The IEEE/RSJ International Conference on Intelligent Robots and Systes October -,, Taipei, Taiwan Toward Understanding the Effects of Visual- and Force-Feedback on Robotic Hand Grasping Perforance for Space Teleoperation Neal Y. Lii, Zhaopeng Chen, Benedikt Pleintinger, Christoph H. Borst, Gerd Hirzinger, and Andre Schiele Abstract This paper introduces a study aied to help quantify the benefits of liited-perforance force-feedback user input devices for space teleanipulation with a dexterous robotic ar. A teleoperated robotic hand has been developed for the European Space Agency by the Geran Aerospace Center (DLR) for a lunar rover prototype. Studies carried out on this telerobotic syste investigated several criteria critical to teleanipulation in space: 1) grasping task copletion tie, ) grasping task difficulty, 3) grasp quality, and ) difficulty level for the operator to assess the grasp quality. Several test subjects were allocated to reotely grasp regular and irregular shaped objects, under different cobinations of visual- and forcefeedback conditions. This work categorized the benefits of visual- and force-feedback in teleoperated grasping through several perforance etrics. Furtherore, it has been shown that, with local joint-level ipedance control, good grasping perforance with rigid hard objects can be achieved, even with liited force-feedback inforation and low counication bandwidth. On the other hand, a perforance ceiling was also found when grasping deforable objects, where the liited force-feedback setup cannot sufficiently reflect the object boundary to the teleoperator. I. INTRODUCTION HERE have been considerable interest in telerobotics Tsyste perforance evaluation in aster-slave teleoperation, due to its wide array of relevant applications. Several studies investigated the added value of inclusion of force reflection and its effects under different visual conditions to operator task perforance. Such studies have been conducted with both real physical [1] [] [3] and siulated virtual [] [5] slave systes. Despite the large nuber of investigations, however, a clear overall consensus or guideline for teleoperation syste design reain unavailable, naely: 1) the influence of forcefeedback quality on task perforance, ) the roles of force and vision in the for of cobined feedback (e.g. feedback doinance-task relationship), and 3) the effects of liited force-feedback inforation on task perforance. Manuscript received March 9,. This work was supported in part by the European Space Agency, AG Noordwijk, The Netherlands. Neal Y. Lii, Zhaopeng Chen, Benedikt Pleintinger, and Christoph H. Borst are with the Geran Aerospace Center (DLR), Oberpfaffenhofen, 3 Weßling, Gerany (phone: +9-()153-1; fax: 9-()153-113; e-ail: neal.lii@ dlr.de). Andre Schiele is with the European Space Agency, AG Noordwijk, The Netherlands (phone: +31()71-55-37; fax: +31()71 55 519; e- ail: andre.schiele@esa.int). The effects of liited force-feedback on teleoperation perforance are the focus of this presented work. It has been observed in previous work force-feedback can indeed be of benefit to teleoperated tasks []. However, a counterpoint has also been raised that liiting force-feedback inforation ay in fact lower operator task perforance [1]. Understanding the level of refineent in force-feedback inforation is especially relevant to space teleoperation. One reason is that a space aster-slave syste is likely to have liited capabilities in bandwidth allocation and transparency, due to overall syste resource constraints and possible significant data transission delay. Syste cost and physical size, both iportant design criteria for space applications, could also further reduce the available forcefeedback inforation for teleoperation. As a result, these systes would be liited to eploying liited perforance feedback to help perfor teleanipulation tasks. Therefore, it is critical to quantify the types and extent of benefit that can be extracted fro a low-perforance syste. A general purpose space robotic syste naed Eurobot is under developent for exploration purposes at the European Space Agency (ESA) []. As a part of defining suitable teleoperation interfaces, it has been deeed necessary to investigate the added value of force reflection under realistic syste conditions. The syste would eploy a dexterous robotic hand and controlled via a data-glove ated to a force-reflection hand exoskeleton. Fig. 1. FFH ipleentation on the Eurobot Ground Prototype (EGP) To help assess the functionality of this teleoperated hand syste for dexterous grasping tasks, this research ais to establish a concise evaluation etric to clarify the perforance gains that can be achieved, under realistic conditions. This paper introduces a study carried out to deterine the effects of force- cobined with visualfeedback on a realistic range of operator perforance etrics as anticipated for real space teleoperation tasks. As teleanipulation tasks perfored in space usually have a cobination of various degrees of visual- and forcefeedback, this work would exaine several aspects of the effects of different feedback conditions through several 97-1--7-//$5. IEEE 375
perforance etrics. Besides a larger range of the perforance etrics for grasp perforance evaluation, this work particularly distinguishes itself fro siilar works [] [3], through its ephasis on the use of a sall nuber of force feedback channels copared to the D.O.F. of the end effector (robotic hand), as well as a low-bandwidth counication channel between aster and slave syste. Finally, this work hopes to serve as an entry into a series of studies to systeatically exaine the effects of various levels of force-feedback on space teleanipulation tasks. II. OVERALL SYSTEM DESCRIPTION The proposed syste is a further developent based on the concepts for teleoperated robotic hand introduced at the Geran Aerospace Center (DLR), Oberpfaffenhofen, Gerany [7]. Its key coponents consist of a DLR/HIT II five-finger dexterous robotic hand (FFH) [] [9] and a haptic operator interface fro CyberGlove Syste, as shown in Fig.. This syste is designed for dexterous object anipulation to be eployed on the Eurobot Ground Prototype (EGP), and operated eventually fro the space station reotely. The EGP, shown in Fig. 1, is a planetary exploration version including a obility and navigation subsyste, of the Eurobot []. The force-feedback exoskeleton coposes of five singledirection cable actuators to exert force on each finger to generate opposing force to finger flexion. The forcefeedback exoskeleton s five channels of force in the pullback direction (one for each finger) are significantly fewer than the D.O.F. for the huan hand, as well as the 15 D.O.F of the FFH. Furtherore, with a bandwidth of ~9 Hz, a liitation is also placed on the data transission rate. These conditions place a significant liit on the forcefeedback inforation available to the reote operator. As the dexterous robotic hand is able to provide three torque easureents for each finger, or 15 in total, an appropriate apping of forces to be reflected fro the robotic hand to the exoskeleton would also be necessary. Exoskeleton force feedback Force to hand apping Force gain calibration CyberGlove calibration user interface CyberGlove -bit sensor data Glove sensor gain and offset calibration GyberGlove to hand apping Hand gain and offset calibration FFH force feedback Calibrated coand to FFH control Fig.. Experiental environent A. Master Syste: User Teleanipulation Input The CyberGrasp [] consists of a data glove for hand gesture easureent, and an exoskeleton for force-feedback. The CyberGrasp hardware can be seen worn by an operator as a part of the experiental environent in Fig.. The data glove utilizes bend sensors with -bit resolution to easure the oveents of each joint in the hand. As each operator s hand size and for can differ significantly, the syste provides the possibility to calibrate the glove sensors to custo fit to each user. Although both the CyberGlove and the FFH attept to iic fully the huan hand, both are unable to fully duplicate all joints in the huan hand. For exaple, the CyberGlove provides only four sensors to easure five finger abductions, whereas the FFH does not have a roll joint for the opposing thub. Therefore, the glove sensor data easureent ust be appropriately apped to the robotic hand for iproved teleanipulation perforance. Fig. 3. Flow-chart of software odules for calibration and apping between data-glove, robotic hand and hand exoskeleton The glove coand and force-feedback apping algoriths have been odeled in Matlab/Siulink, and operate in the QNX real-tie environent via external ode. To help iprove operator coand and feedback perforance, a GUI has been included in the syste, in addition to sensor calibration, to adjust the apped hand response, as well as feedback force gain, as shown on the right hand screen in Fig.. Finally, a diagra is shown in Fig. 3 to help clarify the user anipulation and forcefeedback syste design. B. Slave Syste - Robotic Hand The DLR-HIT II five-finger dexterous robotic hand (FFH) was jointly developed by DLR, and the Harbin Institute of Technology, Harbin, China [] [9]. It is based on the technology and concepts developed on the DLR Hand II [11], and axiized the use of off-the-shelf coponents to help explore the feasibility of series production. The FFH consists of five identical odular fingers, serving as four fingers and one opposing thub. Each finger has three DOF, with coupled interediate and distal joints, and decoupled proxial and abduction actuation with a axiu speed of deg/s. Each finger is capable of a axiu tip force of 9 N. The FFH can be seen as a part of the syste setup in Fig 7. The FFH can be ipleented with Cartesian- or joint-space ipedance controllers [9] for hand grasping tasks. For the 37
application of glove-controlled object teleanipulation on the Eurobot, huan finger joints would be apped onto the robotic hand to siulate the grasp gestures coanded by the operator s hand. In this case, the joint-space controller would be the ore able to closely siulate the intention of the huan operator. The ipleentation of the proposed controller is explained in ore detail as follows: θ s θ d θ & d Trajectory generation The robot hand dynaic odel is as expressed below [1]: M ( q) q& + C( q, q& ) q& + g( q) = τ + τ ext (1) B & θ + τ + τ = τ () f + - D d θ jd θ & jd K d 1 θ j L d 1 j + L θ & d + + K e Motor joint space ipedance controller τ = K( θ q) (3) M (q), C ( q, q&) q&, and g (q) represent the inertia atrices, centrifugal ter, and gravity ter, respectively. The joint torque vector is given by K(θ-q), where θ indicates the vector of the otor angle divided by the gear ratio, and q represents the link side joint angle. K and B are diagonal atrices which contain the joint stiffness, and the otor inertia ultiplied by the gear ratio squared. τ and τ f are external torque vector and friction torque vector, respectively. The generalized actuator torque vector, τ, is considered as the control input. The goal of the ipedance controller is to achieve a desired dynaic behavior with respect to external forces and τ d FFH Finger Fig. The robotic hand joint space odel. Where: θ is the operator desired joint angle, θ s jd is the desired joint angle post trajectory generation, & θ is the desired joint angular rate post trajectory generation, jd θ d is the desired otor angle, & θ is the desired otor angular rate, d K is the desired stiffness, d D is the desired daping, d τ is the actual otor torque, θ is the actual otor angle, & θ is the actual otor angular rate, θ is the actual FFH finger joint angle, j K is the ipedance behavior between the finger joint and external e torque, and L represents the otor-joint forward kineatics. θ τ θ & θ j L θ j θ torques acting on the link side of the finger, given by a desired stiffness paraeter K d, as well as a desired daping paraeter D d [13]. For passivity considerations, in case that the desired ipedance behavior is defined in the joint coordinate systes, a otor position based PD-controller can be ipleented as the following: τ θ θ & = Kd ( s ) Ddθ () where θ s represents a desired configuration in joint space. Together with () and (3), the following closed loop equations can be arrived: M ( q) q& + C( q, q& ) q& + g( q) = τ + τ ext (5) ~ B& θ + D & d θ + K d θ + τ + τ f = () where ~ θ = θ θ s The ipleentation of the proposed controller is as shown in Fig.. The controller is ipleented in Matlab/Siulink, and runs in the QNX real-tie environent via external ode. A graphic user interface allows the operator to control the FFH s syste settings reotely. A reote viewer of the FFH is also built into the FFH GUI, which provides a streaing display of the robotic hand s current gesture based on the position sensors easureents. The reote viewer of the FFH is shown in Fig. as a part of the experiental environent on the onitor to the left. C. Overall Syste Architecture The syste architecture of the CyberGrasp teleoperated FFH consists of two sides: the coander side, and the EGP side, where the FFH is installed. The counication between the sides is provided via a wireless LAN link. The CyberGrasp Syste CyberGlove and exoskeleton are installed on the coand side for user haptic coand input and force-feedback. This side of the syste also provides the operator with the user interfaces both for the CyberGrasp and the FFH, to allow the operator to ake necessary adjustents to the syste for different teleanipulation tasks. A QNX real-tie ini-rack coputer is ipleented on the EGP Side for the control of the FFH. The syste ay be operated in wired or wireless ode, depending on task requireents. The robotic hand controller operates at a rate of 5 KHz. The transission rate to the CyberGrasp is at above 1KHz in wired network configuration, and about 5 Hz in wireless configuration. However the bottleneck of the syste refresh rate lies in the CyberGrasp controller, which updates sensor easureents at 9 Hz intervals. (7) III. EXPLORATORY FORCE-FEEDBACK GRASPING EXPERIMENTAL STUDY AND APPROACH A. Experient Subjects Three huan operator subjects were allocated to perfor a series of teleoperated grasping tasks of different objects. Each subject is of different a level of knowledge on teleanipulation and the experiental setup, ranging fro 377
expert and high failiarity with the syste to a novice user. The spread in expertise allowed a ore full-spectru exaination of the teleoperator response and perforance. condition cobination, scalar ratings of 1) perceived grasp task difficulty, and ) grasp quality assessent difficulty are given by each test subject as subjective etrics. The data collected and observations are analyzed both qualitatively and quantitatively, and discussed in the following sections. Teleoperated FFH grasping soft ball Teleoperated FFH grasping hard ball IV. RESULTS AND ANALYSIS In order to analyze data collected fro the teleanipulation grasping experients, ANalysis Of VAriance (ANOVA) is eployed in this section, which allows the definition of threshold for significant variances in saple eans. In the following, variation probabilities under the 5% level are considered significant. Teleoperated FFH grasping pistol grip tool Fig. 5. Different test grasp objects being grasped as coanded by the CyberGrasp user coand B. Selection of Objects for Teleoperated Grasping Grasping of regular and irregular shaped objects was exained. For this study, the sphere shape was chosen as a representative regular shape. A hard ball was used to siulate a rigid regular shaped object, and a soft ball to siulate a deforable object. The balls are of siilar diensions of about.7 c. The pistol grip tool was selected to represent an irregular shape object for its relevance in space application, particularly in extra-vehicular operations fro the International Space Station (ISS). A power drill was used to siulate a pistol grip tool due to its siilarities in for, size, and trigger position to actual pistol grip tools. Fig. 5 shows the three types of test grasp objects being grasped by the FFH teleoperated through the CyberGrasp haptic user coand. C. Test Sequence and Perforance Metrics In order to exaine the effects of visual- and forcefeedback on grasping perforance, each of the three test subjects were required to teleanipulate grasps of the regular and irregular shaped objects described above, under various cobinations of feedback conditions. These included: Optial vision with force-feedback Obstructed vision with force-feedback Optial vision without force-feedback Obstructed vision without force-feedback Each grasp object-feedback condition cobination was repeated ties to help collect sufficient grasp saples. For each grasp operation, the 1) tie to copletion, ) grasp quality, and 3) grasp assessent correctness were recorded as objective etrics. Finally, for each grasp object-feedback A. Tie to Coplete Grasping Task Possibly due to the relatively low nuber of test subjects, experiental results have shown no significant difference in task copletion tie in relation to different objects, as shown in Fig.. A weak trend is visible towards ore tie required for task copletion as a result of less feedback inforation, both visual and force. Tie to coplete grasp [s] 3 5 15 5 Opt.Vis.+F.-F. Obs.Vis.+F.-F. Opt. Vis. Obs. Vis. Fig.. Box plot of group easureents of tie required until grasp copletion. The experient conditions are shown as: Opt.Vis.+F.-F. (optial vision including force-feedback), Obs.Vis.+F.-F. (obstructed vision including force-feedback), Opt.Vis. (optial vision alone), Obs. Vis. (obstructed vision alone). B. Perception of Task Difficulty Fig. 7 shows a trend of overall increasing difficulty with the copounding loss of feedback inforation. This is particularly clear when left with only obstructed vision of the grasping task. Clear statistical significance can be seen copared to other conditions of optial vision with forcefeedback (p=.3), obstructed vision with force-feedback (p=.), and optial vision setup (p=.). 37
[ easy] Perceived difficulty of task [ difficult] Fig. 7. Box plots of group easureents of perceived task difficulty over the four experient conditions. The difficulty of task is graded by the subjects on a scale of to, with being the easiest, and the ost difficult When exaining vision as the sole feedback criteria, it plays a crucial role in perceived task difficulty. This is clearly deonstrated in Fig., as vision obstruction increases the difficulty of the task significantly (p=.19). All test objects also noted their increased reliance on the robotic hand virtual viewer provided when perforing grasping tasks under obstructed visual-feedback. [=easy; Perceived Task Difficulty [=difficult] 1 1 1 1 1 1 Opt.Vis.+F.-F. Obs.Vis.+F.-F. Opt.Vis. Obs.Vis. Opt. Vis. (w & w/o F.-F.) Obs. Vis. (w & w/o F.-F.) Influence of Vision alone Fig.. Influence of vision alone to perceived task difficulty. Obstructed vision significantly increases the perceived difficulty of the task (p=.19). Both conditions included tests with and without force-feedback. Force-feedback also deonstrates a strong influence on the perceived task difficulty. Without force-feedback, the task difficulty increases significantly (p=.), as shown in Fig. 9. C. Assessent of Grasp Quality The grasp quality assessent reflects the confidence of the operator in his/her knowledge of the state of the grasp. The test subjects were asked to assess whether a grasping task has succeeded or failed upon task copletion. The assessent is iediately confired by an experient facilitator by an attept to reove the grasped object fro the FFH. Perceived Task Difficulty Fig. 9. Influence of force-feedback alone on perceived task difficulty. Without force-feedback, difficulty increases significantly (p=.). Both conditions included optial and obstructed vision. 1) Difficulty level of grasp quality assessent As shown in Fig., the object itself gives clues about grasp perforance. It was easier to assess grasp quality with the pistol grip tool, than with the hard or soft balls (p=.5). This aspect would be further exained in the following subsection. Difficulty to assess grasp quality [=easy; =difficult] Grasp quality assessent difficulty [=easy; =difficult] 1 1 1 1 1 1 1 1 1 Force-Feedback No Force-Feedback Force-Feedback Influence Alone Soft Ball Hard Ball Pistol Grip Tool Fig.. Group ratings on difficulty to assess grasp quality reotely during operations. Dependance on the object to be grasped. Opt.Vis.+F.-F. Obs.Vis.+F.-F. Opt.Vis. Obs.Vis. Fig. 11. Group results for the difficulty levels to assess grasp quality under different experiental feedback condition 379
Fig. 11 shows that a good vision on the worksite, along with the force-feedback allows the best assessent of the actual grasp situation. With decreasing levels of feedback inforation toward obstructed vision without forcefeedback, the assessent of the grasp perforance becoes increasingly difficult. For exaple, obstructed vision without force-feedback is significantly ore difficult than optial vision with force-feedback (p<.1). A ore detailed exaination to isolate the effects of visual- and forcefeedback has shown that visual-feedback played a relatively inor role with p=., whereas force-feedback played a significant role in grasp quality assessent with p<.1. ) Percentage of correct assessents of grasp success As shown in Fig. 1, the pistol grip tool deonstrated % correct assessents. This is likely due to the coplex and rigid shape, which provides the operator with ore clues to arrive at correct judgents. On the other hand, the soft ball was deeed the ost difficult to assess. One ain reason, as pointed out by the test subjects, was the difficulty to tell if a fir grasp has been achieved as a result of a soft, deforable shape. This shows the liitation of a lowperforance force-feedback syste in help the operator to clearly distinguish the object boundary when the object is highly deforable. % Correct Assessents (per trials) 95 9 5 75 7 5 Soft Ball Hard Ball Pistol Grip Tool Fig. 1. Group distribution for the condition of obstructed vision with force-feedback. It shows the percent of correctly estiated grasp situations, within each trial consisting of repetitions for soft ball, hard ball and pistol grip tool. Fig. 13 shows that assessents were ost difficult with obstructed vision alone and best with optial and/or obstructed vision with force-feedback. Upon closer exaination, as shown in Fig. 1, it is clear that assessent difficulty is direct a result of the lack of force-feedback within the task (p=.11), whereas No effect directly due to the different levels of visual-feedback. D. Grasping tasks success rate As shown in Fig. 15, with the aid of force-feedback inforation, teleanipulators can achieve nearly % success in grasping task for objects of different shapes and rigidity. This copares favorably against data with no forcefeedback is available, where error rates as high as % have been observed. % Correct Assessents (per trials) % Correct Assessents (per trials) 95 9 5 75 7 5 95 9 5 75 7 5 Opt.Vis.+F.-F. Obs.Vis.+F.-F. Opt.Vis. Obs.Vis. Fig. 13. Group distribution of different feedback conditions. It shows the percentage of correctly estiated grasp situations, consisting of repetitions within each trial. Force-Feedback No Force-Feedback Fig. 1. Group distribution showing the percentage of correct assessents by the test subjects. A coparison is ade here between feedback conditions with and without force-feedback. % Good graps (per trials) 95 9 5 75 7 5 Opt.Vis.+F.-F. Obs.Vis.+F.-F. Opt.Vis Obs.Vis Fig. 15. Percentage of successful grasps within a repetitions trial. Boxplot for the cobined results of the entire group is displayed over the four different experient conditions for all object types. V. DISCUSSION AND FINAL OBSERVATIONS Both visual- and force-feedback have shown strong influences on perceived difficulty of the grasping task, with vision playing a slightly ore significant role. This is in line 375
with previous findings reported, e.g. in []. When the vision is obstructed, the inclusion of force-feedback helps significantly reducing task load, as copared to optial vision without force-feedback. Force-feedback can also serve the role of a backup feedback source for a fault tolerant redundancy syste, necessary for safety-critical applications such as space telerobotics. In this case, the reaining, unipaired feedback channel changes its role fro a cooperative eber of a cobined feedback to a redundant channel [1]. This also confirs previous findings that forcefeedback akes a bigger contribution to lowering task load with less visual feedback inforation []. The test subjects increased reliance on the virtual robotic hand viewer in obstructed view conditions points to the need for such features in future space teleoperation ipleentations. Fro an investigative point of view, effects of view obstruction, without the aid of a virtual viewer, should be exained, to further distill the effects of force- and visual-feedback. The for closure that can be achieved with the pistol grip, ay have contributed, at least in part, to its associated grasp quality assessent success. This result has also deonstrated the iportance of tool for design to help iprove ease of tasks perfored in space. This paper has shown that force-feedback plays a critical role in helping the operators assess the quality of the grasp. Fig. 1 gives clear evidence of force-feedback as a key factor in correct grasp success/failure assessent. It was also shown that good vision contributes to this assessent ability. Fig. 11 and Fig. 13 show that a good vision, together with force-feedback produces nearly % correct assessents of the grasp success, whereas for other setups without forcefeedback, this rate can drop down to %. It should be specifically entioned here that this is true for a case in which only 9 Hz of up-dated signals are produced by the CyberGrasp finger exoskeleton. The fact that force-feedback contributed less to task perforance iproveent with the deforable object, has shown the liitation of the low-perforance force-feedback device ipleented here. While local ipedance control helped to achieve stability, the low update rate on the asterside still prohibits the rendering of sall changes of contact forces and high dynaic range. Finally, aside fro the data shown here, it should also be entioned that different strategies of grasping have been observed for subjects of different level of expertise in this experient. The expert-level subject seeed to focus during each grasp trial on finding a best and optial grasp. The novice subject in contrast seeed to use the first few grasps to find a best average suitable grasp. Once a good grasp type was deeed discovered, the novice subject repeated the sae grasp for the reainder of the sae experient (e.g. grasp of a soft regular shaped object with optial visual feedback with force-feedback). VI. CONCLUSION This work categorized the benefits of visual- and forcefeedback in teleoperated grasping through a series of grasping perforance evaluation etrics. Two findings in this paper have ade this work worthwhile to this space teleoperation project. First, through the series of test etrics, it has been clearly shown, that with the aid of local jointlevel ipedance control on the end effector, even with liited bandwidth and force-feedback channels, forcefeedback still deonstrated clear benefits in grasping hard objects of regular and coplex shapes. The perforance benefits have been particularly verified in grasp quality assessent, and reducing grasping task difficulty. Second, a perforance border has been identified, where the liitations in bandwidth and force-feedback inforation would liit grasp perforance. This occurs with the grasping of deforable objects, as the operator cannot identify the object boundary, thus aking telegrasping ore difficult. On the other hand, these findings also suggest that tools for telerobots should be designed to for-fit specifically to the end effector to help iprove space teleoperation perforance. VII. FUTURE WORK Moving forward, this project would ai to build a ore clearly defined and extensive set of test etrics, based on the findings fro this study, to further clarify the effects of force-feedback in different operating environents, with ore grasp object shapes, a larger pool of test subjects, and ideally also with a range of perforances of force-feedback inforation. Furtherore, as real grasping tasks can only be perfored effectively with a robotic ar-hand syste. Investigations are underway to get a ore holistic view of the effects of visual- and force-feedback in a hand-ar syste. The user coand would be retrieved by eploying an ar exoskeleton syste developed for such tasks [15]. Finally, as deonstrated fro the experiental results of grasping deforable objects, finer force-feedback inforation can indeed iprove soe grasp operations. Therefore, it would be helpful to investigate the possibilities of iproving the force-feedback inforation available through new device developents, while eeting space teleoperation resource constraints. ACKNOWLEDGMENT The Authors would like to express our deepest thanks to Mr. Florian Schidt of DLR for his tireless support in constructing the proposed syste environent. Mr. Peter Meusel of DLR provided valuable logistics assistance throughout the project. Mr. Ralph Bayer provided the echanical designs for the FFH base adaptor. Dr. Hong Liu of DLR provided useful guidance and suggestions in the application of the FFH. Our sincere gratitude is also goes to Mr. Jan Geerse of Dutch Space, Leiden, The Netherlands, 3751
for providing the test location and on-site support. REFERENCES [1] B. Hannaford, L. Wood, D. A. McAffee, and H. Zak, Perforance evaluation of a six-axis generalized force-reflecting teleoperator, IEEE Transactions on Systes, Nan, and Cybernetics, Vol. 1, No. 3, May/June 1991, pp. -33 [] M. J. Massiino, and T. B. Sheridan, Variable force and visual feedback effects on teleoperator an/achine perforance, in Proceedings of the NASA Conference on Space Telerobotics, Pasadena, CA, USA, 199, pp. 9-9 [3] M. L. Turner, R. P. Findley, W. B. Griffin, and M. R. Cutkosky, Developent and testing of a teleanipulation syste with ar and hand otion Proceedings of the ASME International Mechanical Engineering Congress and Exposition, Dynaic Systes and Controls, 9, pp. 57-3 [] O. Gerovichev, P. Marayong, and A. M. Okaura, "The effect of visual and haptic feedback on anual and teleoperated needle insertion," Proceedings of the Fifth International Conference on Medical Iage Coputing and Coputer Assisted Intervention -- MICCAI, Lecture Notes in Coputer Science (Vol. ), T. Dohi and R. Kikinis, Eds.,, pp. 17-15 [5] P. Richard, G. Burdea, D. Goez, P. Coiffet, A Coparison of haptic, visual and auditive force feedback deforable virtual objects, in Proceedings of the Fourth International Conference on Artificial Reality and Tele-Existence, Tokyo Japan, 199, pp. 9- [] F. Didot, P. Schoonejans, and R. Stott, Eurobot Underwater Model Testing the Co-operation Between Huan and Robots, The 9th European Space Agency Workshop on Advanced Space Technologies for Robotics and Autoation ASTRA, The Netherlands, Noveber -3, [7] C. H. Borst, M. Fischer, S. Haidacher, H. Liu, G. Hirzinger: DLR hand II: experients and experiences with an anthropoorphic hand. IEEE International Conference on Robotics and Autoation, 3 pp.7-77 [] H. Liu, K. Wu, P. Meusel, N. Seitz, G. Hirzinger, M. Jin, Y. Liu, S. Fan, T. Lan, and Z. Chen, Multisensory five-finger dexterous hand: The DLR/HIT Hand II, in IEEE/RSJ International Conference on Intelligent Robots and Systes,, pp. 39 397. [9] Z. Chen, N. Lii, S. Fan, M. Jin, and H. Liu, Experiental study on ipedance control for the five-finger dexterous robot hand DLR-HIT II, anuscript in preparation [] Virtual Technologies Inc., Virtual Hand v.5 Prograer s Guide, 1 [11] H. Liu, J. Butterfass, M. Grebenstein, DLR-Hand II: next generation of a dexterous robot hand, in Proceedings of the 1 IEEE International conference on Robotics I& Autoation, 1, pp: 9 11 [1] R.M. Murray, Z. Li, and S. S. Sastry, A Matheatical Introduction to Robotic Manipulation, CRC Press, 199 [13] A. Albu-Schaffer, C. Ott, and G. Hirzinger, A unified passivitybased control fraework for position, torque and ipedance control of flexible joint robots, The International Journal of Robotics Research, Vol., No. 1, 7, pp. 5-1 [1] W. Elenreich, Sensor Fusion in Tie-Triggered Systes, PhD Dissertation, Technische Universität Wien, Institut für Technische Inforatik, Vienna, Austria, [15] A. Schiele, Case Study: The ergonoic EXARM exoskeleton, in Wearable Robots: Bioechatronic Exoskeletons, J.L. Pons, Ed.: John Wiley & Sons Ltd.,, pp. 55 375