg is requires signi ning an ‘Additionally, this approach is prone to robustness broblen imp! ation. ition between unconstrained and constrained motion. 4 papi” Fare irae. dedien of controller with low sensitivity €6 Parameter Variations Alternatively, the impedance control of compliant motion Proposes that the control objective be the regulation of the “mechanical impedance "Of the end. effector. Thus, the objective of impedance controller is to maintain a desired dynamic relationship between the end-effector position and contact forces, Ths scheme can be robustified using PID controllers. Other robustification schemes can be functional analytical method or Lyapunov-based variable structure approach. The hybrid position/force control and impedance force control, with a thrustto direct adaptation in impedance scheme, are discussed in the following sections There are two basic methor 8.13 HYBRID POSITION/FORCE CONTROL The interaction task between the manipulator and environment is defined in terms of the natural constraints and artificial Constraints with reference to a constraift frame. The control of artificial Constraint variables is required to accomplish the task, as explained in Section 8.11. Because the definition of artificial constrain's Ble degree of freedom g i 5 act with the © contact with e ‘em with mass m in contact Wi ess yesshe environment. Th hi ‘ AVironme; n contanesstc degree nn be controle eat led spring ofr disturbong oo” SPting-m: weak attached to the spring Fy Nance force (due to { illustrated in Fig. 8.17. Unt 5 “ction, backlash, etc:) applied to system ay Control of danipulators B13) jae 8 fae THE variable © be controlled is the force exerted on the iret fe are Fis te control force applied to mass by the actuator. | fe 1—e om VOOIDTO —> fast Fig. 8:17 Single degree of freedom mass-spring system pete balance gives the differential equation ofthe spring-mass system for slacement.x of mass/spring as famtt kext fas (8.72) ‘pe reaction force (f,) exerted by the environment, which is to be controlled, is force from spring fake (8.73) she sping constant k, is typically very high, of the order of 10° N/m. The "problem for the spring-mass system is to maintain, toa desired value fy. sce error, therefore, is defined as e=tanSe (8.74) m dynamics is obtained by differentiating Eq, (8.74) and ubstituting Eq. (8.73) in the resulting equation. ‘The result is The control syste: = hy -fe= fa het (8.75) Substituting ¥ from system model in Eq. (8.72) into Eq. (8.75), gives : . _k, a= fa- Fhe ~ fis) (8.76) ove expression is based on the assumption that environment is fixed and “ly known, Assuming that a PD control scheme is used in the controller so satisfies a desired closed-loop dynamics. The PD control law equation is 6, =—Kyey — Key (8.77) 7 & and K, are proportional and derivative gains. nity feedback, the closed-loop control law in Laplace domain becomes: Hels) + Kgey(s) + Kye (8) =0 (8.78) f PEL A i ky i action, that is the actuator force is obtained by equating right-hand 4 (8.76) and Eq. (8.77) asGIG) Robotics and Control pots + Baer + Keer] Fe * Fae (ay ‘ _itcannot be included direct! used in the control law in pig the disturbance force fais !S know force fis Ce of if desired force Jy is del becomes, Because trol law. However, Cth ) in Eg.(8.79), the control mot pati t Key +Kyej]* fa (8.89) ‘The block diagram for the c' mass system, is illustrated in Fi 7 Manipulator ‘ontrol law in Eq. (8.80), as applied to the spring. ig. 8.18. Fig. 8.18 Force control for spring-mass system In practical applications, the requirements may be quite different from ideal situation in Fig. 8.18. For example, the task may require application of constant forces, which means that the inputs of the control permanently set to zero. Sensed force (f,) is a noisy derivative f, would also be a noisy signal. Alternative approach is to estimate J, from other variables that can be accuratel: of Eq. (8.73) gives system fy and fy ae \d, hence, its y proce: y sensed. For example, differentiation $1) Seek, As joi ity | . ; Joint velocity (+) can be sensed quite accurately, with constant ,. f: can obtained fro1 , : ; noise for use is Eq. (8.81) accurately, This isa better estimation of j, free f0™ ise IN control system. The new control law is, therefore, written _ [K, : fom [se, +k; +n ge) system block diagr: ‘ 7 v lagr 7 hes 8.19. This control law has thea the desired force deriv; : hecomes Jess ScaSitive ( The cont rol am for the new control law is show ‘vantage that it does not require the in" tives. The con em niroller gains a o syste " . gains are chosen such that thes) to variations ink re chosen such that ofControl of Manipulators S15) ; ipulator i + ' ' | -Q—-m 2 al re Irigy (ey iy : k fo Practical force control system for a spring-mass system sybrid position/force control can use PID controller to improve steady-state pefonmance. In addition, systems are made robust with respect to variations in {which is associated with the elastic deformation of the environment and the sunipulator. The environment stiffness k, is not known with exactness and quite often changes with space and time. 33.1 Control Architecture for Hybrid Position/Force Control Scheme Having developed the hybrid position/force control as a force control servo, it can te easily applied to multidegree of freedom systems. Consider a Cartesian configuration manipulator with 2-DOF, having two prismatic joints as shown in Fig.8.20. For simplicity, it is assumed that each link has mass m and slides on frictionless linear bearings. Also, assume that the joint motions are lined up cxactly with the constraint frame {c} as the manipulator follows the environment surface, as in Fig. 8,20, 6, Environment ONG Fig, 8.20, A 2-DOF Cartesian manipulator contacting a surfiice Th Main sd by the environment surface while Foing «contact foree f).'The contaet between the manipuktor end "Mere Muro amed frictionless. This means that the cavironment tion fe : : 3 ena Fee F, bs always normal to the surface, The normal to the sauce ts ‘thy tion ‘ction of frame (e}; heace, foree control is required in that MI position control in.v-dircetion.seoo™—_—_--- i cs and Control BB Robotic sesenve that the de sited UB}CC1OCY i taghag wath desired Value fp THUS, a oyi i ceaiection and a force trajectory an be built such that a comple ar tae toe directions and also 2 Force taject in all three degrees Ot which trajectory will be followed al any given time inate whic compan Crare fr this control strategy is shown in Fig. 8.2 Tspestion conlle ‘and the force controller are s eparated as two independeny Tae rece interact wih each other through the artificial positon and forse constraints. The control mode—position OF foree—to control each joint of the spanipulatoris decided on whether position contro isin effect or force control is in effect, ib rhe control problem is hybrid int ona onitored noe ism the defined independently iD 1” structure of controller while the contact fo" trajectory can yedirection. In 8 eral Two nxn diagonal matrices C and C’ are used to describe the switching ce control modes. Recall that the position control and between position and for forse control due to artificial constraints are mutually exclusive. This fact and the artificial constraints fix the contents of matrix Cand C’. The diagonal elements of the matrices are ones and zeros, such that, where a one is present in C, indicating position control in effect, a zero is present in C’. Where a zero is present in C,a one is present in C’ and force control is in effect. x 1 ge 3 “ie 7) Position | 53 «| | ion Se : as +d —> |x Constraints — ae Manipulator 33. PPE oP Nebel Te £8 f | Fore Be | controller [—>L_S" | 7 ESSE TERE aas a ernie J Fig. 8.21 Hybrid controller for 2-DOF manipulator i fer the 2:DOF Cartesian manipulator discussed above, because x component Position controlled, element (I, 1) is one on the diagonal of C and zero 08 C. Similarly, element (2, 2) i i o * As Zero it i y-direction. The two diagonal matrices ee eee tL 10 c-{y 01 and e-[} 1 (8.83) Note that C’ matri at C’ matrix has the zero and ones on diagonal inverted from C. 8.14 IMPEDANCE FORCE/TORQUE CONTROL The impedance control specifi ies the . position rather than the position dynamic relationship between force a4 and force constraints and switching law betwee”Control of Manipulators i sin hybrid force co ye trol discussed above, ‘The dy we oF the mechanical nic relationship Macon the fowe and position deseribes the imped wines of Hh SYSEETR, | J atte ) | impedance ConEOL provides a unified framework for consid | ing both onsirained and unconstrained motion control problems and possesses certain Yranage for many applications. Kor example, this approach ean often be mpiemented with Tittle Off-Line task planning, it provides robustness: to xqainties and disturbances and can accommod ¢ stable transition betwe ned motion. In impedance control approach, the vance! constrained and cor agnitude of contact force depends on the reference position trajectory for the eadettector as Well as on the environmental stiffness. There is no builtin anism in impedance control to ensure that excessive contact forces are not « sa result of poor choice of the reference position trajectory or inexact estimates of the environmental stiffness, The class of applications that require aecunite regulation of the contact force or demand that the contact force lie within an acceptable range, require trajectory adaptation. In the ideal case, where the parameters of the environment (i.c. location and stiffness) are known exactly, a reference position trajectory cin be synthesized a priori to produce the desired contact force. However, in practical cases, where the environmental parameters ar-not known exactly, the impedance control scheme tends to exhibit poor force racking characteristic. Hence, a mechanism is required to be developed for providing force-tracking capabilities within the impedance control framework This can be done by two simple force-control schemes for automatically generating the reference position trajectory to provide force-tracking capabilities on-line. The first scheme is based on generating the reference position on-line as 4 function of force-tracking error. This is called direct adaptation. The second scheme estimates environmental parameters on-line, using indirect adaptive approach and the required position is computed based on these estimates. 8141 Force Tracking Characteristic of Impedance Control Consider an n-DOF manipulator operating in n-dimensional task space with "$6. When the tool-tip comes in contact with a hard surface, as shown in Fig. 8.22, it cannot penetrate the surface and a natural position constraint is <— Hard surface WIA, Fig. 8.22 Interaction of tool-tip with environment5 ee BIB, Robotics and * 4 1. Itno longer makes sense to attempt [0 ee ey in the Girectig defined. It no longer ma direction in Fig. 8.22. n srpendicular to surface, that is. the x-direction ea hee Ook g {emg ones soto main ont With ea force along. dire ans aie i 7 “ urtificial constraint. Other artificial and natural constraints can be Wentitieg is the artificial constra Harrie force between the tooltip and the surface produces a deforma of the environment (the surface) because no real surface is infinitely hard, The environment deforms inthe vicinity of the point of contact, as shown in Fig 823 deformation or the elastic behaviour of environment can be Modeleg ized spring of stiffness K,. The spring constant is a POSitive-definite This elast asa gener h Xm diagonal stifiness matrix called environment stiffness matrix and is denne as K,=diag{k, ky ky} (ee where the scalar k; denotes the environment stiffness along dimension j. end-effector eo Xe FF Keontact >| he contact force in Eq. (8.85) is reali «cag THe zed by applying t s at joints. T! ce Chapter 5, Ea. apne lorques at j t='@ F (8.86) “auired joint torques are (sJig) is the 6X7 manipulator Jacobian matrix and g is mx 1 joint wrjacement vector The Jacobian also relates the infinitesimal joint dite tments 6g tothe infinitesimal end-effector displacement (X ~X,. se X-X, = Sg) 6q (8.87) ining Eas. (8:85)-(8.87) gives et " 5 8.88) J'(g) K, Jiq))dq (8.8 pe term in brackets, in Eq, (8.88) is called the join stiffness matrix. K.. that S K,=J"q) KJ) (8.89) Note that K, is not diagonal and may degenerate at manipulator singularities. uation (8.88) relates the joint displacement g and the joint torques T with the cwironment stiffness K,, that is, the variables directly relevant to the control ggsem, The impedance control is developed on these relationships. Let E fepesent the displacement error as the difference between the desired end- effecior location and the actual location, that is E=X,-X (8.90) Itisassumed that Z is same a: the infinitesimal deflection of end-effector from itsreference location. A PD-plus-gravity control law to control the dynamic interaction between the manipulator and its environment, is t= J" (q)[K,E + K,E]+ Gq) (8.91) wheteK, and K, are the proportional and derivative gain constant matrices: Gq) isthe gravitational torque matrix; J(q) is the manipulator Jacobian and E is the displacement error vector. The matrices K,, and K, in Eq. (8.91), if assumed to represent “stiffness” and “damping” of the manipulator as seen from the eavironment, ( K,E + Kyi ) becomes the equation for spring- 0 as t + 0, hence, the end-effector position tracks the reference position trajectory Xj, ___ When the end-effector comes into contact with the environment, the dynamic interaction between the manipulator and environment is dictated by the model given by Eq. (8.94). Now to produce a constant force F, desired displacementX, 's specified to penetrate into the environment by a constant amount, which means ; X,=%,=0 (8.97) with this, Eq. (8.94) reduces to tier, the objectiv userspy end-ef chosen as a lines damper-spring-ma ‘ode! for target impedance is, thus To accomplish force trackin MAO + BRO + KK) ~ XC] =B (0) (8.98) 7 In Cartesian space control, eacl independently. Thus, replaced by the lower case letters to re ‘The target impedance model, therefore, h Cartesian variable can be considered mk + Bi + Kxy-x)=e@ From Eq. (8.85) ana HH rend ee with er= fy—f, it becomes . Fe + (Ly elk, fea eo- SETS. = Contr of Manipulators ii substituting for and its derivatives in Eq, (8.99) and rearranging gives miy + Bey + (k+ keer a + By ~My + kK(xy-x,) (8.102) she steady-state force error is obtained by a a gq 03) 10210. Ths gives 'Y equating all time derivatives in -_k Oe a Ry lat kelay ~ ¥)] ‘zai (8.103) which can be written in a concise form as E55 = keg “a ttan (8.104) vere ky i the equivalent stiffness given by = kk. kek, (8.105) From Eq. (8.103), it is easily seen that if xy=.x, + fi/k,s then e,, = 0. In other words, if the precise location (x,) of the environment and the exact value of the environment stiffness (k,) are known a priori, the desired position trajectory xy canbe synthesized accordingly to exert the desired contact force. But, in practice x,and k, are never known exactly, as a result desired force f, cannot be exerted. Inaddition, apart from uncertainties in environmental parameters (x, and k,), there will be residual position errors because of imperfections of robot position control system. If this residual position error is denoted by ¢, it is going to contribute an additional force error as k, (mé + Bé + ke) and the Eq. (8.102) will become més + Bir + (k-+ke)ey = mf, + Bhy + Ky + khe x, + kp[mé + BE + K(E- xq) (8.106) If the position control system is stable, in the steady state, €=&=0. Equation (8.103) will be modified to fa a es ohy[ deer, +$e-Xy (8.107) Itis seen that the residual position error € contributes to the force error e-and ‘causes an increase in e,,, as is expected. This scheme is called pure impedance Scheme and can be implemented by the block diagram shown in aula 9. [srowerd |X lator ronward | spin a "6 mean sf inverse | 26 2)-| _soint |) __ filter kinematics | controller ae ematics) trol | ri aaah 16 \e eS ronment Hal en stiness, Ke 75-825 Pure impedance control system with an inner-loop position-contral system p Oe enesand Control ae ‘ontrol BB Robo 8.14.2 Adaptive C 1 system can be built for the impedance contro, adap ipove. The control system is made adaptive to actually disewstate force errors. AN adaptive control system based oq sdapation scheme is shown in Fig, 8.26. The direct adaptation sehen the application of a desired contact fo An adaptive control Ye Temove ye the ding Me enahy of enviroment parameters, x, and k Xa ae Position x M 5 impedance fiter | -Bicontrohed aol oar manipulator ie Fig. 8.26 Direct adaptive impedance control scheme Underlying concept of direct adaptive control scheme is to generate the reference position x, as a function of the force tracking error. A possible adaptive control law is Xu = g(t) + K, (Ne) + Kye, () (8.108) where K,(#) and K,{f) are the adaptive proportional and derivative feedback gains | acting on the force error ¢,(1) and the error rate é , (1), respectively, and gi) is the auxiliary signal to be generated by the adaptive scheme. Note that the force error rate é,(#) is ordinarily not available because fis noisy signal. One approach is to filter the force/torque signal to remove the high- frequency noise and then differentiate the filtered signal. This approach works well when the robot moves slowly into contact with a soft environment. AM alternative approach is to replace ¢(f) by — k, % as suggested by Eq. (8.10) and noting that f, is constant, In this case, since &, is an unknown pesiti® constant, it can be absorbed in adaptation gains and in the weighing factor. Ths approach is suitable when the sampling rate is high and the end-effector velo" Measurement is accurate. seo ek at mpedance, Eq. (8.92), is used for the adaptive cont! i the error dynamics from Eq. (8.102) becomes més + Bes +(k+ Key = k(~ fy + Kee) + kk Xe Substituting x, from E the error dynamics is [PRR [Mk tte KCL + here) + Rhee 0 m a (3.109) ' raw (08 4. (8.108) in the above equation, the adaption mControl of Manipulators 323| ere This equation represents the “adjustable system” in the model-reference adaprive control (MRAC) framework, If desired behaviour of the force tracking is represented by Bm + Oey + Oe, (8.111) where {and @, are the user-defined damping ratio and undamped natural frequency of the force-error dynamics, respectively, then an adaptation law for (0, K,{0) and K,AD, which ensure that the response of the error dynamics of Bq. (8-110) tends to that of reference model of Eq. (8.111), can be developed by proportional integral (PY) adaptation law. SOLVED EXAMPLES Example 8.1 Linear second order system model Consider the rotational system with a rotational mass at the end of a shaft as shown in Fig. 8.27. The input, a torque ¢, is applied to the disk with a moment of inertia J about the axis of shaft, which is fixed at the other end. The torsional siiffness of the shaft is k and the bearing friction constant is B. What is the condition for the system to be critically damped? Bearing ff 4 (8) Torque (7) Z ie 4 wi \ Aeossvoory | \ 4 3 J 4 1 WHA ye (El 4 aN hg Tho Shaft (k) — \ B inertia J (a) Physical situation (b) Dashpot-spring-mass model Fig. 8.27. Rotating mass at the end of a shaft Solution First, the differential equation (mathematical model) of the system is obtained, Consider the dashpot-spring-mass model of the system shown in Fig. 8.27(b). The torque balance gives the equation 10 =7-plo_ Kg 8.112 BO ( ) or 1894p pga (8.113) dr dt This is the second-order linear differential equation model of the torsional System, The Laplace transform of Eq. (8.113) gives(BB) bots and Conte k O(s)= 4s) 15°6(s) + Bs0(9)* Po patih re ace transform of input and 6(: place tana function of the system is as) -__1___ G>T 5) Is? + Bstk Gis equation for the torsional system is where 7(s) is Lap! Mot output. Then, the transfer and the characteristic: Is? + Bst+k=0 Boke ps saTstya0 11g ‘The condition for critical damping requires the damping factor ¢ equal to 1 ‘The standard form of second-order differential equation, Eq. (8.8), is +20 0,5+@; =0 (1) Comparing the Eq, s (8.116) and Eq, (8.117) gives the system parameters as = | ®, f (8.118) and y 26 0, = a (8.119) Solving Eqs (8.118) and (8.119), the damping factor is found as B SL 8.120 Wk ge ‘Thus, for critical damping, B=2Vik (8.121) Equation (8.121) puts a constraint on the linear second-order system Parameters J, B, and k that must be satisfied if the system is to have a critically damped performance. ; Ifthe inertia Tis 1 kg m s’/tad, stiffness of the shaft kis 10 N m/rad, bearing 2Nmis (amping Coefficient) B is 2 Nm s/rad and a constant input torque Tf 'Sapplied, then the response of the system can be found as follows: Substituting the values of above parameters with 1(s)= 2 for step inptit Eq(8.114), gives : #°8(5)+2s (5) +10 (5) =2 a) ot s Katie 2 oe ‘S@ F254 0) From as (6.118) and gig)” *2°+10) 2 ®=10 or ®, 16 rad/s lime. iyr al peretore, the system iy a J trom Hq. (8.15) Contrl of snipers Hl 660316 yped and the time domain 1 “pon bia (1) © 0.2. 0.2108 e sin 71.57) (4.526) example 8.2 Model-based control tor with a linea \anipul Consider second-order dynamic model as: t=10, + BO, (8.127) where J is total inertia and B is effective friction. The actuator gain is K,, in Nms/rad. The manipulator is required to follow the desired trajectory defined by [by 4 6; |. A model-based control system as shown in Fig. 8.28 is deployed forthe dynamic control of the manipulator. Determine the error dynamics of the system and the condition for critically damped response of the system, 8% +,| Manipulator [ i | Ki dynamics |, + [sacs al a Fig. 8.28 Model-based trajectory following robot control Solution The given dynamic model of the manipulator is = 16, + BO, From the control system block diagram in Fig. 8.28 the torque Tis mining Eqs (8.127) and (8.128) gives t= Kee +6] (8.128) and 1 =8, 4+ Ket Kye (8.129) ‘here error ¢ and its derivative are defined as 0, -0 oA 8.130) 6=0,~0, : Hog + Kye+ Ke) BO, = 10, + BO, (slab) 4eS EE 326| Robotics (8.139) or h ni 8 —8.,. the system's dyna, Defining the seco! performance in error domain & b+ Ke+Ke=0 (hl given by in Bq, (8132) is independent of parameters IB, and, because of cancell: Bat in the model-based control stay employed in the control system in Fig 8.28, signals bropertional to these are provided, requiring perfect knowledge of paramefers 1B, ad Ke vAs pet Eq, (2.132), when excited by initial conditions, the mor andi derivatives, the system dynamics seeks zero stgady-state error, that is, ¢,,=0 ‘This means that the system follows the trajectory demanded through commang inputs 8,04, and 6y. . : ‘The gains K,, and K, ase adjusted for critically damped response, for which they are related as (see Eq. (8.43)), Ky =2YKy ‘The error dynamics in Eq. (8.132) is linear even though the actual robot dynamics may be nonlinear. The nonlinearity effects of robot parameters/, B,and K,, are cancelled out by blocks /K,,and B/K,,, in error dynamics, ‘However, in actual practice robot parameters J and B are imperfectly known and, hence, they have to be estimated online and become variables. for the control system. Therefore, these parameters need to be continuously adjusted requiring adaptive control schemes. Example 8.3 PID control of a joint Consider a servomotor driven joint-link with no viscous damping as shown in Fig. 8.29 initially at @ = 0 rad. The PID controller block diagram for the joint controller is as shown in Fig. 8.30 with the following specific parameter values: Ly = V kgm Kr= 10N nV/A R,=5Q A=20VIV (8.134) K, = 0.1 Virads/s = 10.5 V/100 rpm , = 20.94 V/1000 rpm = 0.2 Virad/s m= kg : Dea ; ; termine the transient performance of the joint if its final position 1/4 radians, that is, 0y(s) = 9-785. s Assuming no disturbance ai control, the transfer function of the joi Solution ° ind no proportional, integral, or derivat®© nt controller is obtainedControl of Manipulators Bf Link drive shat Link | o= #14 o=0 Fig. 8.29 Servomotor driven joint-link Fig. 8.30 PID Controller for a joint AKy As) Ry Outs) {are Kea +aKk,)+ Akt Substituting the value of given parameters Os) O48) a linear second-order sys ‘or the joint control system, without (8.136) ristic ity loading and no PID g % s48.2 944050 (8.137) Paring ng Eq. (8.137) with 1 rad/s (8.138)[= 0.648 (8.139) = = value of damping factor is small (¢< 1) and the system respon. The above value 0 without PID control is oscillatory. The gravity loading for t jointis initially at 8 = 0 rad, het Joad at the motor but as @ increas = n/Athe gravitational force pro loading is considered as a distu disturbance torque, assuming accel 4,(0) = 9.8 sin 0(1) (8.140) “The performance of the control system is studied using MATLAB for computer simulation. For a step input, the response of the joint is obtained for different values of K,, Ky, and Kj, The results are illustrated in Figs 8.31 and 8.32, The first of these figures shows the response with proportional controller and the effect of adding a derivative control. It is observed that with only proportional control (K,, Kj, = 0) there is an overshoot as well as steady state error is present due to gravitational disturbance, as expected. Increasing K,, increases the overshoot and reduces the steady-state error. For a fixed value of proportional gain (K, = 30), the effect of adding the derivative control is a smaller overshoot. As expected, the larger the derivative controller gain X,, the smaller is the overshoot. It is possible to obtain a response with practically no overshoot, that is, critical damping. The steady state error is still present and does not vary with Ky The effect of adding integral control is illustrated in Fig. 8.32 for a fixed Proportional term K, = 30, Adding the integral term in the controller eliminates the steady-state error. he joint is proportional to the sine of 6. Note that the ee the gravitational force produces no additionay ses with time due to command signal to reach duces additional load at the motor. This gravity bance. The time variation of gravitational Jeration due to gravity to be 9.8 kg m/s?, is Joint position (rad) ~ > Ka=0.01; Ky = 30 "> Ka= 0.02; Ky = 30Joint postion (rea) Time, t Fig. 8.32 Step response of system with PID control Itis possible to get the desired performance from the system by choosing the appropriate values of parameters of PID controller. A stepwise thumb-rule for selecting the parameters can be 1. Adjust K, until the system step response is either critically or slightly underdamped with K, = K,=0. 2. Next, adjust K; to reduce the steady-state error to acceptable value (or eliminate completely). Normally, K; is greater than K,, 3. Finally, increase K, until the step response is again critically damped or slightly overdamped. Example 8.4 The one link manipulator Consider the controller design for a single-link manipulator shown in Fig. 8.33. The load carried by the link is modeled as a point massm, at the distal end ofthe link, m, Pig, 8.39 A one-auis manipatator gimcerted perdu ee rr rere eeeCCeecee ee7h eT GGG] Kobatien and Control Fhe dynamic mode eal ay xin manipulator of Fink lengths 1 Tink to be homogenous thin £04, i J of one Solution omit uniformly distribut m ms, eal nO eM HG sy are no velox avity and C, = cos @}. The J centrifugal forces because there is only ration duc (0 £F where gis accele r jolis and teopling, terms due 10 COF axis. fases the link inertia”! (unknown) incre Ihe external load mass) ™, ofimertia due tom, at the distal end of Tink, and gravity load «the joint can be included in the dynamic rondel mL; the mome The viscous friction a by me as yp, with Bas the damping coetFiclent Thus Fe ee (8.142) Let the error equation for the system be 1é+Kyo+K,e=0 (6.143) where I (# +m, JG (8.144) Substituting, e = 6,~ 6), Eq. (8.143) becomes 16, + K,(8, -8,)+K (8, -9)) = 18, (8.145) Hence, the torque to be supplied by the actuator motor is t= 16,+K,(8,-8 zl 1+ KB, 81) + Ky (04 ~0)+ 8,8, +2 el, C, elie If the manipulator has t a desi i 6, 6 st follow a desired trajectory defined by [Bz 81 82). these become the in, i put to the control s: i i Fig. 8.34 is a realization of such dail ayant Pee a Fig. 6.34 Control system for oned palate: ‘ont one-link manips» ae a Control of Manipulators 30H 5 Impedance controt ne 2-DOF shown in Fig. 8.35. Drive the control Law for the manipulator assuming no gravity loading. xainple RR manipulate caste eeaiance © a wt ‘The Jacobian for RR manipulator was obtained in Example 5.4 for w 7 aso! length (Ly ). The significant Jacobian for RR manipulator lengths Ly and Ly is (see Exercise 5.5) LS, - LS LS, JQ= fei ae lige ua] th link (8.147) Y 8.35 The 2-DOF RR manipulator here C, = cos 8,, 5; = sin 8,, Cyp = Cos (6; + 84), and 5,» = sin (8; + 8). This Jecobian indicates singularities at 8, = 0 (fully stretched) and 8, = x (folded). The impedance control law is given as, Eq. (8.91) c= J" @)K,E + K,E]+ Gq) (8.148) Assume that the controller stiffness matrix K,, and the controller-damping matrix K, are diagonal and are given as K, = diagKy Kya] K,=diaglky Ka] (8.149) The error vector is E=[e, eo)" (8.150) Substituting values, the control law, Eq. (8.148) gives FE (L 8) + LaSin) (Kes + Kner) (LG # LC 2 )(Kyaes + Ruse) + a4 (8.151) LyS\y (Key + Kune) + LCi (Kpaer + Rares) ey (8.152) Thee lh comtrotter fined natu Shae atu i sK, and K, aire selected on the basis of the manipulator task nd artificial constraints while keeping track of dam