A Study on a Robotic Arm Contacting with Human Skin Using Tactile Sensing Feedback Strategies

Transcription

1 Proceedings of the 1 IEEE International Conference on Robotics and Biomimetics December 1-18, 1, Tianjin, China A Stud on a Robotic Arm Contacting with Human Skin Using Tactile Sensing eedback Strategies Jingguo Wang and Yangmin Li, IEEE Senior Member Abstract Inspired b human s tactile sensing in dail lives, we present an approach using the tactile sensing plus forcetorque information as the feedback to control the robotic arm interacting with our soft human skin in this paper. Three main tpes of contact between the end-effector of robotic arm and human skin are introduced and the contact model is built up. Hbrid impedance control method is applied to control both the position and force trajectories of the manipulator at the same time. With the feedback of tactile sensing data such as contact state, contact area and so on, several strategies of tactile sensing feedback are included in the control algorithm. Two groups of real eperiments are made using a two-link robotic arm equipped with force/torque and tactile sensors to contact with human skin. The results have confirmed the effectiveness of the proposed strategies. I. INTRODUCTION The human sense of touch has become the main source of insight and inspiration for the development of robotic tactile sensor in man subjects including automatic grasping, edge tracking, and rolling manipulation. Tactile sensation [1]-[8] plas an important role in robotic manipulation in need of high deterit. or some complicated robotic manipulation in uncertain and dnamic environments, open loop control will not suffice, therefore it is essential to know whether contact has occurred and measure the parameters of contact between the sensor and an object such as the spatial distribution of forces. The feedback of tactile sensing can be used to control force and torque at a specific contact location, which is necessar for manipulating objects and to control slippage [1]. The design, fabrication, and testing of a novel membrane tactile sensing sstem with onl four sensing elements are presented in []. A tactile sensing chip is developed, which can be installed on a fingertip of a humanoid robot [3]. A multi-sensor based generic approach is proposed to open doors for a deterous robot [], b combining both forcetorque and tactile sensor information. An integrated tactile feedback sstem for a multi-fingered robot hand is used to enable a human operator to feel contacts/interactions between the robot finger and the environment remotel []. The proposed method [] allows a robot to quantif tactile recognition of a human touched b other people. A novel identification technique of robot perception of impedance is proposed [7], in which the impedance parameters are estimated on-line based on position and force sensing. With Jingguo Wang (Ph.D candidate) and Yangmin Li (Professor) are with Department of Electromechanical Engineering, acult of Science and Technolog, Universit of Macau, Av. Padre Tomás Pereira, Taipa, Macao S.A.R., China Corresponding author: Yangmin Li s YMLi@umac.mo the aid of tactile sensor sstem, a human-interactive robot named RI-MAN [8] can successfull lift up a dumm human. Some related works can be found on the interactions between the robot and human bod using the impedance control method [9]-[13]. Three tactile factors are estimated in the work [13], which are needed to regenerate tactile sensation using tactile displa for a tactile telepresence sstem b one sensor apparatus. As the interaction between robot and environment is an essential part of the robotic manipulation task, a model-based optimization approach is suggested in [9] to deal with impacts, oscillations and constrained motion. An intelligent massage control sstem b using multi-fingered robot hand with hbrid impedance control is proposed [11], which is able to create the movement and the force of robot imitating the human hands massage. Although these works carefull investigate the contact between the robot end-effector and the human skin [17], no tactile sensing feedback techniques are adopted among them. In this paper, we mainl focus on the stud of the tactile sensing feedback to the hbrid impedance control scheme in addition with the force-torque sensing, while tactile sensor can know whether contact has occurred and determine the contact area and geometr from the sensor data. A twolink robotic arm is applied to contact with human skin, whose end-effector is equipped with tactile and force-torque sensors, as shown in ig. 1. The other parts of the paper are organized as follows: In section II, the general information of sstem is introduced. In section III, three general tpes of contact between the endeffector of robotic arm and the human skin are presented and the contact model is built up. In section IV, the hbrid impedance control is proposed to follow both the position and force trajector simultaneousl. In section V, several strategies of tactile sensing feedback are proposed. Eperiments are made and several subjects are discussed in section VI. The concluding remarks are obtained in the last section. II. SYSTEM DESCRIPTION The two-do(degree of freedom) robotic arm made of two PowerCube modules (PR-7 and PG-7 from Amtec robotics GmbH) is applied to interact with human bod, as shown in ig. 1. The force sensor (TCL--8 from SCHUNK GmbH & Co. KG) is mounted between the endeffector and the last joint. The tactile sensor(dsa 91 with controller DSACON3-H from Weiss Robotics) having arclike touching surface is installed at the tip of end-effector, providing a spatial resolution of 3.mm with 7 sensor cells. The working principle of the tactile sensors depends on an /1/$. 1 IEEE 9

2 interface effect between metal electrodes and a conductive polmer covering the sensing electrodes, and the resistance between a common electrode and a sensor cell electrode is a function of the applied load and time []. The output voltage of a resistive sensor cell in the electrical circuit can be detected on real-time, and the pressure or force on it will be obtained from the transducer characteristics. Human skin can be divided into two separate laers, the epidermis and dermis, and the studies show that the dermis is primaril made of collagen, elastin and reticulin fibers, which all contribute to the mechanical behavior of skin [], though the epidermis is stiffer than the dermis. Taken as a soft environment, the human skin is modeled as the massspring-damper sstem in the control scheme. Two PowerCube modules are driven through the data from PC b CAN-bus, and then the measured data of force-torque and tactile sensor are sent back as the feedback to controller through serial communication. CAN-BUS PowerCube modules Human Skin Tactile Sensor (a) Contact with conve surface Tactile Sensor Human Skin (d) Contact with flat surface Tactile Sensor Human Skin (g) Contact with concave surface (b) Conve surface of human skin (e) lat surface of human skin (h) Concave surface of human skin (c) Conve surface of of rigid object (f) lat surface of rigid object (i) Concave surface of rigid object ig.. ig. (a), ig. (d) and ig. (g) are three contact tpes, correspondingl, the teel-voltages of tactile sensor cells contacted with three tpes of human skin surfaces are shown in ig. (b), ig. (e) and ig. (h) while ig. (c), (f) and (i) demonstrate the case of contact with rigid object. PC PCI-card ig. 1. RS-3 Tactile sensor -ais force/torque sensor Configuration of the whole robot sstem III. CONTACT PROPERTIES O HUMAN SKIN It is ver difficult to determine human skin properties and to model its behavior, since human skin is a nonhomogeneous, anisotropic, non-linear viscoelastic material whose properties also var with age, race, and living area, therefore it is a ver comple biological structure to stud. Therefore in this paper, we consider onl three main different kinds of human skin surfaces and build up its contact model. A. Three general contact tpes In general terms, considering the shape of human skin surfaces, three tpes of contact will be divided as conve, flat and concave surfaces as shown in ig. (a), ig. (d) and ig. (g) respectivel. The eperiments are made on three different positions (with conve, flat and concave shapes) of human hand to get the performances of tactile sensor. Under the same vertical force eerted (=N), the corresponding teel-voltages measured from the contacted tactile sensor cells are shown in ig. (b), ig. (e) and ig. (h) respectivel. In order to show the difference between the soft human skin and hard object, the eperiments are also made respectivel on three kinds of surfaces(flat, concave and conve) of hard object with the same force (=N) and the teel-voltages of tactile sensor cells are shown in ig. (c), (f) and (i). Among the eperimental results, it is easil found that the teel-voltages of the sensor cells are different although the same force at the end of the end-effector is eerted, which is determined b the contact tpe and contact area. The biggest teel-voltages of the contacted sensor cells appear in the conve tpe as shown in ig. (b) and ig. (c), but the concave tpe has the smallest values in three contact tpes as shown in ig. (h) and ig. (i), since it has more contact areas than others. As epected, the flat tpe arrives at a compromise. Between ig. (b) and ig. (c), the former has more large-voltage cells because the soft human skin has bigger contact area. B. Modeling of human skin The contact model between the end-effector of the robotic arm and the human skin is built up as a spring-mass-damper model, as shown in ig. 3, where t is the end-effector position of the robot, M k, K k and D k are the mass, spring and damping coefficient of the skin. The human skin model can be represented as: M k ẍ t + D k ẋ t + K k t + s = e (1) e is the force of end-effector eerting on the human skin and s is the eternal force. To evaluate the uncertainties of the impedance estimates and to detect the discontinuous changes of the impedance, a novel identification technique of constraint condition that the environment imposes on the robots end-effector is proposed

3 in [7], which is based on position and force sensing during arbitrar manipulation. Inspired b this, an identification method of human skin muscle to decide the parameters of impedance controller is discussed carefull in the work [11], and the performance inde to determine skin muscle parameter is given out. d d d d Outer loop Trajector generator X t Kinematics solution Inner loop t q Computed torque controller Tactile_Sensing q q ARM Hum_Skin M k K k ig.. Block diagram of hbrid impedance control robot e ig. 3. D k s skin The model of human skin IV. HYBRID IMPEDANCE CONTROL O ROBOTIC ARM The whole sstem in this paper can be regarded as a rigid manipulator in soft contact with a compliant environment and the desired impedance is assumed to be a mass-springdamper sstem. This kind of environment is usuall modeled as a linear spring K, which is sometimes in parallel with a dashpot B. This will belong to the field of impedance control. With regard to the contact with soft human bod, either position impedance-based control or force control is not enough for the task, and the hbrid impedance control works for the solution, which is a combination of two fundamental force control strategies, both the positions are commanded and a command force trajector is followed [1] [1]. A. Hbrid Impedance Control The desired impedance of robot arm has the form as follows: M d (ẍ ẍ d )+B d (ẋ ẋ d )+K d ( d )= e () where, ẋ and ẍ are the position, velocit and acceleration of the robot end-effector. d, ẋ d and ẍ d are the reference position, velocit and acceleration. M d, B d and K d R m m are positive-definite matrices representing inertia, damping and stiffness respectivel, which are usuall chosen to be diagonal, rendering the sstem decoupled, and e is the environmental reaction forces. or the robot Cartesian space control, one form of robot dnamics is considered: τ u = M(q)J 1 (q)(ẍ J q)+n(q, q)+j T e (3) where q, q R n denote the joint position and velocit respectivel; M(q) R n n the inertia matri; N(q, q) R n the centrifugal, Coriolis, gravitational, and friction forces; J(q) R n n the Jacobian; and τ R n the input torque. Combining () and (3), the control input for the impedance control can be epressed as: τ u = M(q)J 1 (q)(md 1 ( e B d (ẋ ẋ d ) () K d ( d ) J q)+n(q, q)+j T e However the impedance control is onl a position control scheme, and it has no an attempt to follow a command force trajector. Therefore, the hbrid impedance control proposed is defined as follows: M d (ẍ Sẍ d )+B d (ẋ Sẋ d )+K d S( d ) (I S) d = e () where S is diagonal with ones and zeros on the diagonal while ones on the diagonal relate to position-controlled subspaces and zeros relate to the force-controlled subspaces. Here the task space is divided into orthogonal position and force controlled subspaces using the selection matri S. d is the desired force. When S =, the desired equation of motion in the position-controlled subspaces is identical to the following equation: M d ë p + B d ė p + K d e p = e () where e p is the position tracking error and e p = X X d. When S = 1, the desired impedance is defined in the forcecontrolled subspace b M d ẍ + B d ẋ d = e (7) The control input etended to hbrid impedance control has the following form: τ u = M(q)J 1 (q)(md 1 ((I S) d e B d (ẋ Sẋ d ) (8) K d S( d ) J q)+n(q, q)+j T e B. Control law of robotic arm As shown in ig., the inner and outer loop control strateg can be combined to achieve the closed-loop dnamics. The outer loop outputs the acceleration trajectories reflecting the desired impedance in the position-controlled subspaces as pointed in (), and the desired force in the force-controlled subspaces as pointed in (7). The inner-loop control is to select an input to the actuators which makes the end-effector track the desired trajectories generated b the outer loop. The input torque of the robotic arm dnamics is defined as: τ = M(q) q +C(q, q) q + G(q) J T e e (9) 1

4 where M(q) R n n is a smmetric and positive definite joint inertia matri, C(q, q) q R n is the centripetal and Coriolis force, G(q) R n is the gravit force and e R m is the contact force eerted b the end-effector on human skin. V. TACTILE SENSING EEDBACK(TS) STRATEGIES IN THE CONTROL ALGORITHM Besides the force-sensing feedback, tactile sensing can provide some useful and crucial information to the robotic control sstem when the end-effector of robotic arm will make a contact with the unknown environment geometr. Since tactile sensors determine various aspects of the environment through direct contact with objects within that environment [1], such sensors can be used to ascertain the distribution of forces over an area, and can also be used to sense shape parameters, location, and orientation of objects touching with the sensor. Tactile sensors produce as output a digital image that alwas can be input to a conventional image processing sstem, as might be used for analzing images from a vision camera. But in this paper, no vision sstem is applied and the contact data (teel-voltage) of tactile sensor will be directl transferred into the control algorithm and some of the strategies are proposed in the table I. Strateg I: Strateg II: Strateg III: Strateg IV: TABLE I TACTILE SENSING EEDBACK STRATEGIES To detect whether the contact occurred To judge the general contact information b the distribution of contact points To know the contact area from the number of tactile sensing cells To avoid large pressures of some critical points Safet is alwas the first to be considered if the contact happens between the robot and human bod, like massaging. The first strateg will ensure the safet and be in case of emergenc just based on working states of sensor cells. rom the second strateg, the contact tpe will be acquired from the pressure-value (or teel-voltage) distribution of sensor cells and contacted points can be known easil. In third strateg, the size of contact area can be calculated directl b summing up the number of contacted sensor cells, but note that the error will be decided b the size resolution of tactile sensor cell. Because the teel-voltage value of sensor cell can be obtained easil, some corresponding strategies should be made to avoid large pressures on some points and make a quick response when the occur. Therefore, based on the above tactile sensing strategies, the tactile sensing feedback (TS) will pla an important role to contact control between the robot and human skin. VI. EXPERIMENTAL INVESTIGATIONS In order to show and verif the importance of the tactile sensing feedback(ts) in the contact with human skin, two groups of eperiments are designed. The first aims to know the difference of contact at initial step with and without the aid of tactile sensing when the hbrid impedance control is applied to track both the position and force trajectories. The second is to contact with human skin in soft and hard positions respectivel in order to test its force-tracking abilities in two different environments. A. Eperimental setup A two-link robotic arm actuated b PowerCube modules are equipped with force-torque and tactile sensor on the endeffector. The lengths of two links are a1 =.38m and a =.81m, the masses are m1 = 1.8kg and m =.8kg, and the initial velocities will be set zero. The motor modules are communicated and controlled through CAN bus, while force and tactile sensing data is transmitted through RS 3 cables and then will be sent back to PC control. B. Eperiment I or the contact work, it will be better to know first whether there is a contact occurred at the beginning, and then decide a corresponding response. Two kinds of eperiments are made to determine the effectiveness of the TS strateg. The end-effector of the robotic arm is desired not onl to track a short straight path(8 mm) on the forearm, but also to keep a constant force ( N) in the direction vertical to operated human skin surfaces. Hbrid impedance control method is applied, and two eperiments are done with and without the consideration of TS respectivel. The results are shown in ig. (a). rom the results, it is easil found that in the case of considering TS, the reaction force changes slowl. The desired force is tracked well considering TS but there are differences between the desired values and the measured ones, because of the resolution of the tactile senors. In order to know more details at the ver beginning of robotic end-effector contacting with human skin, the tactile sensing data are investigated carefull with and without considering TS respectivel as shown in ig. (b) and ig. (c). Since the tactile sensor is composed of a matri of sensor cells, each of them will be responded in contacting. Therefore, from the number of contacted sensor cells, the contact area can be marked as the red-bar figures shown in ig. (b) and ig. (c). The teel-voltage of each contacted sensor cell is also be measured and its value is summed up as the blue-bar figures shown in ig. (b) and ig. (c). It can be found from ig. (c) that after t =.8s, the response of tactile sensor comes stead or even constant, which will be an eplanation of stead force values (in blue line) in ig. (a). The force values (in red line) vibrates during all the time of the investigation period, which can also be indicated b the change of teel-voltages in blue-bar figure of ig. (b). C. Eperiment II Two following eperiments are made using end-effector of robotic arm to contact with soft and hard human skin respectivel and the difference between them is eploited.

5 Because of two different kinds of human skin [11], different impedance parameters of inertia, skin elasticit and damping coefficient are estimated. 1) Eperiment on soft human skin: A soft part (modeled as M k =.1;D k = ;K k = ) is chosen at the forearm of the subject and the end-effector is desired to contact with this point while a changing force (the changing range is 1 N) is eerted. The desired force is followed as shown in ig. (a) and the teel-voltages of all contacted sensor cells are summed up as shown in ig. (b). The changing trend of the curve in ig. (b) identifies well with the force measured from si-ais force/torque sensor. ) Eperiment on hard human skin: Similarl, a hard part (modeled as M k =.3;D k = 11;K k = 13) is chosen at the back of the subject s hand and the same force trajector is required to be followed. The similar eperimental results are showninig.7. Comparing between these two eperiments, it is not difficult to find that the average teel-voltages produced in the eperiment of hard human skin is larger than the former eperiment, which confirms again that contacting with the hard object will produce large teel-voltage but less contacted area as discussed in Section III. D. Discussions Usuall the contact tpe is modeled as point contact, the surface contact is presented in this work, which is equivalent to the sum of a group of point contacts. The contact area can be calculated b the arra number of contacted cells. The tactile sensor with arc-like shape is used in the eperiments, which has such advantages in contacting with human skin as the contacted area will become larger if the eerted force will increase bigger, which means that the contact area is sensitive to the eerted force. However, under high pressure, the contacted sensor cells are eas to reach saturation, where their teel-voltages are no more changed even though a large force is eerted. Therefore, tactile sensing feedback gives helpful information if the eternal force locates in a reasonable range as demonstrated b the eperiments. The TS strategies pla an important role mainl at the beginning phase of contact as demonstrated in the first eperiment since a lot of information about contacting can be acquired. VII. CONCLUSIONS The paper presents a wa of contacting with human skin b a two-link robotic arm with tactile sensing feedback. Since the human skin is ver complicated and its properties are difficult to determine, it is modeled as a mass-springdamper sstem. Hbrid impedance control is applied to manage both the motion and force of the end-effector of robotic arm. Since tactile sensor provides much important information about contacting, several strategies of tactile sensing feedback are proposed thereafter are considered in the control algorithm. Real eperiments are made on soft and hard human skin respectivel in order to test the effectiveness of the tactile sensing, and the results demonstrate that tactile sensing feedback is helpful for the contacting tasks. ACKNOWLEDGMENTS This work is supported b Macao Science and Technolog Development und under Grant no. 1/8/A1. REERENCES [1] J. Tegin and J. Wikander, Tactile sensing in intelligent robotic manipulation: A review, Industrial Robot: An Int. J., Vol. 3(1), pp. 7,. [] A. Mirbagheri, J. Dargahi, S. Najarian, and. T. Ghomshe, Design, abrication, and Testing of a Membrane Piezoelectric Tactile Sensor with our Sensing Elements, American Journal of Applied Sciences, Vol. (9), pp., 7. [3] R. S. Dahia, G. Metta, and M. VaIle, Development of ingertip Tactile Sensing Chips for Humanoid Robots,IEEE Int. Conf. on Mechatronics. Malaga, Spain, April 9, pp.1-. [] A.J.Schmid,N.Gorges,D.Göger and H. Wörn, Opening a door with a humanoid robot using multi-sensor tactile feedback, IEEE Int. Conf. on Intelligent Robots and Sstems, Pasadena, CA, USA, 8, pp [] W. T. Lo, Y.T. Shen, and Y. H. Liu, An integrated tactile feedback sstem for multifingered robot hands, IEEE Int. Conf. on Intelligent Robots and Sstems, Hawaii, USA, 1, pp [] H. Iwata, and S. Sugano, HumanCrobot-contact-state identification based on tactile recognition, IEEE Trans. on Industrial Electronics, Vol. (), pp ,. [7] R. Kikuuwe and T. Yoshikawa, Robot perception of environment impedance, IEEE Int. Conf. on Robotics and Automation, Washington, DC, USA,, pp [8] T. Mukai, M. Onishi, T. Odashima, S. Hirano, and Z. W. Luo, Development of the tactile sensor sstem of a human-interactive robot RI-MAN, IEEE Trans. on Robotics, Vol. (), pp. 1, 8. [9] G. Prokop and. Pfeiffer, Snthesis of robot dnamic behavior for environmental interaction, IEEE Trans. on Robotics and Automation, Vol. 1(), pp , [1] Y. Ohmura, Y. Kunioshi, and A. Nagakubo, Comfortable and scalable tactile sensor skin for curved surfaces, IEEE Int. Conf. on Robotics and Automation, Orlando, L, USA,, pp [11] K. Mouri, K. Terashima, P. Minong, H. Kitagawa and T. Mioshi, Identification and hbrid impedance control of human skin muscle b multi-fingered robot hand, IEEE Int. Conf. on Intelligent Robots and Sstems, San Diego, CA, USA, 7, pp [1] R. J. Anderson and M. W. Spong, Hbrid impedance control of robotic manipulators, IEEE Journal of Robotics and Automation, Vol. (), pp. 9, [13] S. Okamoto, M. Kono, Y. Mukaibo, T. Maeno, and S. Tadokoro, Real-time estimation of touch feeling factors using human finger mimetic tactile sensors, IEEE Int. Conf. on Intelligent Robots and Sstems, Beijing, China,, pp [1] Y. Hasegawa, T. Ootsuka, T. ukuda,. Arai and Mi. Kawaguchi, A relaation sstem adapting to user s condition -identification of relationship between massage intensit and heart rate variabilit, IEEE Int. Conf. on Robotics and Automation, Seoul, Korea, 1, pp [] S. L. Evans, C. A. Holt, Measuring the mechanical properties of human skin in vivo using digital image correlation and finite element modelling, The Journal of Strain Analsis for Engineering Design, Vol. (), pp , 9. [1] J.-G. Wang and Y. Li, Massaging Human eet b a Redundant Manipulator Equipped with Tactile Sensor, IEEE Int. Conf. on Advanced Intelligent Mechatronics, Montreal, Canada, 1, pp [17] J.-G. Wang, Y. Li, Impedance control of a spatial redundant manipulator used for relaing muscle fatigue, IEEE Int. Conf. on Mechatronics and Automation, Changchun, Jilin, China, 9, pp

6 force(n) 1 1 a period for investigation reference without TS with TS force(n) 1 Reference force and contact force measured in the eperiment(soft) measured force reference force (a) Reaction force measured from force sensor compared with reference force in two eperiments cell number sum of teel voltages 1 numbers of contacted sensor cells(no TS) Total teel voltages of contacted sensor cells(no TS) (b) Contacting area and teel-voltages of sensor cells measured in the eperiment without considering TS cell number sum of teel voltages 1 numbers of contacted sensor cells(ts) Total teel voltages of contacted sensor cells(ts) (c) Contacting area and teel-voltages of sensor cells measured in the eperiment considering TS ig.. The end-effector of robotic arm contacting with human skin in two eperiments without and with considering the TS respectivel force(n) sum of teel voltages(mv) (a) Desired force and the force measured from force sensor 1 Total teel voltages of contacted sensor cells(soft) (b) Total teel-voltages of contacted sensor cells measured ig.. 1 The eperiment of contacting with soft human skin Reference force and contact force measured in the eperiment(hard) measured force reference force sum of teel voltages(mv) (a) Desired force and the force measured from force sensor Total teel voltages of contacted sensor cells(hard) (b) Total teel-voltages of contacted sensor cells measured ig. 7. The eperiment of contacting with hard human skin