A Versatile Hand for Manipulators

Transcription

1 A Versatile Hand for Manipulators Paolo Fiorini ABSTRACT: This paper describes current research at the Jet Propulsion Laboratory on a versatile mechanical hand for robotic arms. This mechanical hand is a self-contained, autonomous system capable of executing high-level commands from a supervisory computer. The mechanism consists of parallel fingers powered by a dc motor and controlled by a microcomputer embedded in the hand housing. Sensors are integrated in the hand for measuring grasp force of fingers and forces and torques applied between the arm and the surrounding environment. Fingers can be moved in position, velocity, or force control modes, and all collected data are transmitted to the supervisory computer. The hand described in this paper represents a new development in the area of end-effector design because of its multifunctionality and its autonomy. Data about performance and effectiveness in teleoperation experiments are also presented. Introduction Sensor-based, computer-controlled end-effectors for mechanical arms are receiving more attention in robotic industries and research laboratories because current grippers are adequate only for simple pick-and-place tasks and do not have capabilities necessary for complex tasks. For this reason, innovative mechanical designs, new sensors, and advanced computer control need to be integrated into design to achieve better performance and a higher degree of smartness. The complexity of the design is increased because the hand must be easily integrable with other robotic components, and because of the operator interface for direct remote control of the device (teleoperation). In fact, in critical activities such as space use of robotic systems, an operator is required to supervise or substitute for the robotic controller during the evolution of the task. In these activities, there must be a natural way to give new commands to the hand, and a complex input device must be developed to transfer An early version of this paper was presented at the 1987 International Conference on Industrial Electronics, Control, and Instrumentation, Cambridge, Massachusetts, November 3-6, Pa- 010 Fiorini is with the Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Drive, Pasadena, CA the operator s hand and finger motions to the robotic end-effector. At present, most advanced end-effectors are primarily research tools that have not been integrated with complete robotic or teleoperated systems. Therefore, despite the many existing designs, an incremental approach to the design of a mechanical hand has been chosen because it permits a shorter development time and a rapid integration with other robotic components in different experimental test beds. The industrj-standard, one-degree-of-freedom gripper has been modified and upgraded to the current family of smart hands used at the Jet Propulsion Laboratory (JPL). This approach permits the use of a well-known mechanical technology, such as gear and spindle transmission, an embedded microcomputer to make the end-effector self-contained and portable, and a number of sensors to acquire dynamic data of the end-effector. The device has been integrated successfully in the advanced teleoperation system at JPL, and its features have allowed extensive experiments and evaluation. The name smart hand is used with this family of end-effector because it is a self-contained system with advanced features. It is provided, in fact, with some dexterity and with a set of control and communication options that can be selected by the on-board processor. Background In the past few years, much effort has been put into the design and test of advanced endeffectors for mechanical manipulators. Several research programs are under way by different organizations to reach the goal of an effective manipulating device, with dexterity and sensing comparable to those of a human hand. In the large spectrum of advanced designs that have been proposed, a few major directions can be discerned. Some researchers have given an anthropomorphic structure to their robotic end-effector. Others prefer a modular design in which the hand can be reconfigured as a function of the required task. Other groups have chosen to emphasize the postural knowledge necessary in manipulative tasks, while still other laboratories have decided to investigate the hand as a part of complete robotic systems. Some of the anthropomorphic hands presently studied are described in [1]-(61. Typ- ical of this approach is the large number of active degrees of freedom that have to be controlled and, therefore, the complexity of both hardware and software. Modular hands [7]-[9] are a very attractive solution because of their intrinsic reconfigurability ; however, their development is still continuing, and they seem more suited for recognition or exploration tasks rather than manipulation. The complexity of fully controllable anthropomorphic hands can be reduced by employing both active and passive degrees of freedom [IO]. In this case, passive degrees of freedom are not controlled directly but move according to some relaxation law, as in the human hand. In addition to their complexity, these previous designs do not have a corresponding input device to control the hand in teleoperation. Integrable end-effectors are described in [11]-[17]; they use a more mature technology that makes them reliable enough to conduct system-level experiments. Input devices such as joysticks or triggers easily control one-degree-of-freedom motion and, therefore, can be used immediately with a hand where finger motion is in one direction only. In the context of the research in robotics and teleoperation in our laboratory, the design of the end-effector takes into consideration the complete system [ 181 and does not overtax its development. For these reasons, all the members of the family of mechanical hands built to date share the same basic, albeit evolving, approach of a parallel finger gripper with local sensory and processing capability. Recently, a smart hand was designed and tested for NASA Marshall Space Flight Center for potential use in the future Orbital Maneuvering Vehicle system [ 191. The end-effector described in this paper has been designed to accommodate the smallpayload requirement of the Puma 560 mechanical arm used in our laboratory for experiments on teleoperation. A similar endeffector also has been built for NASA Goddard Space Flight Center for their Flight Telerobot Servicer ground-test facility. JPL Smart Hand This paper describes the smart hand (Fig. 1) designed for a small mechanical arm such as the Puma 560. The structure of the hand can be divided into three parts: the mechanism, the sensors, and the electronics. The / $01 00 (c 1988 IEEE 20 I Control Systrrns Magoiine

2 Fig. 1. Partially assembled smart hand, from left to right, the parts are: electronics housing, forcekorque sensor, and finger mechanism. mechanism consists of a parallel finger gripper powered by a dc motor through gears and recirculating ball spindles. Fingers move on rails supported by linear bearings to minimize the effect of friction. Each finger consists of three parts: (1) a moving support, (2) a grasp force sensor on top of it, and (3) an interchangeable part, the fingertip, which can be application dependent. Behind the fingers structure and base there is the force/torque sensor, which consists of a maltese cross structure instrumented with strain gauges [ 131. Strain-gauge readings are acquired by the local microprocessor, formatted, and transmitted to the supervisory computer, where the manipulator control program is executed. The local electronics is housed in the bottom shell and consists of two printed circuit boards, one for the digital electronics and one for the analog input-output. The microprocessor selected in the current design is the Intel A model of the mechanism has been developed that takes into account some of the nonlinearities affecting the system. This model has been used to design the controller for the fingers motion in the firmware of the hand. This embedded software has a multitasking architecture, handling a sequence of intermpts. The background task consists of the processing of input-output messages to and from the supervisory computer. Foreground tasks include control loops, data acquisition routines, and communication drivers. A real-time clock at 500 Hz drives the control loops and the data acquisition routines. A small degree of autonomy is implemented in the control software. In the case when the fingers hit an unexpected object, a protective action takes place. A grasp force limit can be specified by the supervisory computer, or the operator, such that excessive grasp force will automatically switch the finger control to force mode, and the specified force limit will be sustained. Mechanism and Sensor Description This end-effector consists of two parallel fingers powered by a dc brush motor through a gear train and a recirculating ball bearing lead screw (Fig. 2). A six-axis forceltorque sensor is located below the finger assembly for reading forces and torques induced by the interaction with the environment. The controlling electronics is located under the sensor, in a bell-shaped housing, on which the connector to the Puma arm is mounted. The gear system has a reduction ratio of 4 and the linear spindle has a lead of The fingers move on linear bearings powered by the spindle, and the force sensors and fingertips can be replaced to adapt to different tasks. The removable part of the fingers is shaped to permit a solid grasp of small objects and alignment with the central axis of the hand. This feature provides for a limited amount of dexterity in grasping objects and simplifies telemanipulation. The grasp force sensor consists of a square beam instrumented with a strain-gauge bridge, and it senses the squeeze force applied by the fingers to an object. Each finger has an independent sensor so that off-center objects can be detected. The hand is designed to generate 20 Ib of force at the fingers, and the grasp force sensors have been designed for the range of 0-30 lb. The position of the fingers is measured by a linear potentiometer, and a tachometer mounted on the motor shaft measures motor velocity. The forceltorque sensor has a maltese cross design dimensioned for 15 Ib of force and Fig. 2. Mechanical structure of smart hand. 120 in.-lb of torque. This sensor physically forms the interface between the fingers base and the electronics housing. All forces and torques applied by the manipulator to the external world, and vice versa, are transmitted by the sensor s beams, equipped with straingauge bridges. The housing for the electronics holds the printed circuit boards and connects the mechanical arm to the finger mechanism. The weight of the end-effector is approximately 3.5 Ib, with an envelope of 4.5 x 9 in. Modeling Equations The hand system can be described by the following set of equations [20], relating electrical and mechanical parameters, where B, is the motor viscous damping coefficient; B, the tachometer viscous damping coefficient; c, the constant relating torque to force in the spindle; J, the motor moment of inertia; J, the tachometer moment of inertia; K the stiffness of the fingers; K, the motor voltage constant; K, the motor torque constant; i the winding current; L the motor inductance; M the mass of both fingers; N the gear ratio; R the windings resistance; Tf the nonlinear friction torque; T, the torque generated by the motor; Tp the torque at the pinion; V the applied voltage to the motor; x the finger position; xo the object position; p the unity step function; and 8, the motor shaft position. L di(t)ldr + Ri(r) + K, do,(t)/dt = V(t) (1) Forceitorque sensor Electronics boards October

3 T,,,(O = KAt) (2) (J,,, + J,) d20,(t)/dt2 + (B,,, + BO. do,(t)/dt + Tf(t) - T'(t) = T,,,(t) (3) Position M d2x/dt2 + K(x - xo) p(x0 - X ) = c,ntp (4) Let TI be the load torque defined from the following equation, and let the position of the finger be derived from the motor angular position, where L, is the lead of the spindle. Fl = K(x -,YO) /L(x~ - X) Ti = F1/crN x = -[L,/(~TN)]O,~ M(c,N)-' d2x/dt2 + T, = Tp The preceding equations can be combined to give [ J,, + Jl + ML,(~TC,N~)-'] d20,,/d2t + (B, + B,) do,,,/dt = T,,, - + TI (5) Equations (2) and (5) can be used to obtain the transfer-loop functions relating the input voltage to the motor (V) to the variables of interest-namely, finger position x, finger velocity dx/dt, and clamp force F,. The corresponding equations have been derived and the expressions follow, where the new symbols are V,, the commanded motor voltage; Vo, output of the feedback potentiometer; and V,, the output of the motor tachometer. Vo(s)/V,(s) = -62.l/[~(~ )] (6) V,(s)/V,.(s) = 68.36/(s ) (7) F, (s)lv,.(s) = /(s s ) (8) Controller Design and Simulation Results The digital controller implemented in this end-effector is a classical design, where the system's control modes have been kept separate and implemented as position, rate, and force control. The design of the controllers has been based on the transfer function equations [(6)-(8)], and their performance has been simulated on the nonlinear model of the gripper. Figure 3 represents the complete block diagram of the system, while Fig. 4 shows the simulation of the impact force generated by the fingers closing on an object. The fingers are moving toward the object at Fig. 3. Block diagram of the control loops: position, rate, and force. The block determining the mode selection is commanded by the remote host or by a change in grasp force in./sec, with a force set point of $I oz. Upon contact, the impact force is approximately 120 oz, corresponding to an overshoot of 60 percent. Electronics and Software The electronic system is built around the Intel 8097 microcontroller, which permits an effective management of a real-time multitasking environment, because of its high number of built-in functions and its many input-output lines. The electronics is subdivided into two boards, the first one contains microprocessor and program memory, and the second contains analog circuits for data acquisition, dc/ dc power supply, and motor driver. The electronics occupies a total surface area of approximately 35 in.*, with a volume of ap- N' P I I '7 l- a $ i I o Fig. 4. Simulation of the impact force of the fingers closing on an object. proximately 43 in.3. The eight lines of the on-chip analog-to-digital converter are multiplexed to read 14 input signals: ten strain gauges for grasp and forcehorque readings and four for the fingers' data and set point. The communication channel consists of a full-duplex serial line transmitting at a baud rate, and power to the motor is controlled by using the on-chip pulse width modulated generator. The two circuit boards form a completely self-contained system, which can interface to any robotic arm by means of a four-wire connector, and do not require any intermediate amplification unit or any dedicated board in the arm controller. The software system consists of a background process for message analysis and generation, and interrupt-driven routines for the real-time functions of the controller. The background task cames out the syntactic analysis of the incoming messages and updates the operating mode and set point for the control algorithm. It also creates the outgoing messages by formatting fingers and forcekorque data, according to the transmission protocol. The message format has been designed for variable-length messages, error control with a 2-byte check-sum and message tag, and a continuous data stream from the hand to the supervisor computer. The hand receives commands and set points while sending back control-mode information and sensor data. The transmission can operate in a normal mode, with a message stream consisting of fingers' data alternating with force/ torque data. A high-speed transmission mode is also supported, in which the transmission consists of forcekorque messages only, with a simplified packet format. This mode is used 22 I Control Systems Mogorrne

4 when sensor data are used to generate force feedback information for the operator. The real-time tasks are the communication driver and the control algorithms, both activated by interrupts generated internally by the built-in serial communication port and the real-time clock. The communication module does the synchronization between the hand and the host computer, the low-level processing of the input-output packets, and the acquisition of the force/torque data. This last function has been attached to this module to assure that forcekorque data are collected just before their transmission, to reduce the delay of the data for the host computer to the transmission time only. The real-time clock in the microprocessor generates an interrupt every 2 msec, which activates the control module. This routine acquires position and velocity of the fingers, decodes the control mode, and executes the corresponding control algorithm. Three separate control loops have been implemented, one for each of the control modes: position, rate, and force. The switching between modes is done under control of the supervisory computer by means of a command message, or autonomously when the onboard computer detects a force on the fingers larger than the preset limit. Application of the JPL Smart Hand This end-effector has been incorporated into the Advanced Teleoperator Research System at JPL and is currently used in experimental evaluation studies of teleoperation. This system [21] consists of two separate locations connected via a parallel interface: a command site for the operator and a remote site for the manipulator. The command site consists of the operator's interface, composed of a six-degree-of-freedom universal controller for mechanical arms capable of force reflection and gripper control, and of several TV monitors. The operator is presented with different views of the remote site, with a display of the forces and torques generated during the execution of a task, and with a computer graphic simulation of the manipulator to help compensate for delays between commands to the remote site, and their execution. The remote site consists of the manipulator arm, a Puma 560, the smart hand, and three TV cameras in different viewing angles. Each side has a multiprocessor computer to servo the manipulator and its controller in position and force and to handle the flow of information in both directions (Fig. 5). Data from the remote site consist Fig. 5. Task board, manipulator, and smart hand during experiments in teleoperation. of the position of the manipulator end-effector in the operational space coordinate system, of force and torque readings at the endeffector, and of the position and forces of the end-effector fingers. Incremental set points are sent from the operator's site to position the arm and to command the opening and closing of the fingers. The integration of the smart hand with the teleoperation system required that the transmission rate from the smart hand matched the requirements for robust force control on the operator's site. With the high-speed transmission protocol, a bandwidth of 125 Hz has been achieved on the force and torque data, which, combined with transmission and processing delays, allows for a gain of 0.1 in the force reflection loop. This global performance has proven to be quite adequate for telemanipulation tasks, and experiments are currently characterizing the teleoperation system. Experimental Data The smart hand has been characterized in terms of its frequency response in the three control modes of position, rate, and force, and in practical grasping tasks. Force and torque data generated during a task have been recorded during the experiments and are used to evaluate the performance of the teleoperation system. In Fig. 6, the Bode plots of the three control modes of the fingers are presented. The bandwidths are approximately 6.5 Hz for the force control, 5 Hz for the rate control, and 1.8 Hz for the position control. Figure 7 is a recording of the contact force in a grasping task: fingers are moving toward an object of 0.5 in. in di \ -;I 0 VELOCITY A FORCE 0 POSITION -9 b -10 I I I l l I I l I l I FREQUENCY, HZ (a) O r I I1 I ', I ~ I I I I -60 t \ \!,, I l 1! l 1 I I I 1 1 I Fig. 7. Recording of the impact force in a grasping task. ameter located at the center of the gripper, with an approach velocity of approximately 1.5 in./sec. The impact force measured on one of the fingers is 22.5 Ib for a set point of 15 Ib. The smart hand has been used for about 20 hr of intensive experiments in which it had to withstand high force transients. The results collected thus far show that the profile of the fingers helps the operator in developing efficient grasping strategies and in making these fingek more dexterous than those with flat surfaces. Figure 8 is the recording of the force along the X axis of the hand reference frame during a peg-in-hole (withdrawal, translation, insertion) task. This axis is the approach direction of the manip- October i

5 I L l Fig. 8. Recording of the force profile along the approach axis of the smart hand during a peg-in-hole task. ulator and is perpendicular to the task board surface. With these force/torque measurements, it will be possible to evaluate the effectiveness of the teleoperation system by comparing them to the recordings of a human performing the same operation on a task board mounted on the forceltorque sensor of the smart hand. Conclusion A new design for a manipulator end-effector has been described. It combines local processing capabilities with a modular mechanical design to cover a wide range of task requirements. This system has been characterized both in the stand-alone mode and as a part of the teleoperation laboratory at the Jet Propulsion Laboratory, where it is currently used for telemanipulation experiments. Its portability and simplicity of interface with a robotic system make it a very useful tool for robotic research. Acknowledgments Many people contributed to this project: Dhemetrios Boussalis with his control analysis, Blake Hannaford with his constructive comments, Richard Killion with his computer programming, Steve Mihalko with the mechanical assembly, Mark Nguyen with the electronics assembly, and Zoltan Vigh with the mechanical design. The support and encouragement from my Technical Manager and Technical Supervisor, Anta1 Bejczy and Edwin Kan, respectively, are much appreciated. The research described in this paper was performed at the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. References M. Belletti, C. Bonivento, and G. Vassura, Development of a Dexterous Hand for Industrial Robots, Conf. Proc. SIRI, Milano, Italy, Mar S. C. Jacobsen, J. Wood, D. F. Knutti, and K. B. Biggers, The Utah/MIT Dextrous Hand: Work in Progress, Int. J. Robotics Res., vol. 3, no. 4, Winter S. C. Jacobsen et al., Design of Tactile Sensing System for Dextrous Manipulators, IEEE Contr. Syst. Mug., vol. 8, no. I, Feb H. Kobayashi, Control and Geometrical Considerations for an Articulated Robot Hand, In:. J. Robotics Res., vol. 4, no. I, Spring T. Okada, Object-Handling System for Manual Industry, IEEE Trans. Sysr., Man, Cybern., vol. SMC-9, no. 2, Feb J. K. Salisbury, Kinematic and Force Analysis of Articulated Hands, Ph.D. Dissertation, Stanford University, CA, P. Dario and G. Buttazzo, An Anthropomorphic Robot Finger for Investigating Artificial Tactile Perception, Int. J. Robotics Res., vol. 6, no. 3, Fall A. H. Mishkin and B. M. Jau, Functional Requirements and Proposed Designs for Space-Based Multifunctional End-Effector Systems, JPL Pub , Apr S. 0. Leaver, J. M. McCarthy, and J. E. Bobrow, The Design and Control of a Robot Finger for Tactile Sensing, Technical Report 88-4, May 88, School of Engineering, University of California, Irvine. R. Tomovich, G. Bekey, and W. Karplus, A Strategy for Grasp Synthesis with Multifingered Robot Hands, IEEE International Conference on Robotics and Automation, A. Bejczy, E. Kan, and R. Killion, Integrated Multi-Sensory Control of Space Robot Hand, Proc. A IM Guid. Contr. ConJ, Snowmass, CO, Aug A. Bejczy and B. Jau, Smart Mechanical Hands for Teleoperation in Earth Orbit, Proc. 19th Aerospace Mechanisms Symp., NASA Ames Research Center, May A. BeJczY, Smart Hand-ManiPulator Control Through Sensory Feedback, JPL Pub. D-107, Jan. 15, [I41 F. S. Buchner and P. T. Wolfe, A Servoed Instrumented Gripper, Proc. Sensors [I51 86, Detroit, MI, Nov P. G. Goumas, A Universal Gripper for Small Parts Assembly, Lord Corp., Tech. Article , [I61 Telerobotics, Inc., EP General Purpose End Effector Documentation, (171 G. J. Vachtsevanos et al., Development of a Novel Intelligent Robotic Manipulator, IEEE Contr. Syst. Mag., vol. 7, no. [I81 3, June A. Bejczy and Z. Szakaly, A Laboratory System for Advanced Teleoperator Reseach, JPL Pub , Apr B. Hannaford, Task-level Testing for the JPL-OMV Smart End Effector, Proc. Workshop on Space Teleroborics, JPL Pub , vol. 2, pp , [20] D. Boussalis, Modeling and Controller Design for a Robotic Hand, Proc. Annual Pittsburgh Con$ Modeling and Simularion, Apr [21] A. Bejczy, B. Hannaford, and Z. Szakaly, Multi-Mode Manual Control in Telerobotics, Con$ Proc Romansy. Udine, Italy, Sept Paolo Fiorini received the laurea degree in electrical engineering from the University of Padova, Italy, in 1976, and the M.S. degree in 1982 from the University of California at Irvine. He is presently working toward a Ph.D. degree in the Manufacturing Engineering Program at UCLA. He has held several engineering and research positions, both in Italy and in the United States, developing advanced automation systems for consumer and industrial applications. He joined the Jet Propulsion Laboratory in 1985 and has been involved in research on robotic end-effectors and force-reflecting input devices for their control. His current research interests include control procedures for telerobotics and geometrical representation for multivariable systems. 24 IEEE Control Systems Mogorrrir