fulltext_139

Upload: infodotz

Post on 14-Apr-2018

215 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/30/2019 fulltext_139

    1/9

    J. Cent. South Univ. Technol. (2010) 17: 807815

    DOI: 10.1007/s117710100560y

    Dynamic surface control-backstepping based impedance control for

    5-DOF flexible joint robots

    XIONG Gen-liang()1, XIE Zong-wu()1, HUANG Jian-bin()1,

    LIU Hong()1, 2, JIANG Zai-nan()1, SUN Kui()1

    1. State Key Laboratory of Robotics and System, Harbin Institute of Technology, Harbin 150001, China;

    2. Institute of Robotics and Mechatronics, German Aerospace Center (DLR), Wessling 82230, Germany

    Central South University Press and Springer-Verlag Berlin Heidelberg 2010

    Abstract: A new impedance controller based on the dynamic surface control-backstepping technique to actualize the anticipant

    dynamic relationship between the motion of end-effector and the external torques was presented. Comparing with the traditional

    backstepping method that has explosion of terms problem, the new proposed control system is a combination of the dynamicsurface control technique and the backstepping. The dynamic surface control (DSC) technique can resolve the explosion of terms

    problem that is caused by differential coefficient calculation in the model, and the problem can bring a complexity that will cause the

    backstepping method hardly to be applied to the practical application, especially to the multi-joint robot. Finally, the validity of the

    method was proved in the laboratory environment that was set up on the 5-DOF (degree of freedom) flexible joint robot. Tracking

    errors of DSC-backstepping impedance control that were 2.0 and 1.5 mm are better than those of backstepping impedance control

    which were 3.5 and 2.5 mm in directions X, Y in free space, respectively. And the anticipant Cartesian impedance behavior and

    compliant behavior were achieved successfully as depicted theoretically.

    Key words: Cartesian impedance control; dynamic surface control; backstepping; PPSeCo; flexible joint robots

    1 Introduction

    In contrast to traditional industrial robots, people in

    robotics domain transferred their interest into the service

    robot in the past several years, such as medical robots,

    mobile robots and exploration robots. These robots will

    work in the laboratory environments or practical

    environments, even in the space. The compliant behavior

    of the manipulator is predesigned whenever a robot is

    supposed to perform some manipulation tasks such as

    picking and placing operation in practice. In order to

    achieve the compliant behavior by a control method, animpedance control method, a classical issue, which could

    provide a unified framework for achieving compliant

    behavior when robot contacted with an unknown

    environment, was brought up.

    Impedance control was theorized by HOGAN [1]

    and experimentally applied by KAZEROONI et al [2].

    Based on a singular perturbation approach, flexible joint

    robot was controlled by impedance control [3], the

    feedback of the joint torques was therein considered as

    the control input of a fast inner control loop that received

    its set point values from an outer impedance controller.

    HUANG et al [4] also proposed an impedance controller

    for the flexible joint robot based on the singular

    perturbation theory by DSP (digital signal processing)/

    FPGA (field programmable gate array) hardware

    structure. However, the main drawback of the singular

    perturbation method is lacking of theoretical justification

    for proving the stability due to the limitation of

    Tychonovs theorem [5]. Therefore, ALIN and GERD [6]

    proposed Cartesian impedance control techniques for the

    torque control of light-weight flexible joint robots, using

    local stiffness control to enhance the impedance control.

    CHRISTIAN et al [7] developed decoupling based

    Cartesian impedance control of flexible joint robots anda formal analyzed the stability of the proposed controller.

    CHRISTIAN et al [810] investigated the Cartesian

    impedance control of the light-weight flexible joint

    robots of DLR with torque feedback, gravity

    compensation and complete static states feedback, and

    proved asymptotical stability based on passivity theory.

    OZAWA and KOBAYASHI [11] proposed a new

    impedance control concept for elastic joint robots with

    programmable passive impedance devices in the

    transmission. The concept allows the user to use the

    same index for free motion and contacting task, but it

    Foundation item: Project(2006AA04Z228) supported by the National High-Tech Research and Development Program of China; Project(PCSIRT) supported

    by Program for Changjiang Scholars and Innovative Research Team in University

    Received date: 20091219; Accepted date: 20100401

    Corresponding author: XIONG Gen-liang, PhD; Tel: +8645186412042; E-mail: [email protected]

  • 7/30/2019 fulltext_139

    2/9

    J. Cent. South Univ. Technol. (2010) 17: 807815808

    was applied to one-DOF elastic joint robots. GIANNI et

    al [12] presented an impedance controller for elastic joint

    industrial manipulators. Special attention was paid to all

    aspects that qualify an industrial robot, including

    decentralized proportional integral derivative positioncontrol, torsional flexibility, and friction at the joints, etc.

    CHIEN and HUANG [13] proposed a regressor-free

    adaptive impedance controller for an n-link flexible joint

    robot, the function approximation technique (FAT) was

    employed to transform the time-varying uncertainties

    into finite combinations of orthogonal basis functions.

    HUANG et al [14] proposed an adaptive impedance

    controller for flexible joint robot based on the friction

    model. The friction model includes viscous friction,

    payload and motor position based friction. The closed

    loop stability was investigated. LIU et al [15]investigated the Cartesian impedance control and

    nonlinear compensation for a harmonic drive robot based

    on joint torque sensors, and the imperfect Cubic model

    for harmonic drive friction was detected according to

    friction identification experiments. A Cartesian

    impedance control law was introduced by virtual

    decomposition [16] to realize the compliance control

    which incorporated with three means to make the

    flexible manipulator come into compliant contact with

    the objects.

    Compared with the above singular perturbations,passivity theories and decoupling methods, the

    backstepping technique was barely applied to the flexible

    joint robot control except the bcakstepping method on

    Cartesian impedance control of flexible joint

    manipulators [17] even though it represented a complete

    solution of impedance control problem for flexible joint

    robot model including the tracking case and inertia

    shaping. On the contrary, the backstepping technique was

    widely used to design output or state feedback controller

    for flexible joint robots in the tracking case [1826].

    Because of these issues, a new impedance controllerbased on dynamic surface control-backstepping to

    actualize the anticipated dynamical relationship between

    the motion of end-effector and the external torques was

    presented. The dynamic surface control technique was

    able to resolve the explosion of terms problem in

    backstepping controller. The Lyapunov function

    guaranteed global stability of the proposed controller.

    Moreover, a DSP/FPGA-FPGA hardware structure was

    established to support the proposed impedance controller.

    2 Dynamics of flexible joint robots

    Generally, the reduced dynamic model of an n-link

    flexible joint robot refers to robot dynamics and actuator

    dynamics, which could be described as the following

    form proposed by SPONG [27]:

    ext( ) ( , ) ( )+ + = +M q q C q q q g q&& & & (1)m

    + =B

    && (2)( )= K q (3)

    where nq R and n R denote the link and motor

    side positions, respectively; ( ) ,n n

    M q R ( , ) C q q&

    Rn and ( ) ng q R denote the inertia matrix, centripetal

    and Coriolis vector, and gravity vector, respectively;n

    R denotes the joint torque; extn

    R denotes the

    external torque that shown by the manipulators

    environment; mn

    R denotes the motor torque served

    as the control input; the constant positive definite

    diagonal matrices n nK R and n nB R represent

    the joint stiffness and the actuator inertia, respectively.

    Moreover, two well-known properties of the robot model

    utilized in the following sections are described as.

    Property 1: Link inertia matrix M(q) is symmetric

    and positive definite:

    T T( ) ( ) , ( )=M q M q y M q y0, , 0 n q y R (4)

    Property 2: Matrix ( ) 2 ( , )M q C q q& & is skew

    symmetric, if ( , )C q q& is chosen suitably using the

    Christoffel symbol, then:

    T ( ( ) 2 ( , )) 0, , , n = y M q C q q y y q q& & & R (5)

    These properties were proved by SCIAVICCO and

    SICILIAND [28].

    3 Backstepping based Cartesian impedancecontrol

    The most important contribution of this work was to

    develop a design of impedance controller. Similar to the

    state transformation, Eqs.(1)(3) could be rewritten in

    the Cartesian coordinates x=f(q) with the Jacobian

    J(q)=f(q)/q, Cartesian velocities ( ) ,=x J q q& &

    accelerations ( ) ( )= +x J q q J q q&&& && & and torque variables

    and & as:

    T Text( ) ( , ) ( ) ( ) ( ) ( )

    + + = +x x x x x J q g q J q && & & (6)

    1m

    + = BK Bq&&&& (7)

    where matrices ( )x and ( , ) x x& were given in

    Ref.[17].

    Considering that only the non-redundant and

    non-singular case was treated in this work, thus it isassumed that the manipulators Jacobian J(q) has full

    row rank in the considered region of the workspace.

    External torques ext should be related to the vector

  • 7/30/2019 fulltext_139

    3/9

  • 7/30/2019 fulltext_139

    4/9

    J. Cent. South Univ. Technol. (2010) 17: 807815810

    1 2

    T T1 2 1 1 2 3 ext

    3 4

    14 m 3

    ( ) ( , ) ( ) ( ) ( ) ( )

    ( )

    =

    = + +

    =

    =

    x x

    x x x x x J q g q J q x

    x x

    x KB Bq x

    &

    & &

    &

    & &&

    (15)

    First, define the multiple dynamic surfaces S1, S2, S3

    and S4 as follows:

    =

    =

    =

    =

    4d44

    3d33

    2d22

    d111

    xxS

    xxS

    xxS

    xxS

    (16)

    where x1d denotes the desired trajectory; and x2d, x3d and

    x4d denote the stabilizing functions for the subsystem

    consisting of dynamic surfaces S2, S3 and S4, respectively.

    According to Eq.(15), the derivatives of Eq.(16) could be

    expressed as follows:

    1 1 1d 2 2d

    1 T2 2 2d 1 1 1 2

    T3 ext 2d

    3 3 3d 4 3d

    14 4 4d m 3 4d

    = ( )[ ( , ) ( ) ( )

    ( ) ( )]

    = ( )

    = =

    = +

    +

    = =

    =

    S x x x x

    S x x x x x x J q g q

    J q x x

    S x x x x

    S x x KB Bq x x

    & & & &

    & & & &

    &

    & & & &

    & & & && &

    (17)

    To stabilize dynamic system (17), i.e. S1, S2, S3 and

    S4 0, the following control algorithms could be

    obtained:

    2 1 1 1d

    T3 1 1 2 ext

    1 2 2 2d

    4 3 3 3d

    1m 3 4 4 4d

    = ( ){ ( , ) ( ) [ ( ) ]

    ( )( )}

    ( )

    q

    = +

    + +

    +

    = + = + + +

    x S x

    x J q x x x J q g

    x S x

    x S x

    Bq x BK S x

    &

    &

    &

    &

    && &

    (18)

    where 1, 2, 3 and 4 are positive design parameters.

    However, distinct from the approach of backstepping

    that was described in the previous section, x2d was

    obtained by 2x through a first-order filter with time

    constant t2,

    2 2d 2d 2

    2d 2(0) (0)

    t + =

    =

    x x x

    x x

    &

    (19)

    Similar to x2d, x3d and x4d could be obtained. So,

    actual control vector m could be described as

    1 4 4dm 3 4 4

    4t = + + +

    x xBq x BK S&& (20)

    Substituting expression x3= into Eq.(20) leads to

    1 4 4dm 4 4

    4t

    = + + +

    x xBq BK S&& (21)

    Comparing controller (14) with controller (21), the

    proposed impedance controller (21) for flexible joint

    robots could resolve the explosion of terms problem ofbackstepping impedance controller (14) by evaluating

    the repeated derivatives of the virtual controllers. That is,

    the proposed DSC-backstepping control system for

    flexible joint robots used the outputs of first-order filters

    with 32 , xx and 4x as the inputs, which were

    required in place of 2 3,x x& & and 4x

    & in the

    backstepping design procedure. Thus, the proposed

    controller design procedure based on DSC-backstepping

    technique could be very simple.

    4.2 Stability analysis

    In this section, the stability analysis of the proposed

    control approach on the 5-DOF flexible joint robots was

    given. First, choose the Lyapunov function as follows:4 3

    T T

    1 1

    , , , ) ( )1

    (2

    i i i i i i ii i

    V= =

    += + e x t S S e eS %

    T T T T1 1 2 1 2 3 3 4 4[ ( ) ]

    1

    2+ + +x x x x x x x x x% % % % % % % % (22)

    where Si denotes the surface error; ei denotes the

    boundary layer error; ( )i i i i= x x x x% % denotes the

    observation error; and ix denotes the the estimation

    value ofxi.

    Lyapunov function (22) was differentiated with

    respect to time, and Si, ei, ,~

    ix triangle inequality and

    property P2 were utilized. The derivative could be

    obtained as follows:

    T1 2 1 1 1 2

    T 1 T2 1 3 2 2 2 1 2

    T T3 4 3 3 3 4 4 4 4 2 4

    3T T T T

    1 1 2 1 2 3 311

    , , , ) ( )

    [ ( ) ( ) ( ) ]

    ( ) ( )

    ( ) ( )

    (

    i i i

    ii i

    ii

    k

    k

    V

    +=

    + + +

    + + +

    + + + + +

    + + + + +

    =

    e x t S e S x

    x J q S e S x

    S e S x S x

    ee x x x x x x x

    t

    S S

    S

    S S

    % %

    %

    % %

    & & &% % % % % %

    &

    T T4 4 2 1 2

    1 ( )2

    +x x x x x& &% % % % 4 32 2

    1 1

    i i i i

    i i

    = =

    S e

    322 2 2 2

    1 1 2 2 3 3 4 41

    ii

    l l l l c=

    + =x x x x% % % % V + (23)

    where l1, l2, l3, l4, k1 and k2 represent design parameters;

    , i and ci represent positive constants; and , V

    represent design function.

    The derivative implied that , , , )( i i iV e x tS %& 0

    when V=p and /p. Therefore, Vp was an invariant

    value, in other words, ifV(0)p, then V(t)p fort0.And the surface errorSi could be arbitrarily small to

    ensure stability of the controller.

    The proposed DSC-backstepping Cartesian

  • 7/30/2019 fulltext_139

    5/9

    J. Cent. South Univ. Technol. (2010) 17: 807815 811

    impedance control scheme for flexible joint manipulators

    could be summarized by the DSC system. Accordingly,

    our control system could be designed more easily and

    simply than the Cartesian impedance control system via

    the backstepping technique.

    5 Experiment analysis

    5.1 Structure and physical parameters of 5-DOF

    flexible joint robots

    In order to demonstrate the effectiveness of the

    proposed DSC-backstepping Cartesian impedance

    controller, firstly, the structure of 5-DOF flexible joint

    manipulators is shown in Fig.1 and the kinematic

    parameters and frame assignments are shown in Fig.2.

    The flexible joints of the robot are identical in the macro

    structure, which are driven by brushless DC andharmonic drive gear combined planet gear (gear ratio

    1:12 000). A potentiometer and three hall sensors were

    equipped to measure the absolute angular position of the

    joint and the relative angular position of the motor. Eight

    strain gauges were fixed crossly to the output shaft of the

    harmonic drive gear to construct two full-bridges that

    measured the joint torques. The measurements of the

    angular position and the torque were fed to the Cartesian

    impedance controller. Moreover, the five joints were

    connected in series from the base toward the tip, joint i,

    connected link i with link i1. Five frames, {Li}, i=1,

    2, , 5, were defined in a way that frame {Li} was fixed

    to linki with itsZ-axis coincident with the ith joint axis,

    then the rotations by X-, Y-axes were constrained to the

    coordinate origin.

    The robot parameters of kinematics and dynamics

    Fig.1 Five-DOF flexible joint robot

    Fig.2 Kinematic parameters and frame assignments for 5-DOF

    flexible joints of robot

    were very precisely computed using 3D mechanical

    CAD programs. Finally, K could be calculated by Eq.(3)

    in the joint impedance control when each joint contacted

    a rigid environment in which q was constant. The

    manipulator parameters are listed in Table 1. Where ai, i,

    di and i represent Denavit-Hartenberg (D-H) parameters;

    mi represents the joint quality; Bi represents the joint

    damp; andKi represents the joint stiffness.

    5.2 DSP/FPGA-FPGA based hardware system

    The hardware system based on DSP/FPGA-FPGA

    as shown in Fig.3 was given to realize the proposed

    controller. In order to minimize cabling and weight of the

    5-DOF flexible joint manipulator, a fully mechatronic

    design methodology was introduced to develop the

    hardware system. All the analog signals were converted

    into proper digital signals and serially transmitted into

    joint FPGA board and further to PCI (peripheral

    component interconnect)-based central processor. The

    hardware system consisted of PCI-based DSP/FPGA

    board configured as a Cartesian level and joint FPGA

    board for five-joint control configured as a joint level.

    The control algorithm is illustrated in Fig.4. Joints

    FPGA board (Slave) took charge of the joint level

    controller, and a PCI-based DSP/FPGA board (Master)executed as Cartesian level. In the joint control level, the

    FPGA technology was chosen to achieve a more flexible

    implementation of the joint controller with a high control

    rate and a small sized joint electronics.

    To implement real time control of the robot, the

    Table 1 Manipulator parameters

    Frame ai/mm i/() di/mm i/() mi/kg Bi/(kgm2) Ki/(Nmrad

    1)

    {L1} 0 90 110.76 0 0.800 0.85 82.175

    {L2} 530 0 0 0 1.090 0.85 68.236

    {L3} 470 0 0 0 1.069 0.85 68.236

    {L4} 0 90 135.66 0 0.700 0.85 68.236

    {L5} 0 90 75.26 0 0.700 0.85 68.236

  • 7/30/2019 fulltext_139

    6/9

    J. Cent. South Univ. Technol. (2010) 17: 807815812

    Fig.3 Hardware system based on DSP/FPGA-FPGA

    Fig.4 Block diagram of controller

    Cartesian level needed the feedback information of

    positions, velocities and torques of the joints and

    calculated the required torques swiftly. At the same time,

    the joint level should update the input data in time

    especially for the transient state. Therefore, a high speed

    data bus of point-to-point serial communication (PPSeCo)

    was designed for this requirement, in which the cycle

    time is less than 200 s and communication rate is up to

    25 Mb/s. The communication and other control programs

    for FPGA were written in VHDL and run in FPGA.

    5.3 Experiments

    To illustrate the validity of the proposed methods,

    the following three experiments on manipulator were

    carried out.

    The first experiment was to use the 5-DOF flexible

    joint robot to track sine curve inX- and Y-directions, and

    remain static in Z-direction for the case of free motion

    (Fext=0) by backstepping impedance control and

    DSC-backstepping impedance control in Cartesian space.

    Fig.5 shows Cartesian coordinate positions and tracking

    errors in different directions under the methods of

    backstepping impedance control (BIC) and DSC-

    backstepping impedance control (DSC-BIC). From

    Figs.5(b) and (d), it can be seen that the tracking errorswere 3.5 and 2.5 inX- and Y-directions by backstepping

    impedance control, while the tracking errors were 2.0

    and 1.5 mm inX- and Y-directions by DSC-backstepping

  • 7/30/2019 fulltext_139

    7/9

    J. Cent. South Univ. Technol. (2010) 17: 807815 813

    Fig.5 Cartesian coordinate positions and tracking errors in different directions: (a) Positions in X-direction; (b) Errors inX-direction;

    (c) Positions in Y-direction; (d) Errors in Y-direction

    impedance control, respectively. The control parameters

    of two controllers are listed in Tables 2 and 3. According

    to Fig.5, the accuracy of the tracking errors was

    improved evidently by using DSC-backstepping

    impedance control.

    In addition, the trajectory tracking accuracy of

    impedance control was lower than that of position

    control because the impedance control sacrificed some

    Table 2 Parameters of backstepping impedance control

    Joint No. KP,i KD,i

    1 33.955 3.480

    2 27.354 2.804

    3 27.354 2.257

    4 27.354 2.325

    5 27.354 2.325

    Table 3 Parameters of DSC-backstepping impedance control

    Filter No. Positive design parameter (i) ti/s

    1 14.0

    2 0.8 0.01

    3 1.5 0.01

    4 16.0 0.01

    tracking accuracy. Moreover, the more flexible the joint,

    the lower the accuracy of track, even not meeting the

    requirements. The tracking accuracy would be improved

    when the joint with little flexibility was regarded as a

    rigid joint, but the impedance performance would be

    down. Therefore, the above experiment compromised the

    impedance performance and the tracking accuracy. And

    that the phenomenon of the error was asymmetric about

    X-axis in Fig.5 was caused by the imperfect gravity and

    friction compensation.

    The second experiment was Cartesian impedance

    experiment made on a 5-DOF flexible joint manipulator.

    Firstly, the robot was placed on a virtual equilibrium

    position CD=[0, 0, 0]. The anticipant stiffness (Ki) and

    damping (Di) in Table 4 were used. Then, the robot was

    pulled in different directions shown in Fig.6. Finally, the

    robot overcame the gravity and returned to the CD as

    soon as the force was released. Fig.6 shows the

    corresponding Cartesian forces along with the Cartesian

    position. It could be concluded that the theoretic

    Cartesian impedance behavior was achievedsuccessfully.

    In Cartesian impedance control, the desired

    Cartesian impedance stiffness could be set a small value

  • 7/30/2019 fulltext_139

    8/9

    J. Cent. South Univ. Technol. (2010) 17: 807815814

    Table 4 Impedance parameters in workspace

    Coordinate Ki/(Nm1) Di/(Nsm

    1)

    X-axis 2 000 482.6

    Y-axis 2 000 482.6

    Z-axis 2 000 78.3

    Fig.6 Variation of Cartesian coordinates in DSC-backstepping

    Cartesian impedance control: (a) Position; (b) Force

    (smaller than 20 N/m). Under this condition, the robot

    could be pushed freely using a small force, and when the

    force was removed, the robot could stay at the last

    position stably, which was called zero-force control.

    The third experiment, which was based on DSC-

    backstepping Cartesian impedance controller, was to test

    whether the end-effector could process a compliant

    behavior under a constrained space. The end-effector of

    the manipulator anticipant position was to make a circle

    motion with a cylindrical obstacle during the trajectory

    in Fig.7. During the motion, robot maintained a larger

    contact force in the tangent direction of the obstacle,

    while a smaller retention in the normal direction. Then,

    by the geometric calculation, the forces of the tangent

    direction and normal direction were converted to the

    contact force of directions X and Y on X-Y plane.According to Fig.7, it could be concluded that the

    DSC-backstepping based impedance controller realized

    the compliant behavior of flexible joint robot when it

    contacted with stiffness environment. The control

    parameters are listed in Table 5, whereFi represents the

    external force.

    Fig.7 Circle tracking in X-Y plane constrained space by

    DSC-backstepping based impedance controller

    Table 5 Impedance parameters in constrained space

    Coordinate Ki/(Nm1) Di/(Nsm

    1) Fi/N

    X-axis 1 500 20 5

    Y-axis 1 500 20 5

    Z-axis 1 500 20 5

    Therefore, if expectation torques of the tangent

    direction and normal direction of end-effector were kept

    within an acceptable range, not only the robot itself and

    the object could be protected, but also the smooth surface

    obstacles could be avoided effectively.

    6 Conclusions

    (1) The DSP/FPGA-FPGA based special hardware

    system with PPSeCo is established. The hardware

    structure is designed not only to achieve thecommunication between joint level and Cartesian level,

    but also to calculate acceleration q&& and the jerk q&&&

    which cannot be measured directly, and actualize the

    control algorithm.

    (2) Two Cartesian impedance controllers based on

    backstepping and DSC-backstepping technique are

    proposed, the design of the latter is easier and simpler

    than that of the former because of the introduction of

    dynamic surface control technique which resolves the

    explosion of terms problem of backstepping-based

    impedance controller. And aformal stability of the DSC-backstepping impedance controller is proved based on

    Lyapunov function.

    (3) Experimental results justify that the proposed

  • 7/30/2019 fulltext_139

    9/9

    J. Cent. South Univ. Technol. (2010) 17: 807815 815

    DSC-backstepping impedance controller gives

    satisfactory tracking performance. And the impedance

    performance in free space and the compliant behavior in

    constrained space are achieved on the 5-DOF flexible

    joint robot.

    References

    [1] HOGAN N. Impedance control: An approach to manipulation:

    Theory (part ); Implementation (part ); Applications (part

    ) [J]. Journal of Dynamics Systems, Measurement, and Control,

    1985, 107(11): 124.

    [2] KAZEROONI H, SHERIDN T B, HOUPT P K. Robust compliant

    motion for manipulators: The fundamental concepts of compliant

    motion (part ); Design method (part ) [J]. IEEE Journal of

    Robotics Automation, 1986, 2(2): 83105.

    [3] ALIIN A, CHRISTIAN O, UDO F, GERD H. Cartesian impedance

    control of redundant robots: Recent results with the DLR-light-weight-arms [C]// Proceedings of the 2003 IEEE International

    Conference on Robotics & Automation. Taipei: IEEE, 2003:

    37043709.

    [4] HUANG J B, XIE Z W, LIU H, SUN K, LIU Y C, JIANG Z N.

    DSP/FPGA-based controller architecture for flexible joint robot with

    enhanced impedance performance [J]. Journal of Intelligent and

    Robotic Systems, 2008, 53(3): 247261.

    [5] KHALIL H K. Nonlinear systems [M]. London: Prentice Hall, 2002:

    9395.

    [6] ALIN A, GERD H. Cartesian impedance control techniques for

    torque controlled light-weight robots [C]// Proceedings of the 2002

    IEEE International Conference on Robotics & Automation.

    Washington D C: IEEE, 2002: 657663.

    [7] CHRISTIAN O, ALIIN A, ANDREA K, GERD H. Decoupling based

    Cartesian impedance control of flexible joint robots [C]//

    Proceedings of the 2003 IEEE International Conference on Robotics

    & Automation. Taipei: IEEE, 2003: 31013107.

    [8] CHRISTIAN O, ALIIN A, ANDREA K, GERD H. A passivity based

    Cartesian impedance controller for flexible joint robots: Part .

    Torque feedback, and gravity compensation [C]// Proceedings of the

    2004 IEEE International Conference on Robotics & Automation.

    New Orleans: IEEE, 2004: 26592665.

    [9] ALIIN A, CHRISTIAN O, GERD H. A passivity based Cartesian

    impedance controller for flexible joint robots: Part II. Full state

    feedback, impedance design and experiments [C]// Proceedings of

    the 2004 IEEE International Conference on Robotics & Automation.

    New Orleans: IEEE, 2004: 2666

    2672.[10] CHRISTIAN O, ALIIN A, ANDREA K, GERD H. On the

    passivity-based impedance control of flexible joint robots [J]. IEEE

    Transaction On Robotics, 2008, 24(2): 416429.

    [11] OZAWA R, KOBAYASHI H. A new impedance control concept for

    elastic joint robots [C]// Proceedings of the 2003 IEEE International

    Conference on Robotics & Automation. Taipei: IEEE, 2003:

    31263131.

    [12] GIANNI F, GIANANTONIO M, PAOLO R. Impedance control for

    elastic joints industrial manipulators [J]. IEEE Transactions on

    Robotics and Automation, 2004, 20(3): 488498.

    [13] CHIEN M C, HUANG A C. Regressor-free adaptive impedance

    control of flexible-joint robots using FAT [C]// Proceedings of the

    2006 American Control Conference. Minneapolis: IEEE, 2006:

    39043909.

    [14] HUANG J B, XIE Z W, LIU H, SUN K. Adaptive Cartesian

    impedance control system for flexible joint robot by using

    DSP/FPGA architecture [J]. International Journal of Robotics and

    Automation, 2008, 23(4): 251258.

    [15] LIU H, LIU Y C, JIN M H, SUN K, HUANG J B. An experimental

    study on Cartesian impedance control for a joint torque-based

    manipulator [J]. Advanced Robotics, 2008, 22(4): 11551180.

    [16] HUANG J B, XIE Z W, JIN M H, JING Z N, LIU H. Adaptive

    impedance-controlled manipulator based on collision detection [J].

    Chinese Journal of Aeronautics, 2009, 22(1): 105112.

    [17] CHRISTIAN O. Cartesian impedance control of flexible joint

    manipulators [D]. Munchen: Universitat Des Saarlandes, 2005: 80

    82.

    [18] HUANG A C, CHEN Y C. Adaptive sliding control for single-link

    flexible-joint robot with mismatched uncertainties [J]. IEEE

    Transactions on Control Systems Technology, 2004, 12(5): 770775.

    [19] MIN S K, JIN S L. Adaptive tracking control of flexible-joint

    manipulators without overparametrization [J]. Journal of Robotic

    Systems, 2004, 21(7): 369379.

    [20] YANG Y S, FENG G, REN J S. A combined backstepping and

    small-gain approach to robust adaptive fuzzy control for strictfeedback nonlinear systems [J]. IEEE Transactions on Systems, Man,

    and Cybernetics, Part A: Systems and Humans, 2004, 34(3):

    406420.

    [21] WITHI C, PETER H M. Robust observer backstepping neural

    network control of flexible joint manipulator [C]// Proceeding of the

    2004 American Control Conference. Boston: IEEE, 2004: 5250

    5255.

    [22] CHRIS J B M, ELEUTERIO G M T D, MENG M. CMAC adaptive

    control of flexible-joint robots using backstepping with tuning

    functions [C]// Proceedings of the 2004 IEEE International

    Conference on Robotics & Automation. New Orleans, IEEE, 2004:

    26792686.

    [23] WITHI C, PETER H M. Backstepping high-order differential neural

    network control of flexible-joint manipulator [C]// Proceedings of the

    2005 American Control Conference. Portland: AACC, 2005: 1377

    1382.

    [24] WITHI C, PETER H M. N Motion control of two link flexible joint

    robot, using backstepping, neural networks, and indirect method [C]//

    Proceedings of the 2005 IEEE Conference on Control Applications.

    Toronto, IEEE, 2005: 601605.

    [25] YOO S J, PARK J B, CHOI Y H. Adaptive dynamic surface control

    of flexible-joint robots using self-recurrent wavelet neural networks

    [J]. IEEE Transactions Systems, Man, and Cybernetics, Part B:

    Cybernetics, 2006, 36(6): 13421355.

    [26] YOO S J, CHOI Y H, PARK J B. Dynamic surface controller for

    flexible joint robot without velocity measurements [C]// Proceedings

    of 2007 International Conference on Control, Automation andSystems. Seoul: IEEE, 2007: 10981102.

    [27] SPONG M W. Modeling and control of elastic joint robots [J].

    Journal of Dynamic Systems, Measurement, and Control, 1987,

    109(1): 310319.

    [28] SCIAVICCO L, SICILIANO B. Modeling and control of robot

    manipulators [M]. London: McGraw-Hill, 1996: 141143.

    [29] YIP P P, HEDRICK J K. Adaptive dynamic surface control: a

    simplified algorithm for adaptive backstepping control of nonlinear

    systems [J]. International Journal of Control, 1998, 71(5): 959979.

    [30] SWAROOP D, GERDES J C, YIP P P, HSDRICK J K. Dynamic

    surface control of nonlinear systems [C]// Proceedings of the

    American Control Conference. Albuquerque: IEEE, 1997: 3028

    3034.

    (Edited by LIU Hua-sen)