电子工程师手册（英文版六）：ch101

Lasky, T.A., Hsia, T.C., Tummala, R.L., Odrey, N.G. “Robotics” The Electrical Engineering Handbook Ed. Richard C. Dorf Boca Raton: CRC Press LLC, 2000 101 Robotics 101.1 Robot Configuration Cartesian Configuration?Cylindrical Configuration?Spherical Configuration?Articulated Configuration?SCARA Configuration?Gantry Configuration?Additional Information 101.2 Dynamics and Control Independent Joint Control of the Robot?Dynamic Models? Computed Torque Methods?Adaptive Control?Resolved Motion Control?Compliant Motion?Flexible Manipulators 101.3 Applications Justification?Implementation Strategies?Applications in Manufacturing?Emerging Issues 101.1 Robot Configuration Ty A. Lasky and Tien C. Hsia Configuration is a fundamental classification for industrial robots. Configuration refers to the geometry of the robot manipulator, i.e., the manner in which the links of the manipulator are connected at each joint. The Robotic Industries Association (RIA) defines a robot as a manipulator designed to move material, parts, tools, or specialized devices, through variable programmed motions for the performance of a variety of tasks. With this definition, attention here is focused on industrial manipulator arms, typically mounted on a fixed pedestal base. Mobile robots and hard automation [e.g., Computer Numerical Control (CNC) machines] are excluded. The emphasis here is on serial-chain manipulator arms, which consist of a serial chain of linkages, where each link is connected to exactly two other links, with the exception of the first and last, which are connected to only one other link. Additionally, the first three links, called the major linkages, are focused on, with only a brief mention of the last three links, or wrist joints, also called the minor linkages. Robot configuration is an important consideration in the selection of a manipulator. Configuration refers to the way the manipulator links are connected at each joint. Each link will be connected to the subsequent link by either a linear (sliding or prismatic) joint, which can be abbreviated with a P, or a revolute (or rotary) joint, abbreviated with an R. Using this notation, a robot with three revolute joints would be abbreviated as RRR, while one with a rotary joint followed by two linear (prismatic) joints would be denoted RPP. Each configuration type is well suited to certain types of tasks and ill suited to others. Some configurations are more versatile than others. In addition to the geometrical considerations, robot configuration affects the structural stiffness of the robot, which may be an important consideration. Also, configuration impacts the complexity of the forward and inverse kinematics, which are the mappings between the robot actuator (joint) space, and the Cartesian position and orientation of the robot end-effector, or tool. There are six major robot configurations commonly used in industry. Details for each configuration are presented in subsequent subsections. The simplest configuration is the Cartesian robot, which consists of three orthogonal, linear joints (PPP), so that the robot moves in the x, y, and z directions in the joint space. The Ty A. Lasky University of California, Davis Tien C. Hsia University of California, Davis R. Lal Tummala Michigan State University Nicholas G. Odrey Lehigh University ? 2000 by CRC Press LLC cylindrical configuration consists of one revolute and two linear joints (RPP), so that the robot joints correspond to a cylindrical coordinate system. The spherical configuration consists of two rev- olute joints and one linear joint (RRP), so that the robot moves in a spherical, or polar, coordinate system. The articulated (arm-and- elbow) configuration consists of three revolute joints (RRR), giving the robot a somewhat human-like range of motion. The SCARA (Selectively Compliant Assembly Robot Arm) configuration con- sists of two revolute joints and one linear joint (RRP), arranged in a different fashion than the spherical configuration. It may also be equipped with a revolute joint on the final sliding link. The gantry configuration is essentially a Cartesian configuration, with the robot mounted on an overhead track system. One can also mount other robot config- urations on an overhead gantry system to give the robot an extended workspace, as well as free up valuable factory floor space. The percentage usage of the first five configuration types is listed in Table 101.1. This table does not include gantry robots, which are assumed to be included in the Cartesian category. Additionally, this information is from 1988, and does not accurately represent current usage. In general, robots with a rotary base have a speed advantage. However, they have more variation in resolution and dynamics compared to Cartesian robots. This can lead to inferior performance if a fixed controller is used over the robot’s entire workspace. Cartesian Configuration The Cartesian configuration consists of three orthogonal, linear axes, abbreviated as PPP, as shown in Fig. 101.1. Thus, the joint space of the robot corresponds directly with the standard right- handed Cartesian xyz-coordinate system, yielding the simplest possible kinematic equations. The work envelope of the Carte- sian robot is shown in Fig. 101.2. The work envelope encloses all the points that can be reached by the robot arm or the mounting point for the end-effector or tool. The area reachable by an end effector or tool is not considered part of the work envelope. All interaction with other machines, parts, or processes must take place within this volume [Critchlow, 1985]. Here, the workspace of a robot is assumed to be equivalent to the work envelope. There are several advantages to this configuration. As noted above, the robot is kinematically simple, since motion on each Cartesian axis corresponds to motion of a single actuator. This eases the programming of linear motions. In particular, it is easy to do a straight vertical motion, the most common motion in assembly tasks. The Cartesian geometry also yields a constant arm resolution throughout the workspace; i.e., for any configu- ration, the resolution for each axis corresponds directly to the resolution for that joint. The simple geometry of the Cartesian robot leads to correspondingly simple manipulator dynamics. The disadvantages of this configuration include inability to reach objects on the floor or points invisible from the base of the robot, and slow speed of operation in the horizontal plane compared to robots with a rotary base. Additionally, the Cartesian configuration requires a large operating volume for a relatively small workspace. Cartesian robots are used for several applications. As noted above, they are well suited for assembly opera- tions, as they easily perform vertical straight-line insertions. Because of the ease of straight-line motions, they are also well suited to machine loading and unloading. They are also used in clean room tasks. FIGURE 101.1 The Cartesian configuration. (Source: T. Owen, Assembly with Robots, Engle- wood Cliffs, N.J.: Prentice-Hall, 1985. With per- mission.) TABLE 101.1 Robot Arm Geometry Usage Arm Geometry Percent of Use Cartesian 18 Cylindrical 15 Spherical 10 Articulated 42 SCARA 15 Source: V. D. Hunt, Robotics Sourcebook, New York: Elsevier, 1988. With permission. ? 2000 by CRC Press LLC Cylindrical Configuration The cylindrical configuration consists of one vertical revolute joint and two orthogonal linear joints (RPP), as shown in Fig. 101.3. The resulting work envelope of the robot is a cylindrical annulus, as shown in Fig. 101.4. This configuration corresponds with the cylindrical coordi- nate system. As with the Cartesian robot, the cylindrical robot is well suited for straight-line vertical and horizontal motions, so it is useful for assembly and machine loading operations. It is capable of higher speeds in the horizontal plane due to the rotary base. However, general horizontal straight-line motion is more complex and correspondingly more diffi- cult to coordinate. Additionally, the end-point resolution of the cylin- drical robot is not constant but depends on the extension of the horizontal linkage. A cylindrical robot cannot reach around obstacles. Additionally, if a monomast construction is used on the horizontal linkage, then there can be clearance problems behind the robot. Spherical Configuration The spherical (or polar) configuration consists of two revolute joints and one linear joint (RRP), as shown in Fig. 101.5. This results in a set of joint coordinates that matches with the spherical coordinate system. A typical work envelope for a spherical robot is shown in Fig. 101.6. FIGURE 101.2 Cartesian robot work envelope. FIGURE 101.4 Cylindrical robot work envelope. Y max Y min X min X max Z max Z min FIGURE 101.3The cylindrical con- figuration. (Source: T. Owen, Assembly with Robots, Englewood Cliffs, N.J.: Prentice-Hall, 1985. With permission.) Z max Y max Y min Z min X max X min 0? ? 2000 by CRC Press LLC Spherical robots are typically heavy-duty robots. They have the advan- tages of high speed due to the rotary base, and a large work volume, but are more kinematically complex than either Cartesian or cylindrical robots. Generally, they are used for heavy-duty tasks in, for example, automobile manufacturing. They do not have the dexterity to reach around obstacles in the workspace. Spherical robots also do not have fixed resolution throughout the workspace. Articulated Configuration The articulated (or anthropomorphic, jointed, arm-and-elbow) configu- ration consists of three revolute joints (RRR), as shown in Fig. 101.7. The resulting joint coordinates do not directly match any standard coordinate system. A slice of a typical work envelope for an articulated robot is shown in Fig. 101.8. The articulated robot is currently the most commonly used in research. It has several advantages over other configurations. It is closest to dupli- cating the motions of a human assembler, so there should be less need to redesign an existing workstation to utilize an articulated robot. It has a very large, dexterous work envelope; i.e., it can reach most points in its work envelope from a variety of orientations. Thus, it can more easily reach around or over obstacles in the workspace or into parts or machines. Because all the joints are revolute, high FIGURE 101.5 The spherical configuration. (Source: T. Owen, Assembly with Robots, Englewood Cliffs, N.J.: Prentice-Hall, 1985. With permission.) FIGURE 101.6 Spherical robot work envelope. Z max Y max Y min Z min X max X min 0? FIGURE 101.7The articulated configuration. (Source: T. Owen, Assembly with Robots, Englewood Cliffs, N.J.: Prentice-Hall, 1985. With permission.) ? 2000 by CRC Press LLC speeds are possible. The articulated arm is good for tasks involving multiple insertions, complex motions, and varied tool orientations. The versatility of this configuration makes it applicable to a variety of tasks, so the user has fewer limitations on the use of the robot. However, the same features that give this robot its advantages lead to certain disadvantages. The geometry is complex, and the resulting kinematic equations are quite intricate. Straight-line motion is difficult to coordinate. Control is generally more difficult than for other geometries, with associated increase in cost. Here again, arm resolution is not fixed throughout the workspace. Additionally, the dynamics of an articulated arm vary widely throughout the workspace, so that performance will vary over the workspace for a fixed controller. In spite of these disadvantages, the articulated arm has been applied to a wide variety of research and industrial tasks, including spray painting, clean room tasks, machine loading, and parts-finishing tasks. SCARA Configuration The SCARA (Selectively Compliant Assembly Robot Arm) configuration consists of two revolute joints and a linear joint (RRP), as shown in Fig. 101.9. This configuration is significantly different from the spherical configuration, since the axes for all joints are always vertical. In addition to the first three degrees of freedom (DOF), the SCARA robot will often include an additional rotation about the last vertical link to aid in orientation of parts. The work envelope of the SCARA robot is illustrated in Fig. 101.10. The SCARA configuration is the newest of the configurations discussed here, and was developed by Professor Hiroshi Makino of Yamanashi University, Japan. FIGURE 101.8 Articulated robot work envelope. FIGURE 101.9 The SCARA configuration. (Source: T. Owen, Assembly with Robots, Englewood Cliffs, N.J.: Prentice-Hall, 1985. With permission.) X max Y min Y max Z max Z min = 0.0 X min 0? ? 2000 by CRC Press LLC This configuration has many advantages and is quite popular in industry. The configuration was designed specifically for assembly tasks [Truman, 1990], so has distinct advantages when applied in this area. Because of the vertical orientation of the joints, gravity does not affect the dynamics of the first two joints. In fact, for these joints, the actuators can be shut off and the arm will not fall, even without the application of brakes. As the name SCARA implies, this allows compliance in the horizontal directions to be selectively varied; therefore, the robot can comply to horizontal forces. Horizontal compliance is important for vertical assembly operations. Because of the vertical linear joint, straight-line vertical motions are simple. Also, SCARA robots typically have high positional repeatability. The revolute joints allow high-speed motion. On the negative side, the resolution of the arm is not constant throughout the workspace, and the kinematic equations are relatively complex. In addition, the vertical motion of the SCARA configuration is typically quite limited. While the SCARA robot can reach around objects, it cannot reach over them in the same manner as an articulated arm. Gantry Configuration The gantry configuration is geometrically equiva- lent to the Cartesian configuration, but is sus- pended from an overhead crane and typically can be moved over a large workspace. It consists of three linear joints (PPP), and is illustrated in Fig. 101.11. In terms of work envelope, it will have a rectangular volume that sweeps out most of the inner area of the gantry system, with a height lim- ited by the length of the vertical mast, and the headroom above the gantry system. One consid- eration in the selection of a gantry robot is the type of vertical linkage employed in the z axis. A monomast design is more rigid, yielding tighter tolerances for repeatability and accuracy, but requires significant headroom above the gantry to have a large range of z axis motion. On the other hand, a telescoping linkage will require signifi- cantly less headroom but is less rigid, with corre- sponding reduction in repeatability and accuracy. Other robot configurations can be mounted on gantry systems, thus gaining many of the advan- tages of this geometry. FIGURE 101.10 SCARA robot work envelope. Horizontal Reach Work Envelope Swing Horizontal Stroke Horizontal Reach Plan View Elevation Work Envelope Vertical Stroke Vertical Reach FIGURE 101.11The gantry configuration. (Source: T. Owen, Assembly with Robots, Englewood Cliffs, N.J.: Prentice-Hall, 1985. With permission.) ? 2000 by CRC Press LLC Gantry robots have many advantageous properties. They are geometrically simple, like the Cartesian robot, with the corresponding kinematic and dynamic simplicity. For the same reasons, the gantry robot has a constant arm resolution throughout the workspace. The gantry robot has better dynamics than the pedestal-mounted Cartesian robot, as its links are not cantilevered. One major advantage over revolute-base robots is that its dynamics vary much less over the workspace. This leads to less vibration and more even performance than typical pedestal-mounted robots in full extension. Gantry robots are much stiffer than other robot configura- tions, although they are still much less stiff than Numerical Control (NC) machines. The gantry robot can straddle a workstation, or several workstations for a large system, so that one gantry robot can perform the work of several pedestal-mounted robots. As with the Cartesian robot, the gantry robot’s simple geometry is similar to that of an NC machine, so technicians will be more familiar with the system and require less training time. Also, there is no need for special path or trajectory computations. A gantry robot can be programmed directly from a Computer-Aided Design (CAD) system with the appropriate interface, and straight-line motions are particularly simple to program. Large gantry robots have a very high payload capacity. Small, table-top systems can achieve linear speeds of up to 40 in./s (1.025 m/s), with a payload capacity of 5.0 lb (2.26 kg), making them suitable for assembly operations. However, most gantry robot systems are not as precise as other configurations, such as the SCARA configuration. Additionally, it is sometimes more difficult to apply a gantry robot to an existing workstation, as the workpieces must be brought into the gantry’s work envelope, which may be harder to do than for a pedestal-mounted manipulator. Gantry robots can be applied in many areas. They are used in the nuclear power industry to load and unload reactor fuel rods. Gantry robots are also applied to materials-handling tasks, such as parts transfer, machine loading, palletizing, materials transport, and some assembly applications. In addition, gantry robots are used for process applications such as welding, painting, drilling, routing, cutting, milling, inspection, and nonde- structive testing. The gantry robot configuration is the fastest-growing segment of the robotics industry. While gantry robots accounted for less than 5% of the units shipped in 1985, they are projected to account for about 30% of the robots by the end of the 1990s. One reason for this is summed up in Long [1990], which contains much more information on gantry robots in general: Currently gantry robot cells are being set up which allow manufacturers to place a sheet of material in the gantry’s work envelope and begin automatic cutting, trimming, drilling, milling, assembly and finishing operations which completely manufacture a part or subassembly using quick-change tools and pro- grammed subroutines. Additional Information The above six configurations are the main types currently used in industry. However, there are other configu- rations used in either research or specialized applications. Some of these configurations have found limited application in industry and may become more prevalent in the future. All the above configurations are serial-chain manipulators. An alternative to this common approach is the parallel configuration, known as the Stewart platform [Waldron, 1990]. This manipulator consists of two platforms connected by three prismatic linkages. This arrangement yields the full six DOF motion (three position, three orientation) that can be achieved with a six-axis serial configuration but has a comparably very high stiffness. It is used as a motion simulator for pilot training and virtual reality applications. The negative aspects of this configuration are its relatively restricted motion capability and geometric complexity. The above configurations are restricted to a single manipulator arm. There are tasks that are either difficult or impossible to perform with a single arm. With this realization, there has been significant interest in the use of multiple arms to perform coordinated tasks [Bonitz and Hsia, 1996]. Possible applications include carrying loads that exceed the capacity of a single robot, and assembling objects without special fixturing. Multiple arms are particularly useful in zero-gravity environments. While there are significant advantages to the use of multiple robots, the complexity, in terms of kinematics, dynamics, and control, is quite high. However, the use of multiple robots is opening new areas of application for robots. ? 2000 by CRC Press LLC Typical industrial robots have six or fewer DOF. With six DOF, the robot can, within its work envelope, reach arbitrary positions and orientations. At the edge of the work envelope, a six-DOF robot can attain only one orientation. To increase the geometric dexterity of the manipulator, it is useful to consider robots with more than six DOF, i.e., redundant robots. These robots are highly dexterous and can use the extra DOF in many ways: avoidance obstacle, joint torque minimization, kinematic singularity (points where the manipulator cannot move in certain directions) avoidance, bracing strategies where part of the arm is braced against a structure, which raises the lowest structural resonant frequency of the arm, etc. While the redundant manip- ulator configuration has many desirable properties, the geometric complexity has limited their application in industry. For any of the six standard robot configurations, the orientation capability of the major linkages is severely limited. Thus, it is critical to provide additional joints, known as the minor linkages, to provide the capability of varied orientations for a given position. Most robots include a three-DOF revolute joint wrist that is connected to the last link of the major linkages. The three revolute axes will be orthogonal and will usually intersect in a common point, known as the wrist center point. Then, the kinematic equations of the manipulator can be partitioned into locating the Cartesian position of the wrist center point and then determining the orientation of a Cartesian frame fixed to the wrist axes. Conclusions Each of the six standard configurations has specific advantages and disadvantages. When choosing a manipulator for a task, the properties of the manipulator geometry are one of the most important considerations. If the manipulator will be used for a wide variety of tasks, one may need to trade off performance for any given task for the flexibility that will allow the manipulator to work for the various tasks. In such a case, a more flexible geometry should be considered. The future of robotics will be interesting. With the steady increase in compu- tational capabilities, the more complex geometries, including redundant and multiple robots, are beginning to see increased applications in industry. Defining Terms Degrees of freedom: The number of degrees of freedom (DOF) of a manipulator is the number of independent position variables that must be specified in order to locate all parts of the manipulator. For a typical industrial manipulator, the number of joints equals the number of DOF. Kinematics: The kinematics of the manipulator refers to the geometric properties of the manipulator. Forward kinematics is the computation of the Cartesian position and orientation of the robot end-effector given the set of joint coordinates. Inverse kinematics is the computation of the joint coordinates given the Cartesian position and orientation of the end-effector. The inverse kinematic computation may not be possible in closed form, may have no solution, or may have multiple solutions. Redundant manipulator: A redundant manipulator contains more than six DOF. Singularity: A singularity is a location in the workspace of the manipulator at which the robot loses one or more DOF in Cartesian space, i.e., there is some direction (or directions) in Cartesian space along which it is impossible to move the robot end-effector no matter which robot joints are moved. Related Topic 101.2 Dynamics and Control References R.G. Bonitz and T.C. Hsia, “Internal force-based impedance control for cooperating manipulators,” IEEE Transactions on Robotics and Automation, Feb. 1996. J.J. Craig, Introduction to Robotics: Mechanics and Control, Reading, Mass.: Addison-Wesley, 1986. A. J. Critchlow, Introduction to Robotics, New York: Macmillan, 1985. ? 2000 by CRC Press LLC E. Long, “Gantry robots,” in Concise International Encyclopedia of Robotics, R. C. Dorf, Ed., New York: Wiley- Interscience, 1990. R. Truman, “Component assembly onto printed circuit boards,” in Concise International Encyclopedia of Robotics, R. C. Dorf, Ed., New York: Wiley-Interscience, 1990. K. J. Waldron, “Arm design,” in Concise International Encyclopedia of Robotics, R. C. Dorf, Ed., New York: Wiley- Interscience, 1990. Further Information The journal IEEE Transactions on Robotics and Automation is a valuable source for a wide variety of robotics research topics, occasionally including new robot configurations. Additionally, IEEE’s Control Systems Magazine occasionally publishes an issue devoted to robotic systems. The home page for the IEEE Robotics and Auto- mation Society can be found at “http://www.acim.usi.edu/RAS/”. Another journal that often has robotics-related articles is the ASME Journal of Dynamic Systems, Measurement and Control. An additional source of robotics information is The Proceedings of the IEEE International Conference on Robotics and Automation. This conference is held annually. Useful sources on the World Wide Web include the Robotics Internet Resources page, located at “http://piglet.cs. umass.edu:4321/robotics.html”, and Robotics Resources, located at “http://www.eg.bucknell.edu/~robot- ics/rirc.html”. Consult your system administrator for information on this web access. 101.2 Dynamics and Control R. Lal Tummala The primary purpose of the robot control system is to issue commands to joint actuators to faithfully execute a planned trajectory in the tool space. This may involve position control when the manipulator is following a trajectory through free space or a combination of position and force control if the manipulator is to react continuously to contact forces at the tool or end-effector. Control systems can operate either in open loop or closed loop. In open-loop systems, the output has no effect on the input. On the other hand, closed-loop systems continuously sense the output and make appropriate adjustments to the input in order to keep the output at the desired level. The majority of the current industrial robots use the independent joint control method and close the loop around the joints of the robot. The desired joint positions corresponding to a tool trajectory are either taught by using a teach box or generated by solving an inverse kinematics problem. The independent joint control method, however, is effective only at low speeds. As the speeds increase, the coupling effects between the joints increase and warrant the inclusion of these effects in the controller development. Advanced controller devel- opment and implementation based on full dynamics is one of the active areas of current research. New advances in sensor technology, faster computers, advanced robots such as direct drive arms and industrial competition provide new opportunities and motivation for accelerating the development and implementation of advanced controllers for robots in the near future. Independent Joint Control of the Robot The independent joint control method assumes that a single joint of a robot is moving while all the other joints are fixed. A typical joint position control system is shown in Fig. 101.12, where the actuator used is a dc servomotor [Luh, 1983]. In general, any one or a combination of electric motors and hydraulic or pneumatic pistons can be used to move the joint through the desired positions. These motors may be connected directly to the joint or indirectly through gears, chains, cables, or lead screws. The desired joint positions that are inputs to the position loops are obtained from the trajectory planner. The actual position of the joint is obtained by using a position sensor, such as a potentiometer or an optical encoder. An amplifier is used for increasing the system gain, denoted by K a . The velocity feedback K v is used to reinforce the effect of back emf for controlling ? 2000 by CRC Press LLC the damping of the system. This can be done either using a tachometer or computing the difference in angular displacements of the actuator shaft over a fixed time interval. The design of the control system involves fixing the values of K a and K v to achieve the desired response. Consider the closed-loop transfer function of the system shown in Fig. 101.12 (assuming nT l = 0), (101.1) where K a = gain of the amplifier, K T = torque constant of the motor, K b = back emf constant, K q = position sensor constant (volts/rad), R = resistance of the motor winding (ohms), and n = gear ratio. q L = link position (rad) and q m = angular displacement at the actuator side (rad). The effective inertia, J eff , and damping, B eff , are defined as J eff = J m + n 2 J L (101.2) and B eff = B m + n 2 B L (101.3) where J m = total inertia on the motor side, B m = damping coefficient at the motor side, J L = inertia of the robot link, and B L = damping coefficient at the load side. This is a second-order system and stable for all values of K a and K v . The values of K a and K v are selected to achieve a desired transient response by fixing the damping ratio and the natural frequency of the system and are described below. The characteristic equation for the above system is (101.4) This can be conveniently written as (101.5) where the natural frequency w n and the damping ratio z of the system are given as FIGURE 101.12 Closed-loop control of a robot joint. (Source: Adapted from J. Y. S. Luh, “Conventional controller design for industrial robots: A tutorial,” IEEE Trans. Systems, Man, Cybernetics, vol. SMC-13, no. 3, June 1983. ? 1983 IEEE.) q q q q 1 2 () () () s s nK K K s RJ s RB K K K K K nK K K d aT Tb aTv Ta = +++ + eff eff ss RB K K K K K RJ nK K K RJ Tb aTv Ta2 0+ ++ += eff eff eff q ss nn 22 20++=zw w ? 2000 by CRC Press LLC (101.6) (101.7) These systems are designed to operate with critical damping (z = 1) because an underdamped system (z < 1) has fast response but results in an overshoot, whereas an overdamped system (z > 1) is too slow. However, this is not always possible, because the damping ratio given by Eq. (101.7) depends on B eff and J eff which vary during the actual operation of the manipulator. B eff changes with age or repeated use of the manipulator. J eff varies with the payload. For example, the variation of J eff for the Stanford manipulator under various loading conditions is shown in Fig. 101.13. J eff also varies with the configuration of the manipulator during the actual operation. So a compromise solution will be to design the controller such that z 3 1 throughout the intended operation. The undamped natural frequency w n is selected to be no more than half the resonance frequency of the robot to avoid any structural damage to the robot [Paul, 1981]. These resonances are possible due to the flexibilities associated with the links of the robot and the shafts within the drive system, to name a few. These are called unmodeled resonances because they are not explicitly included in the model. In our case, if K eff and J eff are the effective stiffness and the inertias of the joint, respectively, then the resonance frequency w r is given by (101.8) Since K eff is difficult to estimate but constant for a given joint, we can experimentally determine the resonance frequencies for a known inertia and use this information for fixing the gain. Suppose w is the resonance frequency for a given value of effective inertia J, then FIGURE 101.13Variations of link inertias for JPL-Stanford manipulator. (Source: A.K. Bejczy, Jet Propulsion Lab, Pasadena, Calif., American Automatic Control Conference Tutorial Workshop, Washington, D.C., June 18, 1982.) w q n aT nKKK RJ => eff 0 z q = ++RB KK KKK nKKRJK Tb aTv aT eff eff 2 w r K J = eff eff ? 2000 by CRC Press LLC (101.9) To minimize the effects of unmodeled resonances, we use (101.10) The selection of K a and K v depends on selecting z and w n . Using Eqs. (101.6) and (101.10), we can find an upper bound on K a given by (101.11) The upper bound on K v is obtained by setting z 3 1. Using Eq. (101.7), (101.12) Substituting the upper bound for K a from Eq. (101.11), we get (101.13) The steady-state errors to step position commands for the system shown in Fig. 101.12 are zero. However, in the presence of disturbances such as external load torques or gravitational torques, the system will have steady-state errors. For example, if T L is the load torque as shown in Fig. 101.12, the available torque for the joint motion is given by (J eff s 2 + B eff s) q m (s) = T m (s) – nT L (s) (101.14) Using the superposition property, we get q L (s) = F 1 (s)q d (s) + F 2 (s)T L (s) (101.15) where (101.16) W(s) = RJ eff s 2 + (RB eff + K b K T + K v K a K T )s + nK a K T K q Now if T L (s) = C L /s and q d (s) = C q /s, then the steady-state error is w= K J eff w ww n r J J ￡= 22 eff K JR nKK a T ￡ w q 2 4 RB KK KKK nKKRJK Tb aTv aTeff eff ++ 32 q K RJJ RB KK nK JR vTb 3-- ( ) w q eff eff 4 2 w Fs nKKK s Fs nR s aT 1 2 2 () () () () = =- q W W ? 2000 by CRC Press LLC (101.17) Since the value of K a has an upper bound, this error cannot be made arbitrarily small. A possible way to reduce this error is to add a feedforward term, as shown in Fig. 101.14 [Luh, 1983]. The feedforward signal T d (s) is chosen such that the steady-state error is zero. In this case, (101.18) Similar considerations apply for other disturbances such as frictional torques and gravitational torques. Notice from Eq. (101.18) that the feedforward signal is a function of the estimated torque. The burden of determining these torques should not be underestimated. The other factor that was not mentioned earlier is the centrifugal torque, a nonlinear function of velocity. In the case of positioning applications, the velocity tends to zero as t ? ￥. However, if the robot is required to follow a conveyor with constant speed, then the input is a velocity. In this case, the centrifugal contribution will affect the steady-state velocity error. A feed- forward compensation can be used in this case as well. Another method of compensating for the steady-state errors caused by gravitational and load torque disturbances is by adding an integral feedback (PID control), which of course increases the order of the system. The system is no longer stable for all values of the gains and thus adds another constraint in the selection of K a and K v . So far we have considered the control of one joint of the robot while the other joints are fixed. Implementation of this control by successively positioning each joint while the other joints are fixed slows the robot operation and can also result in awkward hand motions, which is undesirable especially when the robot is supposed to follow a continuous path. Simultaneous fast motion of the joints, on the other hand, requires the inclusion of dynamic interactions between the joints. The controllers designed without considering these dynamic interac- tions tend to make the arm move slower and can potentially cause overshoots, oscillations, and path errors. To estimate the dynamic effects, one needs to obtain the equations of motion (dynamic models) of the robot. These equations are, in general, complex and take the form of coupled nonlinear differential equations. Dynamic Models Two of the most popular methods used to obtain dynamic models of the robot are the Newton-Euler method and the Lagrange-Euler method. The equations obtained using Lagrangian formulation are more suitable for the application of modern control theory than the recursive equations obtained using the Newton-Euler method. In the Lagrangian formulation, the dynamic models are obtained using kinetic and potential energies associated with the rigid bodies in motion. The derivation is systematic and conceptually simple. This method yields closed-form dynamic equations that explicitly express joint variables in terms of joint torques. To arrive FIGURE 101.14Feedback compensation method for disturbances. (Source: Adapted from J. Y. S. Luh, “Conventional controller design for industrial robots: A tutorial,” IEEE Trans. Systems, Man, Cybernetics, vol. SMC-13, no. 3, June 1983. ?1983 IEEE.) e nCR KKK ss L aT = q Ts R KKK nTs d TaR L () ? ()= ? 2000 by CRC Press LLC at these equations, one starts with a set of generalized coordinates q i , i = 1, 2, 3, . . .,n, that completely locate the dynamic system and finds the total kinetic energy K and potential energy P of the system [Paul, 1981]. Then the equations of motion are given by (101.19) where T i is the generalized force and L(q,q·) = K – P is the Lagrangian. A simple example is given next to illustrate these ideas. Example. Consider a planar arm with two degrees of freedom, shown in Fig. 101.15. For simplicity, we assume that the masses m 1 and m 2 of the links are represented by point masses at the end of the links. The link lengths are l 1 and l 2 , respectively. The variables q 1 and q 2 are the joint angles. We know that the kinetic energy of a mass m moving at a linear velocity v is given by 1/2 mv 2 and the potential energy associated with a mass m located at a height h in a gravitational field is given by mgh, where g is the gravitational constant. The kinetic energy K 1 for mass m 1 is found by observing that (101.20) Similarly, the kinetic energy K 2 for mass m 2 is given by (101.21) From Fig. 101.15, we have (101.22) FIGURE 101.15Two-degree-of-freedom planar manipulator. d dt L q L q Ti n ii i ? ? ? ? ˙ ,,...-= =for 123 xl yl vxy Kml 111 111 1 2 1 2 1 2 1 1 2 11 2 1 2 = = =+ = cos sin ˙˙ ˙ q q q Kmv vxy 2 1 2 22 2 2 2 2 2 2 2 = =+ ˙˙ x ll yl l vll l 2 11212 2112 12 2 2 1 2 1 2 2 2 12 2 12 2 1 2 12 2 = ++ ++ =+++ + cos cos( ) sin sin( ) ˙ ( ˙˙ ) cos ( ˙˙˙ ) qqq q qqq qqq ? 2000 by CRC Press LLC The potential energies for the masses m i , i = 1, 2, are given by (101.23) The next step is to form the Lagrangian, The dynamic model of the robot is obtained by using Eq. (101.19), (101.24) (101.25) where t i , i = 1, 2, are the joint torques. The equations for a general n-degrees-of-freedom robot can be derived by following the same procedure and are compactly written in the generalized coordinates q as (101.26) where D(q) is the n2n inertia matrix, H(·) is an n21 vector describing the centripetal and Coriolis terms, V is the coefficient of friction and G(q) is an n21 vector describing gravitational torques. For the above example, q 1 = q 1 and q 2 = q 2 . Thus, (101.27) (101.28) (101.29) where (101.30) Pmgl Pmgl l 1111 2211212 = =++ sin [ sin sin( )] q qqq LKP ii i =- = ? 1 2 t ? ?q ? ?q 1 11 = é ? ê ê ù ? ú ú - d dt LL ˙ t ? ?q ? ?q 2 22 = é ? ê ê ù ? ú ú - d dt LL ˙ Dqq Hq q Vq Gq() ˙˙ (, ˙ ) ˙ ()+++=t D lm llm l m m lm llm lm llm lm () cos ( ) cos cos q qq q = ++++ + é ? ê ê ù ? ú ú 2 2 2122 21 2 122 2 2122 2 2 2 2122 2 2 2 2 2 H mll mll mll (, ˙ ) sin ˙ sin ( ˙˙ ) ( sin ) ˙ qq qq q qq qq = -- é ? ê ê ù ? ú ú 212 2 2 2 212 2 1 2 212 2 1 2 2 G ml g m m lg mlg () cos( ) ( ) cos cos( ) q qq q qq = ++ + + é ? ê ê ù ? ú ú 22 2 12 1 21 1 22 1 2 q q q q q q q q q = é ? ê ê ù ? ú ú = é ? ê ê ù ? ú ú = é ? ê ê ù ? ú ú 1 2 1 2 1 2 ˙ ˙ ˙ ˙˙ ˙˙ ˙˙ ? 2000 by CRC Press LLC Notice that the inertia matrix D(q) is a function of only the position q. In general, the inertia matrix is symmetric and positive definite and thus invertible. The diagonal elements of this matrix represent the effective inertias at the respective joints, while the off-diagonal elements represent the coupling inertias. For example, the term m 2 l 2 2 represents the effective inertia at the joint 2, and the term l 2 2 m 2 + l 1 l 2 m 2 cos q 2 represents the coupling inertia between joints 1 and 2, i.e., the effect of acceleration of joint 1 on joint 2. The terms in the matrix H(·) contain all the terms associated with the centripetal and Coriolis forces. The terms that depend upon the square of the joint velocity are centripetal forces. The terms that contain the product of joint velocities are Coriolis forces. In our example, the term –(m 2 l 1 l 2 sin q 2 ) · q 2 2 represents the centripetal force acting at joint 1 due to the velocity at joint 2. Similarly, the term (m 2 l 1 l 2 sin q 2 ) · q 1 2 represents the centripetal force acting at joint 2 due to the velocity at joint 1. The term –(2m 2 l 1 l 2 sin q 2 ) · q 1 · q 2 isthe Coriolis force acting at joint 2 due to the velocities at joints 1 and 2. The term G(q) contains all the terms involving gravitational constant g. Note that these terms depend only on the position of the arm in the gravitational field. If the arm is operating in the gravity-free environment, then these terms become zero. The term V·q reflects the frictional forces present in the robot system. In our example, these terms are assumed to be zero. However, in practical robots a substantial amount of friction stiction can be present that if not considered will overestimate the torque available for accelerating the joints. The above example illustrates that the existence of significant coupling between the joints, if ignored, can cause positioning and tracking errors when the joints are moving simultaneously. However, all these coupling terms become small at low speeds. In this case, independent joint control with appropriate compensations as discussed earlier may be quite adequate. As the operational speeds increase, one needs to take into consideration the full dynamics in the development of control algorithms. Computed Torque Methods Several control algorithms that incorporate dynamics have been developed. Many of these are variations of the computed torque method, which is similar to the feedback linearization method used for the control of nonlinear systems [Spong and Vidyasagar, 1989]. In the computed torque method shown in Fig. 101.16, the required input forces or torques are computed as follows: (101.31) where K v and K p are diagonal matrices with diagonal elements representing velocity and position gains. If this torque is chosen as the input in Eq. (101.26), and assuming that the model is accurate, i.e., D ? (q)= D(q), H ? (q,q · )=H(q,q · )etc., we get (101.32) FIGURE 101.16Computed torque method. Manipulator G(q)K p ? D(q) ? VK v q d q d ? H(q,q) ? q d ++ + – + – ++ + ++ + t q q t= + - + - + + + ? () ˙˙ ( ˙˙ )() ? (, ˙ ) ? ˙ ? ()[]Dqq Kq q Kq q Hqq Vq Gq dvd pd DqqqKqqKqq dvdpd () ˙˙ ˙˙ ( ˙˙ )()[]-+ -+ -=0 ? 2000 by CRC Press LLC Since the inertia matrix, D(q) is nonsingular, we get (101.33) which represents a set of decoupled equations where the error E = q d – q. If we select the values of K v and K p such that the characteristic roots of Eq. (101.33) have negative real parts, then E approaches zero asymptotically. Effectiveness of this algorithm depends heavily on two factors: (1) the accuracy of the model and (2) the ability to compute the coefficient matrices of the equations of motion in real time. If the model is not an exact representation of the system, Eq. (101.33) becomes (101.34) where R is the mismatch between the model and the actual dynamics of the robot. This is given by (101.35) Observe that if the model is an exact match, Eq. (101.34) leads to Eq. (101.33) and the convergence of q to q d can be guaranteed. Even if the model is accurate, the ability to compute the dynamics at sample rate (60 to100 Hz is typical) is still a problem. It is estimated that the Stanford manipulator requires 2000 floating-point additions and 1500 multiplications to compute all joint torques. A way to overcome this problem is to use the control scheme where the model is outside the feedback loop shown in Fig. 101.17, [Craig, 1989]. In this case, the desired torques are calculated a priori using the model given in Eq. (101.26) as follows: (101.36) Then from Fig. 101.17, we get (101.37) If the mismatch between the model and the robot is small, then we get (101.38) FIGURE 101.17Dynamic model outside the feedback loop. Manipulator E K p K v E Dynamic model q d q d q d ++ + + – + +– t t q q ? ˙˙ ˙ EKEKE vp ++=0 ˙˙ ˙ ( ˙˙ , ˙ ,)EKEKERqqq vp ++= Rqqq DqDq Dqq Hqq Hqq Gq Gq( ˙˙ , ˙ ,) ? ()[(() ? ()) ˙˙ ((, ˙ ) ? (, ˙ )) () ? ()] – =-+-+- 1 ? ? () ˙˙ ? (, ˙ ) ? ˙ ? ()t= + + +Dqq Hqq Vq Gq dd dd d d Dq Hqq Vq Gq Kq q Kq q vd pd ˙˙ (, ˙ ) ˙ () ? ( ˙˙ )()+++=+-+-t Dq q Kq q Kq q dvdpd ( ˙˙ ˙˙ )( ˙˙ )()-+ -+ -=0 ? 2000 by CRC Press LLC Since the inertia matrix is nonsingular, we can rewrite the above equation as (101.39) where E = q d – q and can be made to go to zero asymptotically by selecting the gains K v and K p appropriately. This method has a definite advantage over the earlier method because the model need not be evaluated in real time. However, it does not provide complete decoupling because the inertia matrix is not diagonal. Furthermore, since the gains are continuously modified by the inertia matrix, the response is a function of the configuration and the payload. A way to circumvent this problem is to continuously modify the gains K v and K p . This obviously suggests an adaptive control approach. Adaptive Control In an attempt to reduce the errors caused by the mis- match of the model with the real system, several adaptive control schemes have been investigated. Model reference adaptive control (MRAC) is one such approach. Dubowsky and DesForges [1979] were the first to use this method for manipulator control. This method is illustrated in Fig. 101.18. They have chosen a linear sec- ond-order system with desired z and w n as a reference model for each joint. Their scheme works as long as the manipulator changes configuration slowly relative to the adaptation rate. Since then several researchers have extended the concepts well developed for linear systems to manipulator control. Two aspects that are central to all these methods are identification of the plant or its parameters and use of this new information to update the control law. An extensive review of recent work in this area is given by Craig [1988] and Hsia [1986]. In spite of many approaches suggested for this problem, no attempt has been made to implement these methods by the robot industry. Resolved Motion Control So far we have discussed the methods to achieve desired joint motion. In practice, the desired motion is specified in terms of hand motions. Resolved motion control methods such as resolved motion rate control (RMRC) and resolved motion acceleration control have been suggested [papers by Whitney and Luh et al. in Brady et al., 1982]. In these methods, the joint motions are coordinated to achieve coordinated hand motion along any world coordinate axis. Given the relationship between the position and orientation of the hand, x(t), and the joint coordinates q(t) as x(t) = f(q(t)) (101.40) then RMRC transforms the linear and angular velocity of the hand (end effector) to joint velocities using the relationship (101.41) where J(q(t)) is the Jacobian matrix. Using the above equation, the combination of the joint rates for a given hand motion can be obtained. However, special consideration must be given when the inverse of the Jacobian matrix does not exist. This occurs when the dimension of x(t) and q(t) are not the same (robots with redundant ˙˙ ˙ EDKEDKE vp ++= --11 0 FIGURE 101.18Model reference adaptive control. ˙ (()) ˙ qJqtx= -1 ? 2000 by CRC Press LLC degrees of freedom) or when a nonredundant robot loses one or more degrees of freedom in its workspace (singular configurations). Resolved motion acceleration control extends the concepts of RMRC by including desired acceleration of the hand as well. Differentiating Eq. (101.41) twice with respect to time, we get (101.42) where x¨ is the acceleration of the hand and q¨ is the joint acceleration. To reduce the position and orientation errors of the hand to zero, (101.43) By selecting the gains K p and K v we can force the error e(t) = x d (t) – x(t) to zero as before. The desired joint acceleration can be obtained by from Eqs. (101.41) and (101.42), and is given as follows: (101.44) Since the inverse of the Jacobian is involved, this method suffers from drawbacks similar to the RMRC method. Compliant Motion The position control methods described above are not sufficient when the robot has to react continuously to contact forces at the end effector. Consider, for example, a simple operation of sliding a block of wood on a table along a desired path. Pure position control will not work because any small errors orthogonal to the table may result in the block either losing contact with the surface of the table or forcing the block through the table, which can either damage the table or the end effector. To perform this task, we need to control the position in the plane of the table and control force normal to the table. This is called compliant motion control and is required whenever the robot is in contact with its “environment.” To perform the above task, for example, a coordinate system called a compliance frame or constraint frame is defined such that at each instant and along each axis the task can be expressed as a pure position control or pure force control. Suppose we associate a coordinate system with the z-axis normal to the table surface. Then to perform this task, we need to control the position along the x and y directions and force control in the z direction to maintain continuous contact with the table surface. In this case, the position along the z direction is not controlled because one cannot control both position and force in the same direction, just as we cannot control both voltage and current across a resistor. Hence, this framework will provide a natural separation between the axes that need to be position controlled and the axes that need to be force-controlled. This is the idea behind the hybrid position/force control developed by Raibert and Craig [1981]. Another approach for complaint motion control is the imped- ance control [Hogan, 1985]. Impedance control does not control the end-point position or force directly, rather a desired dynamic relationship between position and force (mechanical impedance). For a good comparison between these two approaches, the reader is referred to Asada and Slotine [1986]. In general, compliance motion control is very important whenever the robot is required to make contact with its “environment.” This is true for many assembly tasks. Apart from the active methods of control discussed above, passive methods can be used to introduce the desired compliance. One such passive scheme is the use of a remote center compliance (RCC) device developed at Draper Laboratories. The RCC is a purely mechanical device consisting of a spring with six degrees of freedom that is inserted between the wrist and the end effector. By adjusting the stiffness of the springs, various levels of compliance can be obtained. However, passive methods suffer from lack of programmability achieved through active methods. Active control methods, however, require ˙˙ () ˙˙ ˙ (, ˙ ) ˙ ()xJqqJqqqt=+ ˙˙ () ˙˙ () [ ˙ () ˙ ()] [ () ()]xt xt Kxt xt Kxt xt dvd pd =+ -+ - ˙˙ () () ˙˙ () ˙ () ˙ () () () ˙ (, ˙ ) ˙ () qt J q xt Kxt xt Kxt xt Jqqqt dvd pd = +- ( ) +-- ( ) é ? ê ê ù ? ú ú -1 ? 2000 by CRC Press LLC sensing of contact forces and torques at the end effector. Joint torque sensors, wrist sensors, fingertip tactile sensors, and force pedestals can be used for this purpose. Joint torque sensors, as the name implies, are placed at the joints of the manipulator. If F represent the vector of forces at the end effector, then the corresponding vector of joint torques is obtained by using t =[J(q)] T F, where J(q) is the Jacobian and q are the generalized joint coordinates. Joint torque sensing has some drawbacks. First, to obtain the endpoint forces F, the Jacobian which changes with the configuration has to be inverted in real time. Second, the sensors at the joints not only measure the forces and torques applied at the hand but also those applied at the other points of the manipulator. Wrist sensors are better at reducing this uncertainty because they are placed close to the end effector and below the last powered joint of the manipulator. Several wrist sensors are available commercially with necessary electronics to obtain force/torque measurements at high speeds suitable for real-time force control. Another method for providing information about the gripping forces is by mounting tactile sensors at the fingertips. However, these may not be suitable in situations where high gripping forces are required. Force pedestals are employed when a common platform is used for many tasks. In this case, the platform is instrumented to measure interacting forces and torques. Flexible Manipulators The discussion so far assumed that the links of the robot are rigid. These are designed intentionally to minimize the vibrations. Most of the present-day industrial robots fall into this category. These robots, however, cannot handle objects heavier than about 5% their weight. In contrast, lightweight flexible arms consume less energy, achieve faster speeds, and can potentially perform precision assembly tasks. However, it is not possible to move these arms quickly without the onset of structural vibrations due to inadequate structural damping. Efforts have been underway to increase the damping without substantial increase in weight by using composite materials or actively controlling the vibrations, or both. Defining Terms Centripetal forces: Forces that are present during the robot motion. They depend upon the square of the joint velocities of the robot and tend to reduce the power available from the actuators. Compliant motion: Motion of the manipulator (robot) when it is in contact with its “environment,” such as writing on a chalkboard or assembling parts. Coriolis forces: Forces/torques that depend upon the product of joint velocities. Gravitational torques: Torques that depend upon the position of the robot in the gravitational field. Independent joint control: A method where each joint is controlled as a single input/single output system. The coupling effects due to motion of other joints are either ignored or treated as disturbances. Inverse kinematics: A model that maps end-effector positions and orientations to joint variables. Jacobian of the manipulator: A matrix that maps the joint velocities into end effector velocities. Tool space: Space of a 6 ′ 1 vector representing the positions and orientations of the tool or end effector of the robot. Related Topics 100.2 Dynamic Response ? 101.1 Robot Configuration References H. Asada and J. J. E. Slotine, Robot Analysis and Control, New York: John Wiley & Sons, 1986. M. Brady, J.M. Hollerbach, T.L. Johnson, T. Lozano-Perez, and M.T. Mason, Robot Motion: Planning and Control, Cambridge, Mass.: The MIT Press, 1982. J. J. Craig, Adaptive Control of Mechanical Manipulators, Reading, Mass.: Addison-Wesley, 1988. J. J. Craig, Introduction to Robotics, Reading, Mass.: Addison-Wesley, 1989. S. Dubowsky and D.T. DesForges, “The application of model-referenced adaptive control of robotic manipu- lators,” ASME J. Dyn. Syst. Meas. Control, 1979. ? 2000 by CRC Press LLC K. Fu, R. Gonzalez, and C.S.G. Lee, Robotics: Control, Sensing, Vision, and Intelligence, New York: McGraw-Hill, 1987. T.C. Hsia, “Adaptive control of robot manipulators—a review,” IEEE Conference on Robotics and Automation, San Francisco, 1986. N. Hogan, “Impedance control: An approach to Manipulation, Part I, II, and III” ASME J. Dyn. Sys. Meas. Control, vol. 107, Mar. 1985. A. Koivo, Control of Robotic Manipulators, New York: John Wiley & Sons, 1989. J.Y.S. Luh, “Conventional controller design for industrial robots—a tutorial,” IEEE Trans. Syst., Man and Cybern., vol. SMC-13, no. 3, June 1983. R.P. Paul, Robot Manipulators: Mathematics, Programming and Control, Cambridge, Mass.: The MIT Press, 1981. M. Raibert and J. Craig, “Hybrid position/force control of manipulators,” ASME J. Dyn. Syst. Meas. Control, June 1981. M. W. Spong and M. Vidyasagar, Robot Dynamics and Control, New York: Wiley, 1989. Further Information More information about this subject can be obtained by referring to many of the textbooks available on this subject. These are given in the References. Readers who are interested in current research may refer to several journals published by the Institute of Electrical and Electronics Engineers. In particular, IEEE Transactions on Robotics and Automation, IEEE Transactions on Automatic Control, and IEEE Transactions on Systems, Man and Cybernetics along with the conference proceedings published by the respective societies are useful in this regard. 101.3 Applications Nicholas G. Odrey An important utilization of robotics has traditionally been in manufacturing operations. By their very design and reprogrammable features, robots have enhanced the capabilities for flexibility in automation. Robot appli- cations initially focused on replacing repetitive, boring, and hazardous manual tasks. Such initial applications required minimal control, programming, or sensory capability and have evolved to applications that use enhanced controller designs and sophisticated sensory capability. The first recorded commercial application of an industrial robot was at the Ford Motor Company in 1961 that used a Unimate robot to unload a die-casting machine. Since then, robots have been used in various manufacturing processes, fabrication, and assembly operations. Current issues relate to the degree of integration with the total manufacturing system and to the degree of autonomy and/or complexity one wishes to implement for a robotic system. In potential applications, it is necessary to determine the degree of sophistication that one wishes to implement coupled with a detailed economic analysis. The focus in this section is to present a practical implementation strategy for robots within a manufacturing environment, to review particular applications, and to discuss issues relevant to enhancing robot applications on the manufacturing shop floor. Such issues include sensors and their integration within an intelligent control system, the development of grippers for enhanced dexterity, and integration topics within a flexible cellular manufacturing system. Justification Reprogrammable automated devices such as robots provide the flexible automation capability for modern production systems. To evaluate a potential robotic application within a manufacturing environment, both technical and economic issues must be addressed. Typical technical issues include the choice of the number of degrees of freedom to perform a task, the level of controller and programming complexity, end effector and sensor choices, and degree of integration within the overall production system. Economic issues have typically been addressed from a traditional point of view, but it is important to note that other criteria should also be evaluated before a final decision is made to implement a robotic system. Such criteria may be both quantitative and qualitative. ? 2000 by CRC Press LLC Traditional economic approaches analyze investments and costs to compare alternative projects. Three methods are commonly used: (1) payback period method, (2) equivalent uniform annual costs (EUAC) method, and (3) return on investment (ROI). The payback method balances initial investment cost against net annual cash flow during the life of the project to determine the time required to recoup the investment. Many corporations today require relatively short (1- to 3-year) payback periods to justify an investment. In the current environment with the drive toward shortened product life cycles, it is not unusual to see payback period requirements of no greater than 1 year. The payback technique does not consider the time value of money and should be considered only as a first part attempt at justification. The EUAC and ROI methods consider the time value of money (continuous or discrete compounding) and convert all investments, cash flows, salvage values, and any other revenues and costs into their equivalent uniform annual cash flow over the anticipated life of the project. In the EUAC method, the interest rate is known and set at a minimal acceptable rate of return, whereas the ROI method has the objective to determine the interest rate earned on the investment. Details to such techniques are presented in various engineering economy texts such as those by White et al. [1977] and Thuesen and Fabrycky [1989]. Various more sophisticated approaches have been taken to justify robotic and automated system implemen- tation. Estimates of indirect factors such as taxes, capital gain or losses, variability consideration, and associated expected value analysis along with decision tree analysis and Markovian decision analysis [Michel, 1986] are but a few methods to justify such systems. Other factors to be recognized in robotic justification are that robots are reusable from one project to the next and there is a difference in production rates for a robotic implemen- tation over a manual process. A changeover from a manual method to a robotic implementation would have the potential to affect revenues for any project. Many companies have also developed standard investment analysis forms for an economic evaluation of a proposed robot project. These forms are helpful in displaying costs and savings for a project. Groover et al. [1986] presents one such proposed form and gives several references to examples of forms specifically designed for projects devoted to robotics and related automation areas. The aforementioned techniques are important in performing an economic justification for a proposed robotic installation. Still, in general, there are other issues that should be included in the overall analysis. These issues are of particular importance if one is considering installing a more comprehensive system such as a flexible manufacturing system that may include many robots and automated systems. As noted by Proth and Hillion [1990], these issues give rise to criteria that are both quantitative and qualitative. Quantitative criteria include not only reduced throughput time and work-in-process inventory but also criteria related to increased produc- tivity coupled with fewer resources. Another measurable criterion is the reduction in management and mon- itoring staff as a result of smaller quantities and automatic monitoring by sensors. Quality improvement can also be measured both quantitatively and qualitatively. Qualitative benefits from quality improvement can include increased customer satisfaction, increased competitiveness, simplified production management, and other factors. It should be noted that any benefits and cost reductions for installation of an automated system are difficult to evaluate and reflect a long-term commitment of the corporation. Strategic factors should be incorporated in the overall economic justification process, but they are difficult to access and incorporate due to their inherent complexity. Verk [1990] proposes a general framework that attempts to integrate both qualitative and quantitative factors in an economic justification process. The approach taken is being tested at Cincinnati Milacron and the Mazak Corporation. Implementation Strategies A logical approach is a prerequisite to robotic implementation within a manufacturing firm. The following steps have been proposed by Groover et al. [1986] to implement a robotic system: 1.Initial familiarization with the technology. 2.Plant survey to identify potential applications. 3.Selection of an application(s). 4.Selection of a robot(s) for the application(s). 5.Detailed economic analysis and capital authorization. 6.Plan and engineer the installation. 7.Installation. ? 2000 by CRC Press LLC It should be noted that a particular company may have nuances that could modify the above steps. Also of note is that the underlying issue is systems integration and any robotic application should consider total system impact as well as include the equipment, controllers, sensors, software, and other necessary hardware to have a fully functional and integrated system. Another good source of information on robot implementation is the text by Asfahl [1992]. Critical factors for the introduction of robotics technology within a corporation are management support and production personnel acceptance of the technology. Companies such as General Electric have developed checklists to determine the degree of workforce acceptance. Given that the above two factors are met, a plant survey is conducted to determine suitability for automation or robotic implementation. Two general categories of robot applications may be distinguished: (1) a project for a new plant, or (2) placing a robot project in an existing facility. We focus here on the latter category. General considerations for a robot installation include hazardous, repetitive, or uncomfortable working conditions, difficult handling jobs, or multishift operations. High- and medium-volume production typically has many examples of repetitive operations. It can prove useful to investigate injury (particularly muscular) reports with medical personnel and ergonomics experts to identify potential manual operations that may be alleviated with the aid of robotics or automation. Multishift operations associated with high demand for a product are likely candidates for robot applications. As compared to manual work that typically has a high variable labor cost, a robot substitution would have a high fixed cost which can be distributed over the number of shifts plus a low variable cost. The overall effect of a robot application would then be to reduce the total operating cost. Once potential robot applications are identified, one typically must determine which application is the best to pursue. Economic and technical criteria must both be considered. Usually, a simple application that is easy to integrate into the overall system is a good initial choice. A fundamental rule is to implement any straight- forward application to minimize the risk of failure. The General Electric Company has been successful in choosing robot applications by considering the following technical criteria: ?Operation is simple and repetitive. ?Cycle time for the operation is greater than five seconds. ?Parts can be delivered with the proper POSE (position and orientation). ?Part weight is suitable (typical upper weight limit is 1100 lb). ?No inspection is required for the operation. ?One to two workers can be replaced in a 24-hour period. ?Setups and changeovers are infrequent. A choice of a robot for a selected application can be a very difficult decision. Vendor information, expert opinion, and various sources such as the Robotics Product Database [Flora, 1989] can aid in the selection. Selection needs to consider the appropriate combination of parameters suitable for the application. These parameters or technical features include the degrees of freedom, the type of drive and control system, sensory capability, programming features, accuracy and precision requirements, and load capacity of the selected robot. Various point or weighing schemes can be applied to rate different robot models. The planning and engineering of a robot installation must address many issues, including the operational methods to be employed, workcell design and its control, the choice or design of end effectors and other fixturing and tooling requirements, and sensory and programming requirements. In addition, one needs to focus on safety considerations for the workcell as well as overall systems integration. Computer-aided design (CAD) is very helpful to study potential machine interference and various layout problems as well as estimating various performance parameters. Various commercial CAD software packages exist to analyze such problems. One such example is McAuto’s PLACE System. The study at this stage should consider the basic purpose and function of the planned workcell. Consideration needs to be given to analyzing the cycle time that is basic to determining the production rate. An approach developed by Nof and Lechtman [1982], called Robot Time and Motion (RTM), is useful for analyzing the cycle time of robots. ? 2000 by CRC Press LLC Applications in Manufacturing Robots have proven to be beneficial in many industrial and nonindustrial environments. Here, we focus on applications within a traditional manufacturing (shop floor) setting and, in particular, on applications which fall into the following three broad categories: 1.Material handling and machine loading/unloading 2.Processing 3.Assembly and inspection The discussion that follows is not all-inclusive but rather is intended to present (1) an overview of such applications and (2) a few of the more current topics which are impacting the shop floor, particularly as related to flexible manufacturing systems. In the latter case, such issues include developments in sensor integration, mobility, sensory interactive grippers/hands, and issues pertaining to intelligent machines and robots. An important reference for many if not all robotic topics is the International Encylopedia of Robotics edited by Dorf [1988]. Material Handling and Machine Loading/Unloading Applications in this category pertain to the grasping and movement of a workpart or item from one location to another. General considerations for such applications pertain to the gripper design, distances moved, robot weight capacity, the POSE, and robot-dependent issues pertaining to the configuration, degrees of freedom, accuracy and precision, the controller, and programming features. POSE information is particularly important if there are no sensors (e.g. vision) to provide such information prior to pick-up. Specialized grippers have been designed for various applications in all three of the listed categories [Engelberger, 1980]. Quick-change wrists enabling the robot to change grippers (or tools in processing applications) during the production cycle have also become more common since their introduction [Vranich, 1984], as have multiple grippers mounted turret-like at the end of a robotic arm. Various factors need to be considered in the selection and design of grippers. One such checklist of factors can be found in Groover et al. [1986]. It should be noted that certain applications may require a high degree of accuracy and precision whereas others do not. Higher requirements result in more sophisticated drive mechanisms and controllers with associated increased costs. Material handling applications are typically unsophisticated with minimal control requirements. Two- to four-degrees-of-freedom robots may be sufficient in many tasks. More sophisticated operations such as pallet- izing may require up to six degrees of freedom with stricter control requirements and more programming features. Various criteria that have proved to contribute to the success of material handling and machine load/unload applications can be found in Groover et al. [1986]. In addition, excellent examples and case studies on robotic loading/unloading are given in the text by Asfahl [1992]. Processing Robotic processing applications are considered here to be those applications in which a robot actually performs work on a part and requires that the end effector is a tool. Examples include spot welding electrodes, arc welding, and spray-painting nozzles. The most com- mon robotic applications in manufacturing processes are listed in Table 101.2 [Odrey, 1992a]. Many more processing applications are possible. Spot welding and arc welding represent two major applications of industrial robots. It has been noted that industrial robot usage in welding tasks may be as high as 40% [Ross, 1984]. Spot welding robots have found wide use in automotive assembly lines and have been found to improve weld quality and provide more consistent welds and better repeatability of weld locations. Continuous arc welding is a more difficult application than spot welding. Welding of dissimilar materials, variations in weld joints, dimensional variations from part to part, irregular edges, and gap variations are some of the difficulties encountered in the continuous arc welding processes. Typical arc welding processes include gas metal arc welding (GMAW), shielding metal arc welding (SMAW), i.e., the commonly known “stick” welding, and submerged arc welding (SAW). The most heavily employed TABLE 101.2Most Common Robotic Applications in Manufacturing Processes Spot welding Grinding Continuous arc welding Deburring Spray coating Polishing Drilling Wire brushing Routing Riveting Waterjet cutting Laser machining ? 2000 by CRC Press LLC robotic welding process is GMAW in which a current is passed through a consumable electrode and into a base metal, and a shielding gas (typically CO 2 , argon, or helium) minimizes contamination during melting and solidification. In welding, a worker can compensate automatically by varying welding parameters such as travel speed, deposition rate by current adjustment, weave patterns, and multiple welds where required. Duplicating human welding ability and skill requires that industrial robots have sensor capability and complex programming capability. A wide variety of sensors for robotic arc welding are commercially available and are designed to track the welding seam and provide feedback information for the purpose of guiding the welding path. Two basic categories of sensors exist to provide feedback information: noncontact sensors and contact sensors. Noncontact sensors include arc-sensing systems and machine vision systems. The former, also referred to as a through-the-arc system, uses feedback measurements via the arc itself. Specifically, measurements for feedback may be the current (constant-voltage welding) or the voltage (constant-current welding) obtained by program- ming the robot to perform a weave pattern. The motion results in measurements that are interpreted as vertical and cross-seam position. Adaptive positioning is possible by regulating the arc length (constant-current systems) as irregularities in gaps or edge variations are encountered. Vision systems track the weld seam, and any deviations from the programmed seam path are detected and fed back to the controller for automatic tracking. Single-pass systems detect variations and make corrections in one welding pass. Double-pass systems first do a high-speed scan of the joint to record in memory deviations from the programmed seam path, with actual welding corrections occurring on the second “arc-on” pass. Single-pass systems give the advantages of reduced cycle time and of being able to compensate for thermal distortions during the welding operation. One recent example of a microcomputer-based single-pass system using a welding torch and laser-ranging sensor on a six-axis robot is given by Nayak and Ray [1990]. Their system, dubbed ARTIST for adaptive, real-time, intelligent, seam tracker, has a two-level integrated control system in which the high level contains rule-based heuristics and model-based reasoning to arrive at real-time decisions, whereas the low level enables tracking of a three-dimensional welding seam. It should be noted that arc welding, like many manufacturing processes, is not well enough understood physically that one can formulate an exact mathematical model to describe the process. Attempts to optimize welding schedules for any arc welding process have led to expert systems for such processes [Tonkay and Knott, 1989]. Other examples of such work can be found in publications of the Welding Journal [e.g., Lucas, 1987; Fellers, 1987]. A robotic arc welding cell provides several advantages over manual welding operations. These advantages include higher productivity as measured by “arc-on” time, elimination of worker fatigue, decreased idle time, and improved safety. It is also important to correct upstream production operations to reduce variations. This is best accomplished during the design and installation phase of a robotic welding cell. During this phase, issues to consider include delivery of materials to the cell, fixtures and welding positioners, methods required for the processes, and any production and inventory control problems related to the efficient utilization and operation of the cell. Other processing applications for robot use include spray coating and various machining or cutting opera- tions. Spray coating is a major application in the automotive industry where robots have proven suitable in overcoming various hazards such as fumes, mist, nozzle noise, fire, and possible carcinogenic ingredients. The advantages of robotic spray coating are lower energy consumption, improved consistency of finish, and reduced paint quantities used. To install a robotic painting application, one needs to consider certain manual require- ments. These include continuous-path control to emulate the motion of a human operator, a hydraulic drive system to minimize electrical spark hazards, and manual lead-through programming with multiple program storage capability [Groover, et al., 1986]. Newer schemes have considered geometric modeling, painting mechanics, and robot dynamics to output an optimal trajectory based on CAD data describing the objects [Suh et al., 1991]. The objective of such work is to plan an optimal robot trajectory that gives uniform coating thickness and minimizes coating time. Machining operations utilizing robots typically employ end effectors that are powered spindles attached to the robot wrist. A tool is attached to the spindle to perform the processing operation. Examples of tools would be wire brushes or a grinding wheel. It should be noted that such applications are inherently flexible and have ? 2000 by CRC Press LLC the disadvantage that such operations would be less accurate than a regular machine tool. Finishing operations, such as deburring, have provided excellent opportunities for robotic application. Force control systems have proven particularly useful in regulating the contact force between the tool and the edge of the work to be deburred. One such example for robotic deburring is given by Stepien et al. [1987]. In general, force-torque sensors mounted at the robot wrist have proven extremely useful in many applications in processing and assembly operations. The Lord Corporation and JR3 are two manufacturers of such commercial sensors. Assembly Applications Automated assembly has become a major application for robotics. Assembly applications consider two basic categories: parts mating and parts joining. Parts mating refers to peg-in-hole or hole-on-peg operations, whereas joining operations are concerned not only with mating but also a fastening procedure for the parts. Typical fastening procedures could include powered screwdrivers with self-tapping screws, glues, or similar adhesives. In parts-mating applications, remote center compliance (RCC) devices have proven to be an excellent solution. In general, compliance is necessary for avoiding or minimizing impact forces, for correcting position- ing error, and for allowing relaxation of part tolerances. In choosing an RCC device, the following parameters need to be determined prior to an application: ? Remote center distance (center of compliance). This is the point about which the active forces are at a minimum. The distance is chosen by considering the length of the part and the gripper. ? Axial force capacity. Maximum designed axial force to function properly. ? Compressive stiffness. Should be high enough to withstand any press fitting requirements. ? Lateral stiffness. Refers to force required to deflect RCC perpendicular to direction of insertion. ? Angular stiffness. Relates to forces that rotate the part about the compliant center (also called the cocking stiffness). ? Torsional stiffness. Relates to moments required to rotate a part about the axis of insertion. Other parameters also include the maximum allowable lateral and angular errors as determined by the size of the part and by its design. These errors must be large enough to compensate for errors due to parts, robots, and fixturing. Passive and instrumented (IRCC) devices have been developed for assembly applications. One such device that combines a passive compliance with active control is described by Xu and Paul [1992]. In addition, the SCARA (Selective Compliance Articulated Robot for Assembly) class of robots is stiff vertically but relatively compliant laterally. Many opportunities exist for flexible assembly systems. Many of the issues for such systems have been addressed by Soni [1991]. The reader is also referred to the Design for Robotic Assembly Handbook [Boothroyd and Dewhurst, 1985] for quantitative methods to evaluate a product’s ease of assembly by robots. Carter [1990] presents a method for determining robot assembly task time as derived from tests and industrial experience. Carter also addresses the relationship between product design and robotic assembly cycle time. Some of the current trends in automated assembly include coordinating multiple robots to increase the flexibility and reliability of an assembly cell [Coupes et al., 1989; Zheng and Sias, 1986], interaction with CAD databases to automatically generate assembly plans [Wolter, 1989; Nnaji, 1989] and the application of sensors to automatic assembly systems [Cook, 1991]. Meijer and Jonker [1991] consider an architecture for an intelligent assembly cell and its subsequent implementation. An article by Jarneteg [1990] considers the strategies necessary for developing adaptive assembly systems. Inspection Inspection involves checking of parts, products, and assemblies as a verification of conformation to the spec- ification of the engineering design. With the emphasis on product quality, there is a growing emphasis for 100% inspection. Machine vision systems, robot-manipulated active sensing for inspection, and automatic test equip- ment are being integrated into total inspection systems. Robot application of vision systems include part location, part identification, and bin picking. Machine vision systems for inspection typically perform tasks which include dimensional accuracy checks, flaw detection, and correctness and completeness of an assembled product. Current vision inspection systems are predominantly two-dimensional systems capable of extracting ? 2000 by CRC Press LLC feature information, analyzing such information, and comparing to known patterns previously trained into the system. As documented by Nurre and Hall [1989], various techniques for three-dimensional measurements have also been developed by many researchers. Primary factors to be considered in the design or application of a vision system include the resolution and field of view of the camera, the type of camera, lighting require- ments, and the required throughput of the vision system. Machine vision application can be considered to have three levels of difficulty, namely, that (1) the object can be controlled in both appearance and position, (2) it can be controlled in either appearance or position, or (3) neither can be controlled. The ability to control both position and appearance requires advanced, potentially three-dimensional vision capabilities. The objective in an industrial setting is to lower the level of difficulty involved. It should be noted that inspection is but one category of robotic applications of machine vision. Two other broad categories are identification and visual servoing and navigation. In the latter case, the purpose of the vision system is to direct the motion of the robot based on visual input. The reader is directed to Groover et al. [1986] for further details. Emerging Issues Robotics, by definition, is a highly multidisciplinary field. Applications are broad, and even those applications focused on the manufacturing shop floor are too numerous to cover in full here. The reader is referred to the various journals published by the IEEE and other societies and publishers, a few of which have been listed in the references. Still, it is worthwhile to note a few issues relevant to manufacturing shop-floor applications that could have an impact over the next decade. These issues include gripper development, mobility, and intelligent robots. The objective of this work is the overall integration of a flexible manufacturing system. In a manufacturing process or assembly operation, the actions required of a gripper will vary with the task. Much work has been done in developing multifigured hands such as the Utah-MIT hand, the Salisbury hand, and others with an increasing interest of adding tactile sensory input for dexterous manipulation [Allen et al., 1989]. As noted by Allen and his colleagues, robotic systems need to process multiple source data and be easily programmable for grasping and manipulation tasks. One study focused on capturing a machinist’s skill in working with parts and tools and codifying this knowledge in a grip taxonomy has been done by Cutkosky and Wright [1986]. Their study suggests some general principles for the design, construction, and control of hands in a manufacturing (particularly machining) environment. The reader is also referred to the work of Feddema and Ahmad [1986] for the development of an algorithm for a static robot grasp for automated assembly and the work of Cutkosky [1991] on robotic grasping and manipulation. This latter work considers dynamic contact and the application of dynamic tactile sensors in manipulation tasks. An application to identify and locate circuit board fixtures within a robotic workcell that integrates a vision system with a tactile probe is given by DeMeter and Deisenroth [1987]. Automated guided vehicles (AGVs) currently dominate the movement of parts through a flexible manufac- turing system (FMS). AGVs typically restrict the path to predetermined routes and subsequently decrease the “flexibility” of the system. Work is being done on mobile robots to address this issue. Research by Arkin and Murphy [1990] focuses on intelligent mobility within a manufacturing environment. The reader is also referred to the research of Wiens and Black [1992] who address a mobile robot system within a manufacturing cell as a means to increase the flexibility, capability, and capacity of a robot-based manufacturing cell. The issues involved with intelligent robots have been surveyed by Nitzan [1985], where he notes that future proliferation of robotic applications will depend strongly on machine (robotic) intelligence. Such applications will lead to a greater diversity of applications and will not be just manufacturing oriented. The reader is also referred to work on intelligent machines by Weisbin [1986]. It should be noted that particular interest has been directed toward integration of multiple sensors as a means to enhance robot intelligence [Luo and Lin, 1989; Pin et al., 1991]. The text by Klafter et al. [1989] categorizes the major sensory needs for robotic tasks and gives valuable insights to current and future robotics applications. Intelligent control systems, particularly hierar- chical control systems, are being developed by many organizations and research institutes [Odrey, 1992b]. Such systems are expected to have an impact both at the shop-floor level and the management levels of production facilities well into the next century. ? 2000 by CRC Press LLC ? 2000 by CRC Press LLC ROBOTIC TOOLS obotics and Automation Corpora- tion, Minneapolis, Minnesota, manu- factures equipment for robotic systems, in particular a variety of tools known as “end effectors”, devices attached to the end of a robot arm for picking up, grasping, manipulating, and transferring objects. Among the company’s newer products is the Automatic Robotics Tool-change System (ARTS), a system designed to meet the grow- ing demand for multiple task work cells for welding and plasma spray functions that require grinding and finishing; deburring, deflashing, routing, hole drilling or parts replacement; and multiple tool disk opera- tions. The ARTS systems were designed to work with the company’s CFD (Constant/con- trolled Force Device) product line, a series of end effectors and bench mounted devices for controlling the constant pressure of abrasive tools used to deburr, grind, polish, and finish products fabricated by welding, casting, mold- ing, forging, or machining. Robotics and Automation Corporation’s CFD line includes three end-of-arm devices and two bench-mounted devices. They do not require that the robot apply and control the force, only that it move along a normal pro- grammed path over the work piece; the CFD applies and maintains the required processing pressure. When the surface to be finished is very rough and course, several different grades of finishing media may be needed, as well as dif- ferent speeds and power as the surface finish is transformed. To accommodate this multi- step process within a single work cell, and with a single robot, Robotics and Automation Cor- poration developed the automated tool- change system. The ARTS-I is being used in industrial applications with six tool positions ranging from coarse sanding disks and abrasive wheels to cloth polishing wheels with motors of various horsepower. The ARTS-II allows a robot to exchange a welding torch for a CFD end effector to finish a welded assembly with a welding robot; using a second tool-changer (ARTS-I) enables finishing the surface conditioning process. (Courtesy of National Aeronautics and Space Administration.) R The tool rack of the Automatic Robotics Tool-Change System includes a two-finger gripper; a grinder, a coated abrasive brush, and a welding torch. (Photo courtesy of National Aero- nautics and Space Administration.) The quick disconnect system allows changing tools with hydraulic, pneumatic, or electric power. (Photo courtesy of National Aeronautics and Space Administration.) Defining Terms Cellular manufacturing: Grouping of parts by design and/or processing similarities such that the group (family) is manufactured on a subset of machines which constitute a cell necessary for the group’s production. Decision tree analysis: Decomposing a problem into alternatives represented by branches where nodes (branch intersections) represent a decision point or chance event having probabilistic outcome. Analysis consists of calculating expected values associated with the chain of events leading to the various outcomes. Degrees of freedom: The total number of individual motions typically associated with a machine tool or robot. Intelligent control: A sensory-interactive control structure incorporating cognitive characteristics that can include artificial intelligence techniques and contain knowledge-based constructs to emulate learning behavior with an overall capacity for performance and/or parameter adaptation. Machine interference: The idle time experienced by any one machine in a multiple-machine system that is being serviced by an operator (or robot) and is typically measured as a percentage of the total idle time of all the machines in the system to the operator (or robot) cycle time. Sensor fusion: Combining of multiple sources of sensory information into one representational format. Sensor integration: The synergistic use of multiple sources of sensory information to assist in the accom- plishment of a task. Related Topic 112.1 Introduction References P.K. Allen, P. Michelman, and K.S. Roberts, “An integrated system for dextrous manipulation,” IEEE Interna- tional Conference on Robotics and Automation, 1989, pp. 612–616. R.C. Arkin and R.R. Murphy, “Autonomous navigation in a manufacturing environment,” IEEE Trans. Robotics Autom., vol. 6, no. 4, pp. 445–454, 1990. C.R. Asfahl, Robots and Manufacturing Automation, New York: Wiley, 1992. G. Boothroyd and P. Dewhurst, “Design for Robotic Assembly,” Department of Industrial and Manufacturing Engineering, University of Rhode Island, Kingston, 1985. P.W. Carter, “Estimating cycle time in design for robotic assembly,” J. Manu. Syst., vol. 9, no. 1, pp. 1–12, 1990. J.W. Cook, “Applying sensors to automatic assembly systems,” IEEE Trans. Ind. Appl., vol. 27, no. 2, pp. 282–285, 1991. D. Coupes, A. Delchambre, and P. Gaspart, “The supervision and management of a two robots flexible assembly cell,” Proceedings of IEEE Conference on Robotics and Automation, 1989, pp. 540–550. M.R. Cutkosky, “Robotic grasping and manipulation,” Proceedings of NSF Design and Manufacturing Systems Conference, Dearborn, Mich.: Society of Manufacturing Engineers, 1991, pp. 423–430. M.R. Cutkosky and P.K. Wright, “Modeling manufacturing grips and correlations with the design of robotic hands,” IEEE International Conference on Robotics and Automation, San Francisco, Calif., April 7–10, 1986, pp. 1533–1539. E.C. DeMeter and M.P. Deisenroth, “The integration of visual and tactile sensing for the definition of regions within a robot workcell,” Robots 11/17th ISIR, Chicago, Il., April 26–30, 1987, pp. 10-51 to 10-61. R.C. Dorf, Ed., International Encyclopedia of Robotics, vols. 1–3, New York: Wiley, 1988. J.F. Engelberger, “Robotics in practice,” AMA COM: A Division of American Management Associations, 1980. J.T. Feddema and S. Ahmad, “Determining a static robot grasp for automated assembly,” IEEE International Conference on Robotics and Automation, San Francisco, Calif., April 7–10, 1986, pp. 918–924. K.G. Fellers, “A PC approach to welding variables,” Weld. J., vol. 66, pp. 31–40, 1987. P.C. Flora, Ed., Robotics Product Database, 6th ed., Orlando, Fla.: TecSpec, 1989. M.P. Groover, M. Weiss, R.N. Nagel, and N.G. Odrey, Industrial Robotics: Technology, Programming, and Appli- cations, New York: McGraw-Hill, 1986. ? 2000 by CRC Press LLC B.G. Jarneteg, “FAS control strategies for adaptive assembly systems,” 21st CIRP International Seminar on Manufacturing Systems, Stockholm, Sweden, 1990. R.D. Klafter, T.A. Chmielewski, and M. Negin, Robotic Engineering: An Integrated Approach, Englewood Cliffs, N.J.: Prentice-Hall, 1989. W. Lucas, “Microcomputer systems, software and expert systems for welding engineering,” Weld. J., vol. 66, pp. 19–30, 1987. R.C. Luo and M.-H. Lin, “Intelligent robot multi-sensor data fusion for flexible manufacturing systems,” Proceedings of NSF 15th Conference on Production Research and Technology, University of California- Berkeley, Jan. 9–13, 1989, pp. 73–85. B.R. Meijer and P.P. Jonker, “The architecture and philosophy of the DIAC (Delft Intelligent Assembly Cell),” IEEE Conference on Robotics and Automation, Sacramento, Calif., 1991, pp. 2218–2223. M. Michel, “Justification models for flexible manufacturing,” Robots’ 10 Conference Proceedings, Dearborn, Mich.: Society of Manufacturing Engineers, 1986, pp. 2-55 to 2-81. N. Nayak and A. Ray, “An integrated system for intelligent seam tracking in robotic welding: part 1—conceptual and analytical development; part 2—design and implementation,” IEEE International Conference on Robotics and Automation, 1990. D. Nitzan, “Development of intelligent robots: achievements and issues,” IEEE J. Robotics Autom. vol. RA-1, no. 1, pp. 3–13, 1985. B.O. Nnaji, “RALPH: An automatic robot assembly language programmer: an overview,” Proceedings of Robots 13 Conference, Gaithersburg, Md., May 7–11, 1989, pp. 16-41 to 16-63. S.Y. Nof and H. Lechtman, “The RTM method of analyzing robot work,” Ind. Eng., April 1982, pp. 38–48. J.H. Nurre and E.L. Hall, “Three dimensional vision for automated inspection,” Proceedings of Robots 13 Conference, Gaithersburg, Md., May 7–11, 1989, pp. 16-1 to 16-11. N.G. Odrey, “Robotics and automation,” Maynard’s Industrial Engineering Handbook, 4th ed., W.K. Hodson, Ed., New York: McGraw-Hill, 1992a. N.G. Odrey, “Control systems,” 1992 McGraw-Hill Yearbook of Science and Technology, New York: McGraw-Hill, 1992b, pp. 87–90. F.G. Pin et al., “Robotic learning from distributed sensory sources,” IEEE Trans. Syst. Man and Cybern., vol. 21, no. 5, pp. 1216–1223, 1991. J.M. Proth and H.P. Hillion, Mathematical Tools in Production Management, New York: Plenum Press, 1990. B. Ross, “Machines that can see: here comes a new generation,” Bus. Week., January 1984, p. 118. A.H. Soni, “Flexible assembly systems: Opportunities and challenges,” Proceedings of the 1991 NSF Design and Manufacturing Systems Conference, University of Texas at Austin, Jan. 9–11, 1991, pp. 367–373. T.M. Stepien, L.M. Sweet, M.C. Good, and M. Tomizuka, “Control of tool/workpiece contact force with application of robotic deburring,” IEEE J. Robotics Autom., vol. RA-3, no. 1, pp. 7–18, 1987. S.-H. Suh, I.-K. Woo, and S.-K. Noh, “Automatic trajectory planning system (ATPS) for spray painting robots,” J. Manu. Syst., vol. 10, no. 5, pp. 396–406, 1991. G.J. Thuesen and W.J. Fabrycky, Engineering Economy, 7th ed., Englewood Cliffs, N.J.: Prentice-Hall, 1989. G.L. Tonkay and K. Knott, “Intelligent process specification for robotic arc welding,” Proceedings of World Conference on Robotics Research: The Next Five Years and Beyond. Robotics International of the Society of Manufacturing Engineers, Gaithersburg, Md., May 11–17, 1989. S. Verk, “Strategic optimization cycle as a competitive tool for economic justification of advanced manufacturing systems,” J. Manu. Syst., vol. 9, no. 3, pp. 194–205, 1990. J.M. Vranich, “Quick change system for robots,” SME Paper MS84-418, Conference on Robotics Research—The Next Five Years and Beyond, Lehigh University, Bethlehem, 1984. C.R. Weisbin, “CESAR research in intelligent machines,” SME Paper MS586-772, Robotics Research Conference, Scottsdale, Ariz., Aug. 18-21, 1986. J.A. White, M.H. Agee, and K.E. Case, Principles of Engineering Economic Analysis, New York: John Wiley & Sons, 1977. G.J. Wiens and J.T. Black, “Design for mobility within a manufacturing cell,” Proceedings of the NSF Design and Manufacturing Systems Conference, Georgia Institute of Technology, Jan. 8–10, 1992, pp. 1147–1150. ? 2000 by CRC Press LLC J.D. Wolter, “On the automatic generation of assembly plans,” Proceedings of the 1989 IEEE Conference on Robotics and Automation, 1989, pp. 62–68. Y. Xu and R.P. Paul, “Robotic instrumented compliant wrist,” ASME J. Eng. for Ind., vol. 114, pp. 120–123, 1992. Y.F. Zheng and F.R. Sias, Jr., “Two robot arms in assembly,” IEEE Conference on Robotics and Automation, San Francisco, Calif., April 7–10, 1986, pp. 1230–1235. Further Information Various journals publish on topics pertaining to robots. Sources include the bimonthly IEEE Journal of Robotics and Automation, the quarterly journal Robotics and Computer-Integrated Manufacturing (published by Pergamon Press), Robotics (published by Cambridge University Press since 1983), and the Journal of Robotic Systems (published by Wiley). IEEE has sponsored since 1984 the annual “International Conference on Robotics and Automation.” IEEE conference proceedings and journals are available from the IEEE Service Center, Piscataway, N.J. The Society of Manufacturing Engineers (SME) is another source for robot publications that are concerned with both research issues and applications. Robots 1 through 13 (1989) conference proceedings are available as well as the Robot Research conference proceedings (three to date) of Robotics International (RI) of SME. A directory of robot research laboratories is also available. Contact SME, Dearborn, Mich. The three-volume International Encyclopedia of Robotics: Applications and Automation (R.C. Dorf, ed.), published by Wiley (1988), brings together the various interrelated fields constituting robotics and provides a comprehensive reference. ? 2000 by CRC Press LLC