Micromega Dynamics

EXOSTATION Phase B : 7-DOF Haptic Exoskeleton Pierre Letier, More Avraam, Mihaita Horodinca, Samuel Veillerette(1) , Jean-Philippe Verschueren(2), Tanguy Fautre, Elvina Motard, Michel Ilzkovitz(3), Andre Schiele(4), André Preumont (1) (1) Active Structures Laboratory Université Libre de Bruxelles 50, Av. F.D. Roosevelt, B-1050 Brussels, Belgium Email: scmero@ulb.ac.be (2) Micromega Dynamics Parc Scientifique du Sart Tilman Rue des Chasseurs Ardennais B-4031 Angleur, Belgium Emai : info@micromega-dynamics.com Abstract EXOSTATION is an ESA project aiming at developing an exoskeleton, for an astronaut in microgravity, supporting the telemanipulation with force feedback of anthropomorphic robotic arms. Such robotic arms could be embedded in mobile robots used as first explorers in hostile environments or as assistants for EVA (e.g. Eurobot, Robonaut). A complete 7-DOF haptic control chain is currently under development, for demonstration purpose. The final haptic demonstrator will be composed of a 7DOF arm exoskeleton master, controlling a virtual slave robot interacting with a virtual environment. 1. Introduction In future space missions, robots could be used as first explorers in hostile environments or as assistants for Extra-Vehicular Activities (EVA). This will require a higher level of cooperation between astronauts and robots. For this, the use of a portable device that would procure the robot operator with force-feedback sensations (also called haptic sensations) would highly increase the easiness of the command task. In this context, ESA has launched the development of a humanoid servicing robot, called EUROBOT. Two control modes are foreseen to drive this robot: autonomous control and master-slave manual control. For this last purpose, an arm-exoskeleton has been developed at ESTEC [1][2][3]. However, the proposed (3) Space Application Services Leuvensesteenweg, 325 B-1932 Zaventem, Belgium Email: info@spaceapplications.com (4) European Space Agency Automation and Robotics Laboratory P.O. Box 299, 2200 AG Noordwijk, The Netherlands Emai : andre.schiele@esa.int exoskeleton did not provide force-feedback to the operator. The EXOSTATION project was launched to implement a complete multi-DOF telemanipulation system, including a force feedback exoskeleton master arm communicating with a simulated slave robotic arm. Due to technological complexity, the project was divided in two phases. Phase A was dedicated to the development of a fully integrated 1-DOF haptic chain representing one joint (master and slave) of the future telemanipulation system [4]. The purpose was to validate all the components and their integration. In parallel, several kinds of actuation technologies were compared to select the best fit for the future portable device [5][6]. Based on the results and experience gained in phase A, the aim of phase B is to develop a complete 7-DOF haptic control chain. It is composed of four main components that will be presented in more details in this paper (Figure 1). The first component is a 7–DOF exoskeleton playing the role of the master robot in the haptic loop. It is a fully portable device with a specific kinematics structure and on-joint actuation. The second component is the exoskeleton controller composed of the master controller (ECO) and local integrated joint controllers. These implement the control strategies and manage the communication links with the slave simulator. Figure 1: EXOSTATION Phase B, Overview The final component is a 3D Visualization Client based on OpenGL, which allows one to several users to visualize the state of the slave arm and provides a GUI to remotely control the simulator. The complete system is under development. However several preliminary tests have already been conducted on the exoskeleton master. The results obtained are presented below. 2. System Description 2.1 Exoskeleton Mechanical Design The master device, known as the Sensoric Arm Master (SAM), consists of a fully portable arm-exoskeleton mechanical interface with a serial kinematics, isomorphic to the human arm (Figure 2). SAM contains 7 actuated DOF from the shoulder to the wrist. That corresponds to the minimum number of joint allowing full immersion of the operator while keeping an acceptable mechanical complexity. The system has also 5 passive DOF (sliders) to be able to align each active joint of SAM with the biological joints of the arm in order to fit a wide range of user morphologies. The kinematics structure is chosen to be simple but at the same time, designed to present a maximized workspace without singularities. As shown in Figure 3, if 3 orthogonal rotational joints parallel to the reference axis (X,Y,Z) are used to represent the spherical joint of the shoulder (the initial configuration considered), singularities appear in the middle of the user workspace. In order to move these singularities away from the useful workspace to the frontier of the human workspace, joints configuration on the shoulder has been modified by two successive rotations to reach the final configuration proposed. The same principle has been used for the wrist joints. The weight of the structure is minimized in order to maximize user comfort. FEM analysis has been conducted in order to optimize the weight of the structure while maintaining its mechanical resistance and stiffness in order to achieve optimal haptic rendering. The current weight is 6.5 kg for the master device. Based on the mechanical design and results of the phase A of the project, in order to implement forcefeedback, each joint, is composed of a brushed DC motor coupled with a cable capstan and small reduction ratio gearbox . Each motor/reducer is chosen to be able to provide continuously around 1/20th of the human maximum torque of the corresponding joint except if the human arm torque in daily activities is lower (Table 1). This criterion is commonly used with other haptic devices which are available on the market [8]. Table 1: Torque output of SAM compared to daily activities and maximum human torque [9][10] Typical human arm torque in daylife (N.m) Max human arm torque (N.m) 1/20th max human arm torque (N.m) JOINT The third component is a Slave Simulator simulating on top of ODE (Open Dynamics Engine) [7] a 7-DOF actuated slave arm, cinematically equivalent to the exoskeleton. Continuous/ Stall torque (SAM's joint) (N.m) 1 4,38/19.74 10 134 6,7 2 4,38/19.74 9,6 115 5,75 3 5,97/26.94 3,1 60 3 4 3,64/16.41 3,8 72 3,6 5 2,22/7.74 0,4 9 0,45 6 0,37/1.73 0,38 21 1,05 7 0,37/1.73 0,25 20 1 Figure 2: Schematic view of the Sensoric Arm Master (SAM) Figure 3: Influence of the kinematic structure on the singularity level Due to the use of large size bearing with seals for the joint 3 and 5, these joints present a higher level of friction compared to the others. For this reason joint 3 and 5 present higher continuous torque to be able to provide sufficient friction compensation. The purpose of combining the two types of reducers is to achieve a high enough torque combined with high compactness, low friction and low backlash transmission. To limit the weight and the volume of the actuation part, a compact design is proposed with a full integration of the mechanical interfaces (Figure 4). In order to achieve teleoperation of the slave robot, position and torque measurement are provided at each joint, respectively by an incremental encoder and an integrated torque sensor based on strain gauge technology. All the motors of the arm-exoskeleton have been located on their corresponding joint’s link in order to reduce complexity and number of mechanical parts. Due to the kinematics equivalence, the control strategy is based on a joint to joint approach. Each joint of SAM controls individually the corresponding joint of the slave robot by exchanging the information of position and torque between the master and the slave at a 500Hz sampling rate. Based on the results of Phase A, the general form of the controller for one Master/Slave joint is represented in Figure 5. The principle is to exchange both torque ( hi, si) and position ( mi si) between master and slave to command the opposite side. The actuator set-points ( m and s) are created by local position and torque controllers. This scheme allows implementing control strategies from two channels (impedance control) to four channels that led to the best results obtained on the 1 DOF demonstrator in phase A [4]. Other pictures and videos of SAM can be found at http://www.ulb.ac.be/scmero . Figure 4: Integrated joint with DC actuator, encoder and torque sensor/conditioning 2.2 Exoskeleton Controller Figure 5: General form of the control strategy for one Master/Slave joint 2.2.1. Control Strategy The haptic control loop enables force feedback teleoperation between SAM Exoskeleton and the 7DOF cinematically equivalent slave robot (section 2.3) by the exchange of information between the two systems. Its purpose is to guarantee good performances in terms of human sensations and stable behavior for slave free motion or contact. To limit the impact of friction forces of joint 3 and 5 on the quality of haptic rendering, friction compensation algorithms, based on [11], are also considered in parallel to the previous scheme. Three different methods are implemented. The first one, called Motion Based method, uses a feedforward compensation based on a model of the friction (Coulomb and viscous) using a speed measurement: F c * (sign ) b *  The second, Force Feedback, (dashed line in Figure 5) implements a torque feedback based on a local torque measurement; finally the third one, the Hybrid method, merges the two previous in order to reduce the disadvantages of each of them. The Force Feedback is used for small velocities and the Motion Based for high. Switching between the two methods is done by using high-pass and low-pass filter based on the velocity measurement. 2.2.1. Control Hardware Figure 6 represents the way the control is physically implemented. Each joint is driven by local highly integrated joint controllers mounted on the exoskeleton itself. These communicate trough a joint dispatcher with the global master controller ECO. motor applications. Mainly, it acquires torque sensor and encoder signals. A “smart” torque sensor signal conditioning board, developed in phase A, is embedded in each torque sensor. A PWM current driver is also implemented by comparing the ECO current set point and the current flowing through the motor windings. A custom internal bus is used to connect each joint controller to the joint dispatcher to overcome throughput limitations (compared to a CAN bus as used in phase A) when addressing a multi-joints architecture and to address compactness/very restricted space availability issues. 2.3 Slave Simulator The Slave Simulator provides a virtual slave robot and a set of virtual worlds. It simulates the slave robot, its controllers, and the interactions with the virtual worlds. The slave robot is controlled by direct kinematics, with its 7 DOF corresponding to the seven joints of SAM. When the slave robot is immersed in a virtual world, the Slave Simulator computes the collisions between the robot and itself, and between the robot and its environment. When a collision occurs, the corresponding dynamics and kinematics response is computed. The physics engine (collision detection, and dynamics / kinematics) is built on top of ODE. Figure 6: Exoskeleton controller hardware architecture ECO implements the control strategies presented before for all the joints and the data links with the Slave Simulator (TCP/IP) and the joint dispatcher. It takes also care of the monitoring and triggering of the whole system. To exhibit demanding real-time performance, ECO is implemented on a PC running the real-time QNX™ Operating System. A Sampling frequency of 500 Hz is foreseen. The joint dispatcher has two purposes. Firstly, it receives the motor commands for all the joints from the ECO Controller and dispatches the individual set points to each joint. Secondly, it collects sensor readings from individual joint controllers and transmits the whole set of readings to the ECO Controller PC. Finally each joint controller is implemented on a DSP/Microcontroller TMS320F2812 optimized for The virtual worlds (including the slave robot) are described by Python scripts located on the Slave Simulator. When a simulation starts with a given virtual world, the corresponding code is executed by the Slave Simulator. This allows the user to have a set of different environments and slaves, and to modify them without restarting the whole simulator. In the scope of this project, different worlds will be implemented to demonstrate and measure the capabilities of the system, including the ability of the system to render various stiffnesses, forces and frictions. The simulation of the virtual worlds runs at the same frequency as the remaining of the haptic loop (i.e. 500 Hz). The Slave Simulator runs on a Core 2 Duo 3.0 GHz with a Debian GNU/Linux. The operating system is not real-time, but the simulation is precisely synchronized by ECO to keep up with the 500Hz frequency of the haptic loop. Because of the Simulation Link protocol, ECO and the Slave Simulator are able to resorb a real-time failure without any significant degradation of the haptic feeling (i.e. the haptic loop does not become instable). 2.4 3D Visualisation Client 3. Preliminary Tests The 3D Visualization Client is a remote application which is not part of the haptic chain, but connects to the Slave Simulator to allow visualizing in real-time the state of the virtual world. Several clients can connect to the simulator allowing the simulation to be visualized by several users at the same time, including an optional HMD (Head Mounted Display). 3.1 Haptic rendering The visual description of the slave robot and its environment is automatically sent by the Slave Simulator to the 3D Visualization Client when required. The idea is to avoid any hard-coded or worlddependent visual information to be stored on the 3D Visualization Client. The 3D rendering is done in OpenGL 2.1 and features the shadows (using shadow mapping) of the slave robot to improve depth perception. Additionally an audio feedback is provided when a contact occurs to enhance the haptic sensation. Figure 7 shows the GUI provided by the 3D Visualization Client for supporting the calibration of SAM joints, browsing a list of available virtual environments, starting and stopping a simulation and/or the simulator. The complete 7-DOF SAM is currently under development. Preliminary experiments have already been conducted with the four first joints, from the shoulder to the elbow with a simple two channels control strategy (Figure 8). In order to conduct these tests, the master has been used in combination with a simplified VR (without slave robot simulation) implemented through the exoskeleton controller instead of using the VR platform. The embedded controller being under development, control is currently implemented using a Dspace acquisition board coupled with MATLAB SIMULINK. Results obtained in terms of haptic rendering are very promising. Firstly they showed the possibility for SAM to produce enough force to the operator in order to create a sensation of rigid contact. Secondly, several geometrical forms could be presented to the operator. Figure 9 represents the position of the operator when he was asking to follow a sphere and a cube in space (projection in XZ plane). We can see a good matching between the real position and the virtual object. The variations observed in the cube case are due to difficulties to follow the surfaces near angular discontinuities. Figure 7: Early prototype of the 3D Visualization Client (showing a pull-down menu to browse worlds) Figure 8: Actual prototyping status of SAM (4 joints implemented) low speeds to the Motion Based method at higher speeds in order to achieve a high level of friction compensation at high speeds while maintaining limited friction forces around null speed. Figure 9: Haptic rendering experiments, geometrical form representation 3.2 Friction compensation tests In parallel to the global haptic rendering tests, experiments were also conducted to reduce friction in the third joint of the exoskeleton. As evocated before, several algorithms were implemented. Figure 10 compares the torque to move the joint for an alternative motion (sinusoidal speed pattern), without contact, between the three methods and without compensation. Although the Motion Based method reduces the friction during motion, the friction of the joint is again present around null speed due to the discontinuity in the Coulomb component of the friction model for such speeds. With Force Feedback, stability considerations limit the feedback gains and thus the level of compensation that can be achieved. However, it should be noticed that this level of compensation is independent of the speed of the joint and thus better results are achieved at low speeds than with the Motion Based method. The Hybrid method combines the advantages of the two previous methods while limiting at the same time their drawbacks. The Hybrid method switches from the Force Feedback method used for Figure 10 : Comparison of several friction compensation methods 4. Conclusion This paper has introduced and described the major components of the complete 7-DOF haptic control chain under development in the frame of Phase B of the EXOSTATION project. Preliminary tests have been conducted on both hardware and software and have led to promising results. It has been shown that the four first joints of the exoskeleton are already able to provide high quality 3D haptic rendering when interacting with a simplified Virtual Reality. Preliminary tests conducted on the slave simulator have also shown the ability to command a virtual multi-DOF robotic arm, compute dynamics associated with collisions from various virtual worlds defined in Python scripts. on Advanced Space Technologies for Robotics and Automation, Nov. 2006. Future work will include: further development of the exoskeleton (3 remaining joints for the wrist), implementation of the internal bus and joint dispatcher, further development of the Slave Simulator and the 3D Visualization Client. Once each subsystem will have achieved a sufficient level of development, integration in the final haptic control chain will be performed and experiments will be conducted in order to assess the global system performances. [5] P. Letier, M. Avraam, M. Horodinca, A. Schiele and A. Preumont, Survey of actuation technologies for bodygrounded exoskeletons, Proceeding of the Eurohaptics 2006 Conference, Paris, France, July 2006. 5. Acknowledgments This study is funded by ESA in the framework of a Technology Research Program (contract No. 18408/04/NL/CP) entitled: Control Stations for new Space A & R Applications. 6. References [1] A. Schiele and G. Visentin, The ESA human arm exoskeleton for space robotics telepresence. Proceeding of the 7th International Symposium on Artificial Intelligence, Robotics and Automation in Space, 2003. [2] A. Schiele and G. Visentin. Exoskeleton for the human arm, in particular for space applications, us200323844, dec. 2003, european patent application ep 1364755a1, nov. 2003. [3] A. Schiele and F.C.T. van der Helm. Kinematic design to improve ergonomics in human machine interaction. IEEE Transactions on Neural Systems and Rehabilitation Engineering, Volume 14, Issue 4, Dec. 2006. [4] P. Letier, M. Avraam, J.P. Verschueren, T. Fautre, J.M. Wisley and A. Schiele, EXOSTATION Phase A: a 1-DOF haptic demonstrator, Proceeding of the 9th ESA Workshop [6] P.Letier, A. Schiele, M. Avraam, M. Horodinca and A. Preumont, Bowden cable actuator for torque feedback in haptic applications, Proceeding of the Eurohaptics 2006 Conference, Paris, France, July 2006. [7] ODE (Open Dynamics Engine), http://www.ode.org [8] T.H. Massie and J.K. Salisbury, The Phantom Haptic Interface: A Device for Probing Virtual Objects, Proceedings of the {ASME} Winter Annual Meeting Symposium on Haptic Interfaces for Virtual Environment and Teleoperator Systems, Nov. 1994, Chicago, IL. [9] C. Carigan and M. Liszka, Design of an Arm Exoskeleton with Scapula Motion for Shoulder Rehabilitation, Proceedings of the International Coference on Advanced Robotics, July 2005, Seattle, USA. [10] J. Rosen, J.Perry, N. Manning, S. Burns and B. Hannaford, The Human Arm Kinematics and Dynamics during daily activities - Toward a 7 DOF Upper Limb Powered Exoskeleton, Proceedings of the 12th International Conference on Advanced Robotics, July 2005, Seattle, USA. [11] N.L. Bernstein, D.A. Lawrence and L.Y. Pao, Friction modeling and compensation for haptic interfaces, Eurohaptics Conference, 2005 and Symposium on Haptic Interfaces for Virtual Environment and Teleoperator Systems, p. 290-295, March 2005.