IROS2019国际学术会议论文集 1979

Twin Kinematics Approach for Robotic Assisted Tele Echography Lu s Santos Rui Cortes oandJo o Quintas Abstract This paper discusses a new teleoperation approach for robotic assisted tele echography A teleoperation architec ture in the joint space is presented taking advantage of kine matic similarity between master and slave manipulators Haptic force feedback is provided based on the slave control command torque without using force torque sensing data The slave manipulator is controlled in the joint space using computed torque techniques and featuring Kalman Active Observers AOBs Experiments in a typical telemedice scenario assess and validate the teleoperation architecture in a clinical context with a radiologist performing a robotic assisted abdominal ultrasound examination on a healthy volunteer I INTRODUCTION In the nineties researchers started to develop robotic assisted ultrasound systems being the work of Pierrot et al 1 one of the fi rst approaches to introduce a robot in an echographic procedure According to Priester 2 a robotic assisted ultrasound system can be defi ned as the integration of an ultrasound imaging system in a robotic platform taking advantage of robots characteristics to improve or create new medical procedures based on ultrasound imaging Priester 2 performed a review of robotic assisted ultrasound systems in different medical applications dividing them in three large groups Extracorporeal Diagnostic Systems for ultrasound diag nostic procedures Needle Guidance Systems Intraoperative Surgical Systems ultrasound real time imaging of hidden structures in surgical environments Regarding robotic systems for ultrasound diagnostic proce dures the adopted solutions follow one of two approaches the fi rst one relying on small and portable robots which require an assistant to hold and place the robot in the region under examination with the physician able to control the probe orientation and perform small probe translations near the organ of interest the second approach features independent robotic archi tectures to fully control the probe pose The TERESA robot 3 the OTELO system 4 the PROSIT 1 robot 5 and more recently the commercially available MELODY system 6 are examples of architectures that This work was supported by the project CENTRO 01 0247 FEDER 017958 funded by CENTRO 2020 Portugal 2020 and the European Union by FEDER Lu s Santos is with the Institute of Systems and Robotics University of Coimbra Portugal email luiscvsantos isr uc pt Rui Cortes o is with the Electrical and Computer Engineering Department Institute of Systems and Robotics University of Coimbra Portugal email cortesao isr uc pt Jo o Quintas is with the Instituto Pedro Nunes Coimbra Portugal email jquintas ipn pt fall into the fi rst paradigm while 7 9 are examples of approaches featuring independent robotic architectures This paper discribes the teleoperation control architecture adopted in the ROSE RObot Sensing for tele Ecography project Each ROSE system includes two torque controlled lightweight robots with similar kinematic structure one at the physician side and other at the patient side a set of ultrasound probes with classical functionalities patient database systems located in the cloud and safe internet based communication facilities Merging all these components in coherent way the ROSE system allows the interaction of physician and patient through a compliant telecontrol architecture enabling the physician with haptic perception of the patient body while mitigating shortcomings of traveling from both patient and physician sides In this paper a teleoperation control architecture using the joint space formalism is presented taking advantage of kinematic similarity between master and slave manipulators to simplify the control architecture design Haptic force feedback is provided in the master station without using force torque sensing data in the slave manipulator Furthermore the slave manipulator control architecture is also presented being the teleoperation strategy and the slave control architecture validated by a physician in a typical telemedicine setup The remaining of this paper is organized as follows Section II presents the ROSE system The teleoperation architecture is presented in Section III while Section IV details the slave manipulator control strategies In Section V experimental results are presented Finally Section VI concludes the paper II ROBOTIC ASSISTEDTELE ECHOGRAPHYSYSTEM A robotic assisted tele echography system can be divided in three major parts Master site where the physician is located controlling a remote manipulator through a haptic device while receiving ultrasound and video conference images Slave or remote site where the patient robot and ultrasound device are placed Communication link used to exchange control sound ultrasound and video conference data between sites In Fig 1 the ROSE system is depicted Our approach is to use two commercially available manipulators with similar kinematic structure taking advantage of their similarity in the development of the teleoperation control architecture The manipulators chosen are the Virtuose 6D and Virtuose 6D Desktop two haptic devices from Haption 2019 IEEE RSJ International Conference on Intelligent Robots and Systems IROS Macau China November 4 8 2019 978 1 7281 4003 2 19 31 00 2019 IEEE1339 A Slave Station The remote station has an anthropomorphic arm the Virtuose 6D robot with an ultrasound probe attached to its end effector The Virtuose 6D is a torque controlled 6 DoF lightweight anthropomorphic designed by its manufacturer as a haptic device with high force feedback and large workspace The fi rst three joints are cable driven enabling low friction no backlash and low opposition to external forces while the last three joints motor torques are transmitted by harmonic drives This robot is adequate for robotic assisted tele echography since it has most of its mass located in the base having a small moving inertia Furthermore the forces that the manipulator can apply are limited due to its structure acting as an extra safety measure The robot controller is on an external PC with a Linux Xenomai operating system being connected to the robot through an Ethernet interface The ultrasound probe is a Interson SiMPLiTM GP C01 Medical Fig 1c shows the medical robot with the ultrasound probe attached to its end effector B Master Station The master station features a Virtuose 6D Desktop as a haptic device a small scale version of the slave manipulator The robot controller is also on an external PC with a Linux Xenomai operating system being connected to the robot through an Ethernet interface Fig 1d shows the master station The haptic device together with medical imaging interface are presented for task monitoring assessment and control The haptic device sends joint position commands receiving 6 DoF haptic force feedback C Human Robot Interface Human Robot interfaces constitute another key element of the system since they are the main interface users interact with In our system we considered a common technological framework for the interfaces at both ends Besides typical human computer interfaces i e keyboard and mouse and the haptic device the graphical user interfaces GUI re quired careful analysis and design to ensure an equivalent level of similarity with the interface available in traditional ultrasound devices Therefore our GUI includes features to support authentication of users managing worklists display ultrasound images support video conferencing ultrasound probe command and image annotations The GUI displayed at the master station shows the entire set of features while in the slave station a simplifi ed version is used showing only video conference and ultrasound images if allowed by the physician Communication of images and health information between both stations is performed through a combination of HL7 and DICOM protocols This approach also ensures interoperability with Radiology Information Systems RIS and Picture Archiving and Communication System PACS systems Fig 1b shows the medical interface at the master station SLAVE STATION INTERNET COMMUNICATION CHANNEL MASTER STATION Parameters a b c d Fig 1 ROSE Robotic Assisted Tele Echography Setup a Tele Echography concept b Medical interface c Slave manipulator d Master station d qrqd Gs s Gm s h Control d a Gs s Gm s hQdQr dMr MrKp Kds Kp Kds d b Fig 2 Teleoperation architecture a Teleoperation scheme b Block diagram rearranged to better highlight the local feedback in each system dis a torque scaling factor III TELEOPERATIONARCHITECTURECONCEPTUAL ANALYSIS FOR1 DOF Since both master and a slave devices have a similar geometric structure the teleoperation architecture Fig 2 follows a position position approach in the joint space The slave reference is given by master joint positions while master input is the torque computed by the slave joint position controller In the sequel the teleoperation architecture is analyzed for a 1 DoF system LetGm s andGs s be master and slave station transfer functions respectively with da torque scaling factor The slave station transfer function including manipulator PD controller and environment from a joint position inputqd to the manipulator joint motionqris given by in Laplace domain Gs s Qr s Qd s Kds Kp s2 Kds h Kp Kjs Mr i 1 whereQd s andQr s are respectively the Laplace do main representation ofqd t andqr t Kjsrepresents the 1340 QdQr Tr reaction force physics Kjs Mass Normalization Computed Mr 1 Mrs2 Robot Dynamics Joint Controller Contact Dynamics Kp Kds a QdQr Tr reaction force physics Mass Normalization Computed Mr 1 Mrs2 Robot Dynamics Joint Controller KjsK 1 js Contact Dynamics Kp Kds b Fig 3 Slave station block diagram a Slave station with the reaction force represented in the feedback loop b Block diagram rearranged to better highlight the feedback loops environment stiffness seen in the joint space andMrthe slave mass KpandKdare respectively proportional and derivative PD control gains Fig 3 shows a block diagram of the slave robot in contact Tr s 1is the torque applied by the manipulator on the environment Qr s Tr s Kjs 2 The master station includes the haptic device and the human arm2and can be modeled as a mass spring damper system 10 Gm s 1 Mms2 Dms Km 3 whereMm DmandKmare mass damping and stiffness of the master station respectively Kmis mainly due to the human arm stiffness being assumed that the operator presents a compliant behavior in contact low Km A Telepresence Lawrence 11 states that transparency i e telepresence is achieved when the transfer function from a force input applied by the human operator Th s to the master device position output Qd s matches the compliance of the slave system in contact Qd s Th s 1 Mms2 Dms Km dKjs 4 where dis a scaling factor to enhance environment per ception d 0 The transfer function of our teleoperation architecture is see Fig 2a Qd s Th s Gm s 1 Gm s Kp Kds dMr s2 Kjs Mr s2 Kds Kp Kjs Mr 1 Mms2 Dms Km 1 Kp Kds dMr Mms2 Dms Km s2 Kds Kp Kjs Mr 5 Fig 2b shows a rearranged block diagram of Fig 2a 1This value is not measured in our system 2There is no force controller at master station This approach can only be applied in light weight frictionless master stations 1 Low Frequency Motion The analytical analysis of 5 is diffi cult since it depends on the human arm parameters which are diffi cult to quantify and may vary during task execution 12 However for low frequency motions s 0 typically the motions present in ultrasound examinations when replacing 1 in 5 the transfer function of our teleoperation architecture becomes Qd s Th s 1 Km 1 dMrKpKjs Km Kjs KpMr 1 Km 1 dKh 6 with Kh MrKpKjs Kjs KpMr 7 the perception that the human operator has of the remote en vironment Fig 4 showsKhevolution when the environment on the slave side Ksj changes from free space to a very stiff surface In free space 1 becomes decoupled from the robot dynamics Gs s Qr s Qd s Kds Kp s2 Kds Kp 8 and 6 becomes Qd s Th s 1 Km 9 meaning that there is telepresence since only the master station is felt see 3 In contact for soft environments lowKsj whenKsjincreases Khalso increases showing a quasi linear behavior Equation 6 is similar to 4 meaning the operator is able to perceive the environment on the slave side For stiff environments whenKjs KpMr 6 becomes Qd s Th s 1 Km dKpMr 10 showing no telepresence since the operator only feels the product of Kpby Mr 2 Discussion Equation 7 shows that despite the human operator not having an exact perception of the environ ment the teleoperation architecture does provide a form of telepresence allowing the operator to distinguish between free space soft and stiff contact surfaces For a task such as an ultrasound examination this approach is enough to provide realistic sensations to the physician enabling him to distinguish between soft tissue and surfaces like bones Stability analysis of the teleoperation architecture is out of the scope of this paper IV SLAVEMANIPULATORCONTROLARCHITECTURE This section discusses the slave manipulator control archi tecture The control architecture is based on computed control techniques with each joint controlled at position level through an Active Observer AOB 12 1341 04 0008 000 0 500 1 000 Kjs Kh force a Fig 4 Khevolution when the environmentKsjon the slave side changes from free space to a stiff surface with Mr 1 Kg and Kp 1000 A Manipulator Dynamic Model The dynamic model of a manipulator withngeneralized coordinatesq Rnandngeneralized torques acting on it Rnis given by B q q c q q g q 11 whereB q Rn nis the inertia matrix c q q Rnis the vector of Coriolis and centripetal terms andg q Rn is the gravity term 1 Nonlinear Feedback Linearization The manipulator formulated in 11 is a nonlinear system If the dynamic model is known 13 a linear and decoupled system can be achieved through nonlinear feedback linearization Let the total torques acting on the manipulator be given by c f r 12 where c Rn f Rn r Rnare actuator friction and interaction torques respectively Neglecting joint frictions f 0 and assuming the manipulator is in free space r 0 the commanded torque c3 can be defi ned as4 c c q q g q 0c 13 From 13 and 11 the system reduces to B q q 0c 14 where 0c Rnis the torque vector computed by the control law If 0c is defi ned as 0c B q 15 a linear and decoupled system arises in free space represent ing the dynamics of a unitary mass for each joint variable q 16 where Rnis the new control variable a resolved acceleration in the joint space In contact 14 becomes B q q 0c r 17 due to interaction torques not being compensated since our system does not provide force torque sensing 3In this setup the gravity term g q in 13 is compensated by the manipulator physical structure 4It is assumed throughout the paper that robot kinematic and dynamic parameters are estimated with negligible errors 2 Control Issues Controlling the last two joint is par ticularly problematic Due to its low inertia friction effects are dominant the last two elements inB q main diagonal are at least two orders of magnitude smaller than the other elements Since frictions are not taken into account in the dynamic model f 0 to overcome friction effects high control gains must be employed to control the last two joints which leads the system close to instability To cope with this issue the last two joints are controlled without taking into account their inertia in the control law 0c B q S1 S2 18 with S1 I 4 4 0 002 2 S2 0 4 4 0 0I2 2 19 whereIand0are identity and zero matrices respectively In the sequel the control analysis is done for each joint dimension individually B System Plant Equation 16 describes the dynamics of a unitary mass for each joint dimension in free space Assuming a small time system delayTd the plant of a position controlled robot can be approximated by Gf s e sTd s2 20 The equivalent temporal representation is given by y t u t Td 21 whereu t is an acceleration plant input andy t is a joint position plant output Defi ningx r1 t y t andx r2 t y t 21 can be written in a state space representation as xr t Axr t Bu t Td y t Cxr t 22 with A 0 1 00 B 0 1 and C 1 0 01 whereA R2 2 B R2andC R2 2are respectively state input and output matrices Discretizing 22 with sampling timeh the equivalent discrete time system becomes5 xr k rxr k 1 ruk 1 yk Crxr k 23 C Active Observer Design An AOB is a state space observer that reformulates the Kalman fi lter framework to accomplish model reference adap tive control Based on an explicit pole placement approach to achieve a desired closed loop response the AOB uses the Kalman stochastic design to refl ect the uncertainties of system model and output measurements in the state estimation An extra state active state pk is used to estimate and compensate disturbances referred to the system input Details on AOB design stability and robustness can be found in 12 5Details on discrete matrices can be seen in 12 1342 1 s 1 s AOB qd rk xr k pk T Lr1 System Plant Gf s qr qr e sTd L1 L2 qd Fig 5 Joint position control with an AOB in the loop Each joint position dimension is controlled by the AOB 1 Stochastic Design LetRk E k Tk andQk E n k T k o be respectively measurement and system noise matrix Qkhas the form Qk Q xr k 0 0Qpk 24 kand kare the stochastic inputs due to model and measure uncertainties respectively Let joint position measurements be written as qik qik qik 25 whereqikis theithjoint position measurement corrupted by noise qik Similarly the joint velocity measurement can be represented by qik qik qik 1 h qik 26 with qik qik qik 1 h 27 showing that measurements noise is correlated Since Rk E k Tk E 2 qik qik qik qik qik 2 qik 28 straightforward analysis from equations 26 27 and 28 gives Rk E n 2 qik o 1 1 h 1 h 2 h2 29 The Kalman gainKkdepends only on the relation between measurementRkand system noiseQk and not on their absolute value itself In this way a normalized value ofRk has been use