Torque-controlled light weight arms and articulated hands - do we reach technological limits now?

Transcription

1 Torque-controlled light weight arms and articulated hands - do we reach technological limits now? G. Hirzinger, N. Sporer, M. Schedl, J. Butterfaß, M. Grebenstein DLR German Aerospace Center Institute of Robotics and Mechatronics, D Wessling, Germany Gerd.Hirzinger@dlr.de, Abstract. Based on the long-term goal robonauts for space and the experiences DLR has gained so far in space robotics, the talk describes recent design and development results in DLR s robotics lab towards a new generation of mechatronic ultra-light weight robots with articulated hands. The design of fully sensorized joints with complete state feedback which has led to 7dof torque-controlled, soft robot arms with definable impedance in cartesian space, is outlined. The just recently completed third light-weight arm generation is based on a fully modular carbon fibre shell structure, newly designed motors with exciting characteristics, and it gives the feeling that it comes close to the limits of what is technically feasible today; in the same way the second generation of DLR s most highly integrated 4 finger-hand with its 13 actuators was completed now and adapted to the new arms. Thus it is hoped that major steps towards a new generation of space as well as service and personal robots have been achieved. 1 Introduction The space robotics background When comparing human skills with those of present-day robots of course human beings in general are by far superior, but when comparing the skill of an astronaut in a clumsy space-suit with that of the best available robot technology, then the differences are already going to disappear, the more if there is a remote control and monitoring capability on ground with arbitrarily high computational and human brain power. Although it is not clear today when a multi-fingered robot hand might be as skilled as the human hand without a thick glove and when (if ever) a robot might show up real intelligence and autonomy, it nevertheless is obvious that even with today`s technology and the available telerobotic concepts based on close cooperation between man (e.g. the ground operator) and machine there are many tasks in space, where robots can replace or at least augment human activities with reduced cost at least from a long-term perspective.

2 Thus we are convinced that automation and robotics (A&R) will become one of the most attractive areas in space technology, it will allow for experiment-handling, inspection, maintenance, assembly and servicing with a very limited amount of highly expensive manned missions (especially reducing dangerous extravehicular activities). And the expectation of an extensive technology transfer from space to earth seems to be more justified than in many other areas of space technology, the more as space is a driver for a new generation of ultralight weight robots with minimal power consumption. 2 DLR s early space robot projects Our first big experience with space robotics had been ROTEX ( Fig. 1 ) the first remotely controlled robot in space. It flew with Spacelab-Mission D2 inside shuttle COLUMBIA in April 93 and performed several prototype tasks (e. g. assembly and catching a free-floating object) in different operational modes, e. g. remotely programmed, but also on-line teleoperated by man and machine intelligence. Its success was essentially based on multisensory gripper technologies local autonomy using the above sensory feedback capabilities predictive graphics simulation compensating for 5 7 seconds Fig. 1 ROTEX the first remotely controlled space robot flew with shuttle COLUMBIA in 1993

3 We gained our second big space robot experience with NASDA s ETS VII project, the first free-flying space robot, who was operable for around two years. In April 99 we got the permission by our Japanese friends to remotely program and control their robot from Tsukuba / Japan. The project called GETEX (German Technology Experiment) was again very successful (as was the whole ETS VII mission); our goals in particular had been: To verify the performance of an advanced telerobotic concept (MARCO, see e.g. [10]), in particular concerning the implicit task level programming capabilities as well as the sensor-based autonomy and world model update features. A highlight was indeed the tele-programming of a peg-in-hole task, where in the virtual world we intentionally displaced the standby position of the peg from where the robot had to fetch it. Vision processing on ground using NASDA s tracking markers on the task board and the Jacobian matrix learning beforehand based on real images caused the ETS VII robot to automatically and perfectly adapt to the unexpected situation. The peg-in-hole insertion as such (taking into account the fairly high tolerances) was less critical and of course made use of NASDA S compliant motion commands. To verify 6 dof dynamic models for the interaction between a robot and its free-flying carrier satellite. A major part of the GETEX experiment time was allocated to these experiments, which consisted of a series of manoeuvres carried out by the manipulator while the attitude control system of ETS-VII was switched off. Fig. 2 ETS VII ground control via the task-level programming system MARCO In such a mode of operation, a space robot consisting of a manipulator and a satellite is generally considered to be free of external forces. The robot therefore is assumed to have constant angular momentum, due to the law of the conservation of angular momentum, which means that if the arm moves and thus introduces angular momentum into the system, the satellite reacts with a compensating motion. The amount of satellite rotation produced depends on the mass and inertia of the bodies which constitute the system. The description of a TCP trajectory in orbit-fixed co-ordinates, as it is necessary e.g. for the capturing of a defect satellite, has to ac-

4 count for the satellite reaction. The experiments conducted during the GETEX mission aimed at a verification of the existing models of free-floating space robots and at the identification of the dynamic model parameters such as the satellite inertia tensor. A further goal was to obtain some insight into the nature and importance of disturbances acting on a robotic satellite in low Earth orbit and to gather data for the future design of controllers which will combine the manipulator motion control with the satellite attitude control. Therefore, a variety of different manoeuvres were executed (an example of which is shown in Fig. 3), which include simple point-topoint operations and closed-loop re-orientation manoeuvres, sequences during which only one joint was active at a time as well as sequences during which all joints were moving simu ltaneously. Fig. 3 Example of a Dynamic Motion manoeuvre carried out during the GETEX mission. The shaded robot indicates the reference position. The satellite reaction to the arm motion is scaled by a factor of 10 in this picture. The major constraint, due to mission security aspects, was the maximum satellite attitude error allowed by NASDA, which was limited to ±1.0 around each axis, and the fact that the maximum tool center point velocity was limited, too. Furthermore, the reaction wheels were turning at a very low but non-zero constant velocity during the experiments, which introduced undesired torques into the system. In total, over 110 minutes of dynamic motion experiments have been carried out, of which 52 minutes have been spent in free motion mode. The remaining time was used to repeat the experiments in reaction wheel attitude control mode for verification purposes. Evaluations of the measurement data confirmed the need to account

5 for external disturbance forces acting on the satellite, such as the gravity gradient torque and magnetic. 3 Future steps Germany has made decisions as to where the next steps in space robotics will move. There are two major directions: 3.1 Systems at the international space station ISS The most remarkable system at ISS is Canada s Mobile Servicing Center, a three arm system with a long ( 17m) arm and two smaller arms ( 3,5) on top of it. Canada has put a major part of its space budget into this remarkable technology; nevertheless in the past we have repeatedly criticized, that the arms were only controllable by astronauts typically with 1mm/see, at least in our opinion a waste of time. As an example, in July 2001 (the base arm was just installed) two astronauts needed 6 hours to grasp the so called airload out for the shuttle bay and mount it at the ISS. We are very happy that a close cooperation exists meanwhile between CSA and DLR aiming at an efficient ground control of the arms implying the advanced telerobotic system MARCO. Fig. 4 The Canadian multi-arm ISS robot system (courtesy CSA) Simulated Airlock assembly with our telerobotic system MARCO In addition, a German Space station robotic project ROKVISS has just started (Fig. 5). It aims at the qualification of DLR s newest light weight joints (sect. 6 and 7) by

6 hand of a 2 dof arm on the ISS by the end of 2003 and it also plans a demonstration of telepresence technologies with real-time stereo-video transmission and tactile feedback via the ground station. Fig. 5 The two-joint qualification and telepresence experiment ROKVISS on ISS aims at the verification of basic technologies for future robonaut applications Indeed what we are still missing in space are fast signal transmissions via just one relay satellite yielding coverage times of around 40 minutes and round trip delays of 0,5 sec. thus allowing even haptic feedback. Realistic telepresence will be needed e. g. too, when astronauts have to be remotely investigated using robotically guided 3D-ultrasound probes. Fig. 6 Telepresence is gaining more importance in space

7 3.2 Free-flying systems We have performed an extensive technology study on an experimental servicing satellite (ESS), which applies robotics to solve the problem of servicing a non-cooperative target in or near to a geostationary orbit, a region of space still out of reach to manned spaceflight. A three-month demonstration flight of such an ESS has been planned and all phases of its mission have been defined. These include the acquisition, inspection and servicing of an orbiting satellite through to parking it in a graveyard orbit. For that external servicing task high interactivity between man and machine is required, because the remote environment will be mainly unknown. The telerobotic system MARCO will be used to give the system the local autonomy by intelligent sensor data processing. Because all the satellites, built so far, are not equipped for servicing, the final stages of approach and the subsequent capture of the target are the most critical phases of the mission. To simulate the dynamic behavior of the chaser during robot motions, we have arranged two KUKA robots as shown in Fig. 7. Robot B is used to carry out the capturing task, Robot A emulates the entire dynamic relation between the chaser and the target satellite, where the dynamic coupling with the AOCS is included. Capture Tool and Sensors Video, F/T and Distance Data Model of Apogee Motor Tool Center Point Sensor - control Dyn. Model Robot A of ESS - Robot B Manipulator TCP Difference Dyn. Model Target Motion Dyn. Model of ESS AOCS Model Fig. 7 ESS simulation and testbed The manipulator of ESS, equipped with a multi-sensory capturing tool, must follow the residual movements of a selected object on the target (e.g. the main thruster or apogee motor ) by means of an image processing system whose data are passed through an extended Kalman filtering process. With the robot controller monitoring

8 laser distance sensor values, force, torque and travel, the capture tool is inserted into the cone of the thruster (Fig. 8, Fig. 9). Fig. 8 Tracking of target s apogee as seen from the wrist-mounted hand camera. The wireframe model of the target is projected into the live video image at the currently estimated pose Fig. 9 An artist s view of ESS, catching the apogee of TV-Sat-1, one of whose solar panels did not open The tumbling target would not necessarily be in a geostationary object. Indeed a few space systems no longer controllable have been identified on lower orbits, which might become dangerous for earth, as they will not completely burn out when passing the atmosphere. Thus grasping them with a robot (e.g. with a more articulated hand if no apogee motor is usable) and drawing them down in a well-

9 defined way might become an important service in the framework of future garbage collection systems (Fig. 10). Fig. 10 Catching a worn-out satellite to render it harmless Independent of this special application the next European step should be a new free-flying space robot mission (experimental or operational), for which however no final decisions exist at present. Our proposal for a two-arm/hand freeflyer is shown in (Fig. 11). Fig. 11 DLR s Robonaut concept for a free-flying robot satellite with two arms and two articulated hands 4 Light weight robot concepts What we definitely need for space (as a technology driver) but also for the wide variety of future terrestrial service robot applications, are soft and sensorcontrolled light-weight arms (in contrast to the stiff and heavy industrial solutions) and articulated, multifingered hands, which come closer and closer to the delicate human performance. Two of these arms combined with an arrangement of a stereo camera pair tends to provide such a system with humanoid appearance and thus provokes the robonaut terminology. NASA has recently presented remarkable results in this context.

10 From the experiences we made with ROTEX, we see the need that at least a small size space robot (1-2m size) should not only sustain itself on ground without suspendings, but should also be able to work in a 0g gravity mode, i.e. compensating gravity by imposing appropriate bias torques onto the joints. And a free-flying space robot should be able to make itself compliant when getting in contact with a floating object to be grasped or docked at. These are only a few arguments that cry for a new generation of torque-controlled, 7dof light-weight arms with minimal power consumption. By the way, to compare light-weight actuators in a fair way, it is of crucial importance to take the maximum motion speed in account (as proposed in [2]). With our third generation of light weight arms presented in this paper we try the approach the technological limits of what is feasible today. DLR s light-weight robots from the beginning have been kinematically redundant (7dof) and joint-torque-controlled [2], [3]. LWR I (Fig. 12, left) with its 18kg weight, a nearly 1:2 load to weight ratio and its carbon fibre grid structure links was a highly mechatronic arm, but its doubleplanetary gearings with a 1:600 reduction turned out to be too critical in terms of tolerance-safe manufacturing. In addition its inductive torque sensing was critical, too, in terms of complexity and robustness. In LWR II (Fig. 12, right) we went back to harmonic drive gearings, a strain-gauge based torque measurement system, embedded into a full state measurement and feedback system (motor position, link position, joint torque). Again all electronics (signal, power, control) was fully integrated into the arm weighing 17kg and carrying 8kg. Fig. 12 LWR I (left) and LWR II (right) With this arm, we were fully satisfied with its high-fidelity impedance controllability, but manufacturing and duplicating the arm was not yet simple enough.

11 Fig. 13 New DLR Light-Weight-Robot LWR III, virtual (left) and real (right) The development goal of our light weight robot generation III (Fig. 13) was to realize a system which builds up on all the experiences made with robot I and II, but exploits the remaining potentials to approach the technical limits. 5 Modular arms Thus our new robot arm concept (with future marketing and sales of the arm in mind) aimed at a completely modular assembly system with only a few basic components concerning joint mechanics, electronics (i.e. mechatronics) and links, so that completely different configurations can be composed in a short time. And, as in LWR II the Harmonic drive gearings had been redeveloped in close cooperation with the Japanese manufacturer into a drastically weight-reduced ALU version (60% weight saving), it was clear that now the key issues to be attacked were a new motor generation, ultralight-weight brakes and bringing the links back to carbon fibre technology (LWR I had already a grid-type carbon structure, but as already stated unsatisfactory joint drives). Indeed the new concept is based on a fully modular joint-link-assembly system, with only a few basic components, namely three one-dof robot joint link types and a two-dof wrist joint. Still there remain two alternatives: a symmetrical version (Fig. 14) and an asymmetrical version (Fig. 15) which is particularly interesting if folding the arm (e.g. for transporting it into the outer space) is a key issue.

12 Fig. 14 Symmetrical Robot Fig. 15 Asymmetrical Robot

13 These modularity concepts were supported by a powerful kinematic-dynamic analysis and design software, based on concepts of concurrent engineering, which allow to assemble any type of robot using the link component library (with continuously adjustable link lengths) and a set of (typically 3) available gearing types and a set of (typically 3) available motor types. When an arm has thus been virtually composed (a matter of a few minutes) the arm tip (with a definable mass fixed to it) may be immediately moved around with a 6 dof input device like a Space Mouse or a Phantom, both devices allowing also to impose forces and torques on to the arm s tip; and the kinematic-dynamic simulation in real-time shows the arising forces/torques in the joints (Fig. 16, Fig. 17), and - after passing FEM calculations give hints on necessary material strengths e. g. in bearings and carbonfibre structures. 7-Joint Robot Kinematics Light Weight CFK Link Structures Online Dynamics Simulation and Visualization Online Torque Monitoring Highly Integrated Drives Online Interactive Robot Control Fig. 16 A kinematic-dynamic simulation tool for optimizing the design Fig. 17 Concurrent Engineering for the arm configuration

14 Another result came out from the kinematic-dynamic simulations: a ball-shaped two axis wrist joint (Fig. 18), imitating the human wrist, but showing up much higher mobility. Indeed the most important joints of a robot are the wrist joints. Manipulability of the robot is significantly dependent on kinematic configuration. If the distance between wrist-pitch axis to tool-center-point is short, changes of the orientation at the tool-center-point do not lead to big movements of the lower robot joints. And every gram which can be saved in the design of the wrist is a gain of payload. The new extremely light-weight design of the wrist enables two kinematic configurations. A common configuration with roll-pitch-roll axis and optional a roll-pitchpitch configuration which is easy to reconfigure. The short standard flange is simply to be replaced with an extended 90 bended flange. The cardan joint avoids singularities in a stretched position. This position is often reached while finemanipulating with articulated hands like DLR s 4-finger-hand. (roll-)pitch-roll Fig. 18 Ball like wrist joint (roll-)pitch-pitch In general the modularity concept gives us a number of advantages, e. g: rotation symmetric (Fig. 19) components few single parts, short force transmission from bearing to off-drive connection, pitch and roll joints identical, Big hollow shaft in all joints with up to 30 mm diameter, thus optimal cable and plug links inside the arm.

15 modular drive between two links Fig. 19 Modular drives in the new arm 3D-CAD-design of the modular drive 6 The breakthrough in motor technology As a matter of fact, in the past robot manufacturers have taken the best available motors off the shelf for their robots, without being optimized for robotic applications (comparably slow rotational speed though high dynamics, permanently reversing operation around zero speed) and aiming at minimal weight and power losses. Thus we have gone through a two years concurrent engineering and optimization process who took in account al the electromagnetic and other physical effects, short copper paths, optimal coil winding and coil filling aspects between the magnetic iron poles. It came out of this optimization process that the stator poles had to be subdivided and winded separately (Fig. 20). Fig. 20 Concurrent Engineering for the actuator design

16 The result is DLR s high energy ROBOdrive with only half of the weight and half of the power losses of the (to our knowledge) best available motors (see Fig. 22). Simulative and experimental results differed by only 1-2%. Fig. 21 Comparing DLR s ROBODRIVE with the best commercially available motors Fig. 22 3D CAD-design of DLR s ROBODRIVE Fig. 23 DLR s new ROBODRIVE with one of the carbon fibre links

17 Three different sizes are now integrated into LWR III. Other important features and components as developed for LWR II [4] were transferred into the new arm III: the full state measurement and control in all joints, as are - straingauge-based torque-sensing and torque feedback in a 3 khz rate - motor position sensing using hall sensors - off-drive position sensing (formerly using optical principles, now moving alternatively to low-cost plastic potentiometer or high-end capacitive sensors, respectively) the light-weight ALU-Harmonic drive with 60% weight reduction. The arm has just been assembled and is going to become operable now. Only three cables inside the arm are needed to connect all joints with power supply and the external PC controller. What presently remains to be done is e.g. - a decision on which types of safetybrake should be finally used. While in [4] a weight reduction of a commercial safety brake from 281 to 155 g was reported, we have in the meantime developed piezoelectric brakes weighing only 70g (Fig. 25) Fig. 24 Torque sensor Fig. 25 A piezo-brake with only 70 g overall weight

18 However as some moistness problems are inherent with piezobrakes, we have in addition gone through another concurrent engineering and optimization process leading to an extremely weight-reduced electromagnetic brake version, too ( 30g for the ball-shaped wrist joints) - miniaturizing the arm electronics further, going far beyond well-known SMDtechnologies (with tiny chips anyway) to chipsize-packaging and flipchip (thus microsystems) technologies by support of leading German Fraunhofer- Institutes specialized on microintegration. All electronics (except the central PC controller board), is integrated into the arm. 7 Impedance controlled arms The control structures for this arm have been mainly developed and verified by hand of LWR II (Fig. 26). In particular a flexible joint model is assumed. Fast and reliable methods for the identification of the joint model parameters (joint stiffness, damping, and friction) were developed, while the rigid body parameters are directly generated from the mechanical CAD programs [6]. This leads to an accurate simulation of the robot dynamics, so that the controller structures can be developed and tested directly in the simulation. Fig. 26 controller architecture The first stage in the controller development was a joint state feedback controller with compensation of gravity and friction. The state vector contains the motor position, the joint torques, as well as their derivatives. The conditions for the passivity of this structure were derived in [6]. Under these conditions, the global asymptotic stability of the controller can be proven [7].

19 By the appropriate parameterization of the feedback gains, the controller structure can be used to implement position, torque or impedance control. In the last case, the gains of the controller are computed in every Cartesian cycle, based on the desired joint stiffness and damping, as well as depending of the actual value of the inertia matrix. Hence, this controller structure, fulfils the following functionalities: It provides active vibration damping of the flexible joint structure. It maximizes the bandwidth of the joint control, for the given instantaneous values of the inertia matrix. It implements variable joint stiffness and damping. Based on this joint control structure, three different strategies for implementing Cartesian compliant motion have been realized: admittance control, which accesses the joint position interface through the inverse kinematics; impedance control, which is based on the joint torque interface; and Cartesian stiffness control, which accesses the joint impedance controller. To combine the advantages of the last two methods regarding geometric accuracy and high bandwidth, a new control method was implemented, which consists of an impedance controller enhanced by local stiffness control [8]. This structure consistently takes into account the fact that the cycle of the joint control loop is typically by one order of magnitude faster than that of the Cartesian loop. It uses the high bandwidth of the joint impedance controller to improve the performance of the Cartesian impedance control. The control strategy is strongly motivated by the way in which humans achieve their compliant manipulation skills. It is well known that the mammalian supervisory control loop has a significant time delay. The passivity property of the muscles, as well as the fact that the intrinsic stiffness and damping can be varied according to the stiffness of the environment (in this case by the contraction of antagonistic muscle pairs), seems to play an essential role in preserving the stability and performance of the movement [9 ]. Fig. 27 LBR II balances a pole and reacts with a human

20 Fig. 27 shows a situation, where arm II automatically balances a pole and a human operator touches the arm and deflects the elbow joint using the arm s redundancy and impedance controlled soft behavior. The new impedance controller enhanced by local stiffness control shows up a better performance than classical impedance and stiffness control. Compared to admittance control, it has lower geometric accuracy, but higher bandwidth and impedance range. This makes it well suitable for applications where the robot is in contact with a stiff, unknown environment. 8 Articulated 4-fingered hands In 1997 DLR developed one of the first articulated hands with completely integrated actuators and electronics. This well known hand has been in use for several years and has been a very useful tool for research and development of grasping. The main problems remaining were maintenance and the many cables (400) leading out from the Hand. The experiences with Hand I accumulated to a level that enabled us to design a new hand according to a fully integrated mechatronics concept which yields a reasonably better performance in grasping and manipulation and therefore accelerates further developments. Due to maintenance problems with Hand I and in order to reduce weight and production costs the fingers and base joints of Hand II were realized as an open skeleton-structure. The open structure is covered by 4 semi shells and one 2-component fingertip housing realized in stereolitography and vacuum mold. This enables us to test the influence of different shapes of the outer surfaces on grasping tasks without redesigning finger parts. The main target in developing Hand II from the beginning has been the improvement of the grasping- performance in case of precision- and power-grasp. Therefore the design of Hand II was based on performance-tests with scalable virtual models as seen in Fig. 29.

21 Fig. 28 DLR's Hand I. Fig. 29 Optimization of kinematics with scalable hand model On the other hand performing precision-grasps/fine-manipulation requires huge regions of intersection of the ranges of motion and the opposition of thumb and ring-finger (Fig. 30). Therefore Hand II was designed with an additional degree of freedom which enables to use the hand in 2 different configurations. This degree of freedom is a slow motion type to reduce weight and complexity of the system. This adaptive palm motion of the first and the fourth finger are both realized with just one brushed dc actuator using a spindle gear. In Fig. 32 real precision and power grasp are shown with the 13 dof hand II. Fig. 30 Simulation of Hand II in power grasp and fine manipulation configuration Fig. 31 Differential bevel gear of the new basejoint. The three independent joints (there is one additional coupled joint) of each finger are equipped with appropriate actuators. The actuation systems essentially consist of brushless dc-motors, tooth belts, harmonic drive gears and bevel gears in the base joint. The configuration differs between the different joints. The base joint

22 power converter sensor circuitry power converter sensor circuitry power converter sensor circuitry with its two degrees of freedom is of differential bevel gear type, the harmonic drive gears for geometric reasons being directly coupled to the motors. The differential type of joint (Fig. 31) allows to use the full power of the two actuators for flexion or extension. Fig. 32 Precision and powergrasp with Hand II Since this is the motion where most of the available torque has to be applied, it allows to use the torque of both actuators jointly for most of the time. This means that we can utilize smaller motors. The actuation system in the medial joint is designed to meet the conditions in the base joint when the finger is in stretched position and can apply a force of up to 30 N on the finger tip. Here the motor is linked to the gear by the transmission belt. A dexterous robot hand for teleoperation and autonomous operation needs (as a minimum) a set of force and position sensors. Various other sensors add to this basic scheme. Each joint is equipped with strain gauge based joint torque sensors and specially designed potentiometers based on conductive plastic. Besides the torque sensors in each joint we designed a tiny (20mm diameter, 16mm height) six dimensional force torque sensor for each finger tip with full digital output. The force and torque measure ranges are 10 N for Fx and Fy, 40 N for Fz, 150 Nmm for Mx, My and Mz respectively. sensor ADC circuitry 8 ch sensor circuitry ADC 8 ch Motor ADC 8 ch Motor Motor ADC 16 ch communication controller power supply serial to hand base Fig. 33 The fingertip sensor Fig. 34 Electronics and communication in a finger

23 All electronics needed locally is integrated into the hand. However the control of the fingers and the hand is done by an external computer. In order to use the hand freely on different manipulators and to reduce cables and the possibility of noise in the sensor signals, we decided to design a fully integrated serial communication system. Each finger holds one communication controller in its base unit. This controller is responsible for the collection and distribution of all information of interest. Furthermore it does some reasonable signal processing. It collects the data of all five ADCs per finger with together 40 channels of 12 bit resolution each and transmits these data to the communication controller in the hand base (see Fig. 35 and Fig. 36). On the other hand it distributes the data from the control scheme to the actuators for finger control. The communication controller in the hand base links the serial data stream of each finger to the data stream of the external control computer. By this hardware architecture we are able to limit the number of external cables of DLR's Hand II to a four line power supply and an eight line communication interface since the data is transmitted via differential lines. This interface even provides the possibility of using a quick-lock adaptor for autonomous tool exchange. Reducing external cabling from 400 (in Hand I) to 12 here, is one of the major steps forward in our new hand. fingers 1-4 serial Motor hand dof Communication Controller (FPGA) serial/differential IEEE Fig. 35 The communication controller in the hand base links the fingers to external computers

24 Fig. 36 Signal pre-processing and transfer in the hand II When a robot hand performs any fine manipulation, there is always need that the fingertip should be soft in the direction normal to the contact surface and hard tangential to the contact surface. Thus the impedance should be adaptable to the orientation of the fingertip. Therefore, a cartesian impedance controller has been developed, where the fingertips behave like a programmable spring. Fig. 37 Block diagram of the hand s cartesian impedance controller

25 9 Conclusion After more than 10 years of torque-controlled light weight robot development we are near to finishing the third generation, where we have tried to use all presentday available simulation and computational technologies to approach the technical limits. Taking into account its weight of 13-14kg, its typical power consumption of little more than 100 Watt, its load capability of around 10kg, its motion speed based on maximal joint speeds of 180 /sec, it is probably one of the lightest robots that have been built so far. Arm III is equipped now with DLR s articulated hand II. With its 13 active joints integrated into the palm it allows to verify strong power grasps in the same way as delicate fingertip grasps. A special adapter was designed so that the only 12 wires leading out of the hand are fully guided inside the hollow shafts of the robot joints. To our knowledge this extremely high degree of mechatronic integration in arm and hand has not been realized before. It is the basis for DLR s future space robot developments aiming at systems similar to NASA s robonaut [5]. Torque-controlled arm and hand technologies are also the basis for a new surgical robot system, the more as these new robot systems can be used as "force-reflecting hand controllers", too. This observation led to the decision to combine the above arm and hand technologies to create an innovative surgical robotics system comprising five identical, extremely slender, variably compliant, but also very powerful robot arms, where two arms are used as force-reflecting hand controllers, two as instrument carriers and one as an endoscope guide system (Fig. 8). This system is supposed to help minimally invasive surgery make a major breakthrough in the future, because it not only enables the surgeon to intuitively transfer his finger and hand movements to the inside of the patient, (just like when a large, traumatic, opening is made in the body), but also restores his tactile perception. Fig. 38 A surgical robotics system for the future, based on LWR III and Hand II technologies

26 Acknowledgement We would like to thank the German space agency and Bavaria s ministry of technology, who has made mechatronics a key topic of Bavaria s high tech offensive with light weight robotics and articulated hands as central demonstrators. References [1] G.Hirzinger, B.Brunner, J.Dietrich, J.Heindl, ROTEX - The First Remotely Controlled Robot in Space, IEEE Int. Conference on Robotics and Automation, San Diego, California, May 8-13, 1994 [2] G.Hirzinger, J.Butterfaß, M.Fischer, M.Grebenstein, M.Hähnle, H.Liu, I.Schaefer, N.Sporer,, A Mechatronic Approach to the Design of Light-Weight Arms and Multifingered Hands, IEEE Int. Conference on Robotics and Automation (ICRA 2000), San Francisco, April 2000 [3] G.Hirzinger, A.Albu-Schäffer, M.Hähnle, I.Schaefer, N.Sporer: On a New Generation of Torque Controlled Light-Weight Robots. ICRA International Conference on Robotics & Automation, Seoul, Korea, May 21-26, 2001, IEEE, COEX Seoul Korea, Conference Proceedings ICRA 2001, S , (2001) [4] G.Hirzinger, B.Brunner, K.Landzettel, I.Schaefer, N.Sporer, G.Butterfaß, M.Schedl: Space robotics DLR s telerobotic concepts, lightweight arms and articulated hands. 10 th International Conference on Advanced Robotics (ICAR 2001) Budapest, Hungary, August 2001 [5] M.Bluethmann, R.Ambrose, R.Askew, M.Goza, C.Lovechik, D.Magruder, M.A.Differ, F.Rehnmark: Robonaut: a robotic astronaut s assistant. 10 th International Conference on Advanced Robotics (ICAR 2001) Budapest, Hungary, August 2001 [6] A.Albu-Schäffer, G.Hirzinger: Parameter identification and passivity based joint control for a 7DOF torque controlled light weight robot. ICRA International Conference on Robotics & Automation, Seoul, Korea, May 21-26, 2001, IEEE, COEX Seoul Korea, Conference Proceedings ICRA 2001, S , (2001) [7] A.Albu-Schäffer and G.Hirzinger.: State feedback controller for flexible joint robots: A globally stable approach implemented on DLR's light-weight robots. IROS, 2000 [8] A.Albu-Schäffer and G.Hirzinger: Cartesian Impedance Control Techniques for Torque- Controlled Light-Weight Robots IEEE International Conference on Robotics and Automation (ICRA), May 11-15, 2002 Washington D.C. [9] R.Koeppe, A.Albu-Schaeffer, C.Preusche, G.Schreiber, G.Hirzinger: A New Genereation of Compliance Controlled Manipulators with Human Arm Like Properties. The Tenth International Symposium of Robotics Research, ISRR, Lorne, Victoria, Australia, Nov. 9-12, 2001 [10] Hirzinger, G., Brunner, B., Butterfaß, J., Fischer, M., Grebenstein M., Landzettel, K., Schedl., M.: Space robotics - towards advanced mechatronic components and powerful telerobotic systems. i-sairas 2001, 6th International Symposium on Artificial Intelli-

27 gence and Robotics & Automation in Space, St-Hubert, Quebec, Canada, June 18-22, 2001, Canadian Space Agency, CD i-sairas 2001, (2001) [11] Hirzinger, G., Sporer, N., Albu-Schäffer, A. Hähnle, M., Krenn, R., Pascucci, A., Schedl, M.: DLR s torque-controlled light weight robot III are we reaching the technological limits now? IEEE International Conference on Robotics and Automation (ICRA), May 11-15, 2002 Washington D.C. [12] D.Vischer, O.Khatib: Design and Development of Toruqe-Controlled Joints. In: Experimental Robotics I. ed. V.Hayward and O.Khatib, Springer-Verlag, pp , 1990 [13] P.Dario, E.Guglielmelli, C.Laschi, G.Teti (SSSA): MOVAID: a personal robot in everyday life of disabled and elderly people. Technology and Disability Journal, No. 10, 1999, IOS Press, Olanda, pp [14] M.E.Rosheim: Design of an Omnidirectional Arm, Proceedings of IEEE Conference of Robotics and Automation Magazine June 1996, S