D Computer Technology and Application 4 (2013) 569-574 DAVID PUBLISHING Remote-Controlled Ambidextrous Robot Hand Concept Emre Akyürek, Alexandre Dilly, Fabrice Jourdan, Zhu Liu, Souraneel Chattoraj, Itziar Berruezo Juandeaburre, Michel Heinrich, Leonid Paramonov, Peter Turner, Stelarc and Tatiana Kalganova School of Engineering and Design, Brunel University, Uxbridge, Middlesex UB8 3PH, UK Received: October 08, 2013 / Accepted: October 25, 2013 / Published: November 25, 2013. Abstract: In the development of robotic limbs, the side of members is of importance to define the shape of artificial limbs and the range of movements. It is mainly significant for biomedical applications concerning patients suffering arms or legs injuries. In this paper, the concept of an ambidextrous design for robot hands is introduced. The fingers can curl in one way or another, to imitate either a right hand or a left hand. The advantages and inconveniences of different models have been investigated to optimise the range and the maximum force applied by fingers. Besides, a remote control interface is integrated to the system, allowing both to send commands through internet and to display a video streaming of the ambidextrous hand as feedback. Therefore, a robotic prosthesis could be used for the first time in telerehabilitation. The main application areas targeted are physiotherapy after strokes or management of phantom pains for amputees by learning to control the ambidextrous hand. A client application is also accessible on Facebook social network, making the robotic limb easily reachable for the patients. Additionally the ambidextrous hand can be used for robotics research as well as artistic performances. Key words: Robot hand, pneumatic muscles, remote platform, video streaming, telerehabilitation, hand recognition. 1. Introductionο The aim of the project is to design an anthropomorphic ambidextrous robotic hand, actuated by PAMs (pneumatic artificial muscles) and controllable via a remote platform over internet using hand gesture recognition through a webcam (Fig. 1). Although a number of dexterous robot hands have been developed over last decade, none of them considered the possibility of an ambidextrous design, with fingers performing both in “left hand” and “right hand” behaviour, bending in one way or another to have both a left and a right hand in a single architecture. Consequently, the design allows significant reduction of the required resources for its application areas, such as rehabilitation and physiotherapy after strokes. Phantom limb pain after amputation could also be eased interacting with the ambidextrous hand, as the movements may be Corresponding author: Emre Akyürek, Ph.D. student, research field: electronic and computer engineering. E-mail: emre.akyurek@brunel.ac.uk. interpreted as those of the missing limb. The ambidextrous hand is also accessible from Facebook social network to facilitate its access for the patients. The objectives of the project included both the design of ambidextrous fingers and the creation of a remote control interface. Both of them are introduced in this paper. The content of which is organized as follow: Section 2 discusses different finger prototypes tested and presents experimental data collected for each phalange. Section 3 introduces the control theory used to control the ambidextrous hand movements. Section 4 describes the implementation of the remote control platform and the video streaming of the robot hand. Section 5 gives a conclusion. Finally, Section 6 presents applications under development. Fig. 1 System architecture. Remote-Controlled Ambidextrous Robot Hand Concept 570 2. Prototypes of Ambidextrous Fingers change for the ambidextrous design, all the tests have been focused on flexion/extension over a first phase. 2.1 Outline of the Research Human fingers can make two kinds of antagonist movements: flexion and extension, or abduction and adduction. Flexion and extension control the angular displacement of the three phalanges of a finger, which means as many DOF (degrees of freedom). Abduction and adduction imply lateral rotations of a whole finger and constitute another DOF, which makes a total of four distinct DOFs per finger [1]. These four DOFs must be reproduced to obtain a robotic model as close as possible to a human hand. As abduction and adduction are not essential for a number of applications, many dexterous hands have been developed without taking them into account [2], [3]. Limiting the number of DOFs allows both easing the control of the structure and reducing the number of needed actuators. According to the first investigations, the abduction/adduction of the whole hand can be simulated with two PAMs spreading the five fingers or bringing them closer at the same time. In addition to these movements, using four PAMs to control the flexion/extension of each finger would make a total of 22 PAMs for a robot hand of 16 DOFs and for which the range is twice bigger than the one of a human hand [4]. As showed in Table 1, it is current for dexterous hands to use even more PAMs to imitate human gestures. Given that the main difference with the ambidextrous hand is the range of fingers, the ability to 2.2 Results Obtained with Different Models The tests have been made with standard PAMs, used to control classic robot hand designs. As human distal and middle phalanges do not move independently, a first pair of antagonist PAMs is used to drive them together while a second pair is connected to the proximal phalanges. The tests are done with 150 mm PAMs for the distal phalanges and 122 mm PAMs for the proximal phalange, contracting with compressed air at a maximum pressure of 3.5 bars. Several designs have been investigated to maximise the phalanges range and reach an ambidextrous behaviour. Design A, shown in Fig. 2, consists in coupling two pulleys of different diameters at joint levels. Therefore, for the rotation of the first pulley around a distance π0 with a force πΉ0 , the second pulley rotates around a distance π1 applying a force πΉ1 on the following tendon, theoretically increasing the range by the ratio π1 /π0. Design B, also illustrated in Fig. 2, consists in wrapping the tendons around the pulley of the second joint, to prevent any looseness of tendons during the rotation of the phalanges. The performances of designs A and B are gathered in Table 2, which indicates the maximum angles reached by the metacarpo-phalangeal joint (MCP), the proximal interphalangeal joint (PIP), and the distal interphalangeal joints (DIP), as well as the maximum forces applicable by their respective phalanges, both cover such a large angle is the main point investigated in this paper. As the abduction/adduction does not Table 1 Comparison between PAMs and DOFs of several robot hands. Robotic hands J. Sancho-Bru et al, 2001[5] D. Wilkinson et al, 2003 [6] Y. Honda et al, 2010[7] Festo, 2012 [8] Shadow Robot, 2013 [9] Number of PAMs 25 31 25 26 40 Number of DOFs 20 N/A ~ 20 17 N/A ~ 20 24 (a) (b) Fig. 2 Tendon routings of ambidextrous fingers prototypes, where (a) is Design A and (b) is Design B. Remote-Controlled Ambidextrous Robot Hand Concept 571 Table 2 Experimental data obtained with designs A and B. Models Design A Design B Joints MCP PIP DIP MCP PIP DIP Max. angles reached (°) Left Right 85 80 80 85 40 43 84 82 70 71 52 55 Max. forces applied (N) Left Right 52 47 9 10 3 3 51 49 12 13 0 0 for their left and right positions (the maximum forces have been collected applying pressure on the phalanges until their maximum angles decrease of 10°). As it can be observed, both of these designs reach correct left and right positions. However, none of them apply enough force on the distal phalange to allow a proper control of the finger or to grab a light object. For the Design A, it is due to the loss of force caused by the coupled pulleys. Indeed, because of the range increase, the force provided by the PIP joint is πΉ1 = πΉ0 ∗ π0 /π1. Other tests have been done also coupling the pulleys at the DIP joint, but the force was so weak it even decreased the maximum angles reached. Design B has the same inconvenience because of the friction force of the wrapped tendons. To overcome these defects, the Design C, illustrated in Fig. 3, includes intermediate pulleys which, as it is for Design B, avoid the tendons to get loose, but this time preserving most of the tendon force. The movements of middle and distal phalanges are synchronised because of torsion springs put under the PIP and DIP joints. The same experiments are repeated with Design C, using different size of pulleys at the joints levels. The data collection is provided in Table 3. The better pulley configuration to optimise the angle/force ratio is 15.34 mm for the MCP joint and 7.71/2.95 mm for the PIP/DIP joints. It is seen that the maximum angles can be increased using smaller pulleys, but then the loss force becomes too significant. As a conclusion, even though the experiments are done with the standard PAMs, the ambidextrous design reaches ranges quite close to other models. The ranges reached for right hand behaviour are indicated in (a) (b) Fig. 3 Tendon routing of Design C, (a) is a left position and (b) is a right position. Table 3 Experimental data obtained with different pulleys configuration of Design C. Joints Pulleys’ diameter (mm) MCP MCP MCP PIP/DIP PIP/DIP PIP/DIP PIP/DIP 23.81 15.34 7.71 23.81/15.34 15.34/15.34 7.71/2.95 2.95/2.95 Max. angles reached (°) Left Right 58 54 85 80 90 90 54/10 60/19 73/29 77/36 75/51 80/55 88/68 90/72 Max. forces applied (N) Left Right 91 86 53 49 38 31 59/4 62/11 30/18 35/22 21/10 23/13 12/3 15/4 Table 4 Comparison of joints maximum angles (°). Joints Ambi-hand MCP 80 PIP 80 DIP 55 Shadow Hand [10] 90 90 90 ACT Hand[11] 90 110 70 Table 4 for each joint, including comparisons with the Shadow Hand and the ACT Hand, both known as belonging to the most advanced robot hands in the world. It is noted that the ambidextrous hand can reach better performance using more suitable PAMs. However, the range was large enough to proceed with first tests using control theory. 3. Control of the Hand The use of PAMs instead of motors allows more flexibility to the overall behaviour of the system. Contracting or relaxing according to the amount of compressed air getting in or out, their lengths variation is used to interact with the endoskeleton of the hand, such as real human muscles. However PAMs are rarely used to control robotic structures because of the 572 Remote-Controlled Ambidextrous Robot Hand Concept non-linearity existing between the inflation of the PAM and the force provided [12], [13]. The airflow is controlled from valves opening or closing for a specific time depending on the input voltage they receive. Voltages’ switching is commanded by a PIC microcontroller that converts the control functions into analog signals. The delays between these signals define the different contraction rates of PAMs. Specific gestures of the ambidextrous hand are calculated according to the parallel form of a PID (proportional integrative derivative) control, for which the equation is: π‘ π π’(π‘) = πΎπ π(π‘) + πΎπ ∫ π(π)ππ + πΎπ π(π‘) ππ‘ 0 The error π(π‘) is determined according to the difference between the target value and the current data feedback. The output π’(π‘) is sent to valves, making the PAMs react to reduce the error π(π‘) until the sensors return the target values. Appropriate constants πΎπ , πΎπ and πΎπ allow the system to reach the target in a faster way, minimizing the overshoots and the number of oscillations. The PID control allows dealing with several parameters of the system such as the joints’ angles, the force at the fingertips or the PAMs’ pressure. As the commands must be received from internet, the control functions are recorded on the server that runs on the host computer. The hand movements can be watched online because of the video feedback integrated in the remote control platform. formats. The more suitable video format to implement is MJPEG, because it is compatible with all web browsers and can be streamed, with an appropriate configuration, at a maximum of 10 Mbps of bandwidth, which is available in most of internet connections. The stream is sent using a HTTP server that deals with HTTP requests, sending back the appropriate HTTP headers and XML messages. All these communication processes are integrated in the server, the architecture of which is shown in Fig. 4. It allows the server to communicate with its two different clients. The first one is a Web application, easily accessible from the ambidextrous robot hand’s website or from its Facebook application. The video control options are stored in XML files, got and parsed with Java script. Fig. 4 Server architecture. 4. Remote Control Platform The video streaming and the communication between the robotic limb and its operator are made possible by the development of a client-server interface on a Linux platform. The video is collected from an embedded webcam using the USB Video Class (UVC) protocol and the Linux UVC driver [14], which have an interface for the Video For Linux (V4L) library [15]. The V4L library communicates with all input devices and allows the operating system to decode some video Fig. 5 Finger design driven by 2 passive and 3 active tendons. Remote-Controlled Ambidextrous Robot Hand Concept The second one is software developed on the cross-platform library Qt4 [16], which would allow communicating with the robot hand using other hardware devices, such as EMG signals. It includes a webcam control interface, webcam’s images renderer and the robot hand control interface. Both of these applications are based on TCP/IP protocol. Specific commands can be selected from their respective graphical user interfaces and sent to the server as XML Http Request objects either from Java script or from Qt software. 5. Conclusions The testing of first prototypes has proved that an ambidextrous behaviour is possible for a robot hand. However, because of the range’s limitations and the weak grasp of current designs, it has been decided to use longer PAMs, able to inflate at a pressure up to eight bars, to have a more accurate control of the hand. As for the remote control platform, its functioning has been proved by controlling ambidextrous prototypes from a distance. 6. Future Plan Independent prototypes are still tested and modified recognition. Using previous research in this area [17], the head region is currently recognized with the help of the cascade classifier algorithm from OpenCV [18]. A histogram of colours to repartition is created from this region and used to find skin of the same colour to identify the hand. This application has been integrated in a browser plugin downloadable from the ambidextrous hand website. Contrary to Flash and Java applications, its use implies native code for every operating system, which avoids respectively a low power consumption for image processing and interpretation tasks through Java Virtual Machine. Developed with the NPAPI routines [19], the plugin uses a dynamic library which allows it to deploy very fast. However, the fingers tracking part has not finished being developed yet. Once done, the patient’s movements would then be imitated on the robotic model as a hand of the same side or on the opposite one, to allow mirroring movements. Acknowledgments This research was partially assisted by Shadow Robot Company Ltd (London, UK), which provided a number of pneumatic and electronic devices, as well as hardware to control the first prototypes. according to their kinematic performances. To reduce References the number of PAMs of the whole hands, next testing [1] will include finger designs driven by only three tendons. A global architecture is shown in Fig. 5. A tendon driving the proximal phalange is removed, but [2] the design would allow replacing its effect by pulling the two tendons connected to the middle phalange together. Although the numerical simulations indicate this kind of models is properly controllable, no similar [3] designs have been implemented yet. Once the best behaviour is reached with a specific model, the chosen design will be replicated to the fingers of the whole hand. Over a second phase, next stage consists in increasing the interaction between the operator and the robot hand by developing software for hand gesture 573 [4] [5] A.M. Zaid, M.A. Yaqub, A.M. Rizal, M.S. Wahab, UTHM hand: Mechanics behind the dexterous anthropomorphic hand, Engineering and Technology 50 (2011) 150-154. P.J. Kyberd, C. Light, P.H. Chappell, J.M. Nightingale, D. Whatley, M. Evans, The design of anthropomorphic prosthetic hands: A study of the Southampton Hand, Robotica 19 (2001) 593-600. D. Choi, D. W. Lee, W. Shon, H. G. Lee, Design of 5 D.O.F Robot Hand with an Artificial Skin for an Android Robot, The Future of Humanoid Robots – Research and Applications (2012), 81-96. M. Johnsson, C. Balkenius, Sense of touch in robots with self-organizing maps, IEEE Transactions on Robotics 27 (2011) 498-507. J.L. Sancho-Bru, A. Pérez-González, M. Vergara, D.J. Giurintano, A 3D biomechanical model of the hand for power grip, Journal of Biomechanical Engineering 125 (1) (2003) 78-83. 574 [6] Remote-Controlled Ambidextrous Robot Hand Concept D.D. Wilkinson, M. Vandeweghe, Y. Matsuoka, An extensor mechanism for an anatomical robotic hand, in: Proc. Of IEEE International Conference on Robotics and Automation, 2003, pp. 238-243. [7] Y. Honda, F. Miyazaki, A. Nishikawa, Control of pneumatic five-fingered robot hand using antagonistic muscle ratio and antagonistic muscle activity, in: Proc. of IEEE RAS & EMBS International Conference on Biomedical Robotics and Biomechatronics, 2010, pp. 337-342. [8] Festo (2012) Exo-hand. [9] Shadow Robot Company Ltd (2013) Shadow Dexterous Hand E1P1R. [10] Shadow Robot Company Ltd (2013) Shadow Dexterous Hand E1M3R. [11] A.D. Deshpande, Z. Xu, M.J. Vandeweghe, B.H. Brown, J. Ko, L.Y. Chang, D.D. Wilkinson, S.M. Bidic, Y. Matsuoka, Mechanisms of the anatomically correct testbed (ACT) hand, IEEE/ASME Transactions on Mechatronics 18 (2013) 238-250. [12] E. Kelasidi, G. Andrikopoulos, G. Nikolakopoulos, S. Manesis, A survey on pneumatic muscle actuators modeling, Journal of Energy and Power Engineering 6 (2012) 1442-1452. [13] Z. Varga, M. MouΔka, Mechanics of pneumatic artificial muscle, Journal of Applied Science in the Thermodynamics and Fluid Mechanics 3 (2009) 1-6. [14] http://www.ideasonboard.org/uvc/. [15] UVC specification in video class section, http://www.usb.org/developers/devclass_docs/. [16] http://qt.nokia.com/. [17] G. Molenaar, Sonic Gesture, in: Master Artificial Intelligence, University of Amsterdam, 2010. [18] OpenCV Library, http://opencv.willowgarage.com/wiki/. [19] Mozilla Public License, NPRuntime Samples for the Development of Scriptable Plugins with Mozilla NPAPI, http://mxr.mozilla.org/seamonkey/source/modules/plugin /samples/npruntime/.