Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Proceedings of the 2003 IEEE/RSJ Intl.

Conference on Intelligent Robots and Systems Las Vegas, Nevada October 2003

Analysis of Virtual Fixture Contact Stability for Telemanipulation


Jake J. Abbott and Allison M. Okamura Department of Mechanical Engineering The Johns Hopkins University, Baltimore, MD 21218 {jake.abbott, aokamura}@jhu.edu Abstract
A forbidden-region virtual xture (FRVF) is a constraint, implemented in software, that keeps the slave manipulator of a master/slave telemanipulation system from entering into a forbidden region of the workspace. In this paper, we consider the problem of unstable vibrations of the slave against the FRVF for a general class of telemanipulator control architectures. The master and slave equilibrium positions resulting from a constant human input force are found, and the system is evaluated around this equilibrium. We consider two methods of analyzing the stability of the system around this equilibrium point. The rst method uses tools developed for analysis of two-port networks. The second method converts the system to its discrete state-space form, and then uses the position of the eigenvalues to analyze system stability. We nd that the discrete state-space method agrees with simulations, and can be easily used to design and analyze the stability and transient behavior of the telemanipulator. guration space. A simple method of implementing FRVFs is with an impedance surface, or virtual wall. This is a method common in haptic devices, where the position (velocity, acceleration) of the device through and normal to the impedance surface is used to generate an actuator force on the device. This paper is concerned with the implementation of these impedancetype FRVFs on telemanipulators where both the master and slave devices are of the impedance type (backdrivable, low mass). Implementing sti virtual walls on a haptic device with a digital computer is inherently nonpassive [7]. The maximum stiness of the surface that can be implemented is limited by the sampling rate of the computer, the resolutions of the position sensor and actuator, and the impedance of the mechanical device [5]. Implementing impedance-type FRVFs on a telemanipulator has added complexity because the slave manipulator may be simulateously interacting with the FRVF and a physical environment, especially in the case of a FRVF used to limit the depth of slave motion into an environment [1]. Also, it may be desirable to implement virtual xtures on both the master and slave devices simultaneously [1, 13]. In previous experiments [1], we found that implementing an impedancetype FRVF on the slave of a telemanipulator can lead to unwanted vibrations against the FRVF as the slave manipulator is pulled into the forbidden region by the master (see video). These vibrations are most likely the manifestation of an unstable system with nonlinearities resulting in a limit cycle, but they could also represent a marginally stable system. In either case, any sustained vibration is unwanted. This paper considers the general system where the master and slave devices can each have FRVFs of different stinesses but at the same position in their respective workspaces, and the position of the FRVF can be located at or below the surface of a compliant environment. We consider the underlying telemanipulation control architecture to have any combination of PD, velocity feedback, and force feedforward control. We ask the question: If a user attempts to apply a constant force to the master manipulandum of a telemanipulator, will the system reach a static equilibrium or will it vibrate, and can this system behavior be predicted without simulating or physically implementing
2699

Introduction

Telemanipulation typically refers to a system where a human operator manipulates a master robotic device, and a slave device emulates the behavior of the master, with varying types of feedback to the operator. One focus of current research in telemanipulation is in the eld of minimally-invasive surgery (MIS). In addition to traditional telemanipulation, we are interested in the application of virtual xtures for operator assistance in MIS tasks. The term virtual xture refers to a general class of guidance modes, implemented in software, that help a robotic manipulator perform a task by limiting its movement into restricted regions and/or inuencing its movement along desired paths [1, 4, 11, 12, 13, 14]. The potential benet of virtual xtures is safer and faster operation. Virtual xtures attempt to capitalize on the accuracy of robotic systems, while maintaining a degree of operator control. As their name implies, forbidden-region virtual xtures (FRVF) [13] prohibit the motion of a robot manipulator into forbidden regions of geometric or con0-7803-7860-1/03/$17.00 2003 IEEE

the system? The results of this research indicate that these unwanted vibrations can be predicted given an accurate model of the master and slave devices and some knowledge of the human operator and environment properties.

Ze(s) ZOH Xe(s) Zs-1(s) Cs(z) _ _

Fe*(s) + + Fe(s)

System Model
ZOH C4(z) C3(z) ZOH _ _ + Fh(s) Fh*(s) + _ Zh(s) _

++

We will adopt a system model similar to that of Lawrence [10]. The human and environment are modeled with the LTI models
Fh Fh Fe Fe

ZOH C2(z)

C1(z) ZOH

= =

Zh Xh Ze Xe

(1) (2)

Zm-1(s) Xh(s) Cm(z) ZOH

where Fh and Fe are the human/master and slave/environment interaction forces, respectively. They are dened positive in compression. Fh and Fe are dened as the exogenous human and environment inputs, respectively. We will only consider interactions with a passive environment (Fe = 0). Xh and Xe are the points where the human makes contact with the master device, and where the slave makes contact with the environment, respectively. Xe = 0 when the slave is just touching the surface of the environment (the environment is uncompressed). The systems are also assumed to be geometrically similar (Xh = Xe represents perfect position tracking). The human and environment are modeled as linear mass-spring-damper systems [15]. Written in Laplace form, they are Zh Ze = mh s2 + bh s + kh = me s2 + be s + ke (3) (4)

Figure 1: General telemanipulator implemented on a digital computer.

systems are written in z-domain and preceded by zeroorder hold (ZOH) blocks to indicate where sampling occurs. If the ZOHs are lumped with the controllers they precede, the master and slave actuator forces are given by Fma Fsa = C4 Xe + C2 Fe + Cm Xh = C1 Xh + C3 Fh Cs Xe (9) (10)

The master and slave device dynamics are governed by Fh Fma Fsa Fe = = Zm Xh Zs Xe (5) (6)

Equilibrium Positions

where Fma and Fsa are the master and slave actuator forces, respectively. The master and slave impedances are modeled as mass-damper systems. Zm Zs = m m s2 + b m s = ms s + bs s
2

(7) (8)

The equations above are often written in terms of Vh and Ve rather than Xh and Xe . This simply scales the impedances by a factor of s. Working in terms of velocity often simplies the analysis of stability of telemanipulators, but we consider positions here to explicitly keep track of the positions of the master and slave with respect to their FRVFs. Figure 1 shows the general telemanipulator system we consider. It is similar to that in [10], but we explicitly consider the digital nature of the controller; continuous systems are written in s-domain, and digital
2700

When a slave device interacts with a stable impedance-type FRVF, the slave continues to move forward if the master moves forward, establishing a new equilibrium point that balances the eects of the telemanipulators control system, the FRVF, and the environment. Experiments show that when the slave device vibrates on an unstable virtual xture, the center of the vibration moves forward if the master moves forward. It is also possible to generate vibrations that exist entirely within the FRVF. For this reason we believe the unstable vibrations occur not on surface of the FRVF, but rather around some unstable equilibrium below the surface of the FRVF. The equilibrium positions are found explicitly in this section, and whether or not they are stable or unstable is considered in Section 4. The goal here is to nd the equilibrium positions

for the general class of telemanipulation control architectures that can be described by the equations Fma Fsa = = Kpm (Xh Xe ) + Kdm (Xh Xe ) (11) +Kvm Xh + Kf m Fe + KmV F XmV F Kps (Xh Xe ) + Kds (Xh Xe ) Kvs Xe + Kf s Fh KsV F XsV F (12)

Environment Surface

FRVF Xh

Master

dVF Xe

XmVF

where Kpm and Kps are the position gains of the master and slave, Kdm and Kds are the derivative gains of the master and slave, Kvm and Kvs are additional velocity feedback gains of the master and slave, Kf m and Kf s are the force feedforward gains of the master and slave, KmV F and KsV F are the stiness gains of the master and slave FRVFs, and XmV F and XsV F are the depth of the master and slave through their respective FRVFs. This general class includes many common control architectures [1]. We will only nd the equilibrium positions of this class of controllers, but the methods used here could be applied to other control systems. We consider the case where the human operator has moved the telemanipulator to a position where the slave has at least reached its FRVF, and the opera tor is now trying to apply a constant force Fh to the master manipulandum. The actual force felt between the human and master will be a function of the feedback system, and will deviate from the desired force by Fh . The force Fe will also have an equilibrium value. Fh Fe = = Fh + Fh Fe + Fe (13) (14)

XsVF (a) KfmFe Kpm(Xh-Xe) KmVFXmVF Slave K eX e KsVFXsVF (b) Master

Slave

Fh

KfsFh Kps(Xh-Xe)

Figure 2: Telemanipulator equilibrium resulting from constant Fh : (a) position (b) static force balance.

15 and 17 into Equations 1 and 2, respectively.


Fh Fh Fe

Let dV F be the depth of the virtual xture below the surface of the environment. We consider the case where dV F 0. Let Xh and Xe be the equilibrium positions of the system. They will be dened by the relationships Xh Xh Xe Xe = Xh + Xh = dV F + XmV F = Xe + Xe = dV F + XsV F (15) (16) (17) (18)

= =

Kh Xh e Ke X

(21) (22)

The equilibrium position of the slave manipulator can now be found as a function of the control system gains and Fh . The equilibrium position of the master device can then be written compactly as a function of XsV F . XsV F = [Fh (Kf m Ke (23) +(Kpm + KmV F )Ke /Kps )dV F ] /[Kf m Ke Kpm + (Kpm + KmV F ) = (Kps + Ke + KsV F )/Kps ] Ke dV F /Kps (24) +(Kps + Ke + KsV F )XsV F /Kps

and will be functions of Fh and the control system gains. Let us consider the equilibrium position shown in Figure 2. The master and slave devices are in static equilibrium governed by the equations Fh KmV F XmV F = Kf m Fe + Kpm (Xh Xe ) Ke Xe + KsV F XsV F = Kps (Xh Xe ) + Kf s Fh (19) (20)

XmV F

These equations assume that the master device has some form of haptic feedback. In the case where the master is completely unactuated, a constant force Fh will not lead to an equilibrium. In this case the human user can control XmV F directly, and Kps XmV F Ke dV F XsV F = Kps + Ke + KsV F
2701

At equilibrium, the forces experienced by the human and environment are found by substituting Equations

(25)

We would now like to analyze the stability of the system in Figure 1 around the equilibrium point asso ciated with a given Fh . The actuator control signals of Equations 11 and 12 do not meet the required form of Equations 9 and 10. This stems from allowing the FRVF to be in a dierent location than the surface of the environment (allowing dV F = 0), making the system nonlinear and prohibiting the use of currently available analytic stability techniques. But, by substituting Equations 13-18 into the original system equations of Section 2, and making use of the fact that Zm and Zs are mass-damper systems with no spring terms, resulting in Fh = Fma and Fsa = Fe at equilibrium, the system can be rewritten as Zh Xh Zm Xh Zs Xe Ze Xe = = = = Fh Fh C4 Xe C2 Fe (Cm + KmV F )Xh C1 Xh + C3 Fh (Cs + KsV F )Xe Fe Fe (26) (27) (28) (29)

reduced. Second, even if the equilibrium is found to be stable, this still may not lead to desirable system behavior. We would like any oscillations in the system to have small magnitude and decay away very quickly. With these two points in mind, we may need a more conservative design criterion than equilibrium stability to assure that the model is predicting true system behavior, and the system is well-behaved when interacting with the FRVF.

4.1

Two-Port Method

First we consider the two-port circuit representation of a telemanipulator [8]. We will use stability techniques developed for linear systems in the frequencydomain, so the digital components of the system of Figure 1 must be converted to Laplace form. The digital blocks C1 (z), C2 (z), C3 (z), C4 (z), Cm (z), and Cs (z) all are converted to s-domain using z-Transform inversion. The zero-order hold can be written exactly as a transfer function 1 esT (30) s where T is the sampling period of the control computer [6]. We will then convert the linear impedance and admittance transfer functions back into functions of Vh and Ve , rather that Xh and Xe . This is done by scaling the transfer function by a factor of s. Writing in terms of velocity will facilitate the use of stability techniques that are currently available. Hashtrudi-Zaad and Salcudean [9] give a hybrid two-port form of the telemanipulator in Figure 1. The hybrid two-port ZOH(s) = Fh Ve can be formed with = h11 h21 h12 h22 Vh Fe (31)

Equations 26-29 represent a dual system of the one given in Figure 1 and the original system equations. The dual system is written in terms of positions and forces around the equilibrium, and the master and slave FRVFs are lumped with local master and slave controllers. Because this dual system, which includes virtual xtures, can be written as a linear system around an equilibrium position, the stability of this equilibrium can be analyzed using techniques already available in the telemanipulation literature. Note that this dual system has no exogenous force inputs (compare Equations 26 and 29 to Equations 1 and 2, respectively).

Equilibrium Stability

h11 h12 h21 h22

We will compare two methods for analyzing the stability of the equilibrium discussed in the previous section. First we consider a two-port circuit representation [8] of the telemanipulator, and we will use Llewellyns absolute stability criteria [3] to evaluate the stability of the two-port. Second, we will consider a state-space representation of the telemanipulator, and we will use the system eigenvalues to determine stability. We now note two key observations that should be considered in stability analysis. First, the system model presented in the equations of Section 3 and Figure 1 assume that the human is always connected to the master, and that the environment is always connected to the slave, which generally will not be the case. For a well-designed FRVF, a constant Fh should lead to no unwanted oscillations, but if large oscillations are present the accuracy of the model will be
2702

Zcm Zcs + C1 C4 Zcs C3 C4 C2 Zcs C4 = Zcs C3 C4 C3 Zcm + C1 = Zcs C3 C4 1 C2 C3 = Zcs C3 C4 =

(32) (33) (34) (35)

where Zcm = Zm + Cm and Zcs = Zs + Cs . In the preceding equations, the ZOHs have already been lumped with the transfer functions that they precede. Llewellyns absolute stability criteria give the necessary and sucient conditions for unconditional stability of the hybrid two-port network of Equation 31 [3]. Unconditional stability means that there exists no set of passive human and environment one-ports that will make the combined system unstable. Llewellyns absolute stability criteria are Re [h11 (i)] 0 (36)

Vh'

Vh

Zh

Zh,max

Ve + Telemanipulator + Two-Port Fh Fe Ze,max Network _ _ H

Ve'

characteristics of eigenvalues in the z-plane are related to the those in the s-plane by [6] z = esT s = ln(z) T (43)

Ze

H'
Figure 3: Shunting the human and environment impedances with their maximum expected values, forming a new hybrid two-port matrix.

Also, we use the fact that for stable eigenvalues in the s-plane, the damping ratio is found by = sin tan1 |Re[s]| |Im[s]| (44)

We will consider the eigenvalues location and damping ratio in the next section. 2Re [h11 (i)] Re [h22 (i)] (37) |h12 (i)h21 (i)| + Re [h12 (i)h21 (i)] for all 0. Allowing for any possible passive human and environment will give an overly conservative result, so [9] showed it is possible to shunt the human and environment impedances Zh and Ze with impedances that bound their maximum values, Zh,max and Ze,max . This results in the system shown in Figure 3. The shunt impedances can be lumped into the the original hybrid matrix, resulting in a new hybrid matrix, where Fh Ve h11 h12 h21 h22 = = = = = Vh h11 h12 (38) Fe h21 h22 h11 Zh,max (39) Zh,max + h11 h12 Zh,max (40) Zh,max + h11 h21 Zh,max (41) Zh,max + h11 h12 h21 1 h22 + (42) Zh,max + h11 Ze,max

Comparison to Simulations

Llewellyns absolute stability criteria can now be applied directly to this new hybrid two-port network, and the results should no longer be overly conservative.

4.2

State-Space Method

We have created an algorithm for numerically generating a state-space formulation for the discrete system of Figure 1 and Equations 11 and 12 [2]. This state-space model has two inputs (sampled Fh and Fe ) and ten states (Xh , Xh1 , Xh2 , Xh3 , Xe , Xe1 , Xe2 , Xe3 , Fh1 , and Fe1 ). A subscript n indicates the value of that state n samples ago. This discrete state-space model was designed as a discrete system from the beginning, and is not simply a digital emulation of a continuous state-space model. This discrete system has ten eigenvalues, but at least two of them are identically equal to zero in the z-plane, due to the structure of the state-space model. The location of the remaining eigenvalues reveal stability and transient properties of the system. Equivalent
2703

To validate the models developed in the previous section, we now compare Simulink simulations with the predictions from Llewellyns absolute stability criteria and the eigenvalues of the discrete state-space model. The Simulink simulation we have developed runs the continuous time at 10,000 Hz, and the controller sampling time at 500 Hz (T = 0.002). The simulation implements the continuous and discrete subsystems as they are given in Figure 1. The simulation is not written about the equilibrium point of Section 3, and does not assume constant contact. The simulation should accurately represent a physical system for the purpose of validating the stability techniques of Section 4. For any analytical model, we would like to have no false positives for stability, meaning that if the model predicts the system will be stable, then the system will be stable. It is acceptable, from a safety standpoint, if the analytical model gives some false negatives for stability, but ideally these would be minimized. Using simulations to predict if a system is unstable is a dicult (if not impossible) task. If an unstable simulation is found, then the system can be considered unstable, but if no unstable simulation is found, we cannot conclude that the system is stable. It may be the case that we simply have not stumbled upon an unstable case.

5.1

Two-Port Method

For the two-port stability model to be useful, if the stability criteria returns a stable result, the simulation must be stable for all valid human and environment impedances. By valid, we mean impedances with parameters less than or equal to the maximum upper-bound impedances that were introduced in Section 4.1 and incorporated into the two-port network H of Figure 3. It is also desirable for an unstable result to correspond to an unstable simulation for some valid human and environment. To mitigate the inadequacy of using simulations to validate this analytical model,

Re(h11) 25 20 15 10 5

0.15

0.1

0.05

0 4 10

10

10

10

10
22

10

10

10

10

2Re(h )Re(h ) |h h | Re(h h )


11 12 21 12 21

0.05

2.5 2 1.5
0.15 0.1

1 0.5 0 4 10

Xh Xe 0.2
3 2 1 0 1 2 3 4

0.1

0.2

0.3

0.4

0.5 Time

0.6

0.7

0.8

0.9

10

10

10

10

10

10

10

10

Figure 4: Llewellyns absolute stability criteria.

Figure 5: Unstable Simulation corresponding to stable prediction of two-port method.

5.2
we simulate systems where Zh = 0 or Zh = Zh,max and Ze = 0 or Ze = Ze,max . By considering the four possible combinations of these impedances, we should hopefully nd a worst-case system that will test the eectiveness of the model. To evaluate Llewellyns stability criteria, we numerically calculate Equations 36 and 37 for a large range of > 0, and analyze the resulting plots, which should be positive for all if the system is stable. Because the model is continuous, the values of Equations 36 and 37 asymptote to constant values in the limit as 0 and . We need only plot the range of where the values from Equations 36 and 37 are changing. Figure 4 shows Llewellyns stability criteria for mh,max = 1, bh,max = 89, kh,max = 2000, me,max = 1, be,max = 63, ke,max = 1000, mm = ms = 1, bm = bs = 1, Kpm = Kps = 4000, KsV F = 5000, and all other control gains equal to zero. Notice that both plots are positive for all values of , meeting the requirement for unconditional stability. This means that any valid human and environment should be stable for this telemanipulation control system and virtual xture. But, Figure 5 shows a simulation for a system with mh = 0.1, bh = 8.9, kh = 200, me = be = ke = 0, dV F = 0.01, and Fh = 100, where the system clearly becomes unstable. This simulation was not hard to nd; if fact, these types of false positives for stability are very common with this method. And, because the numerical values seen in Figure 4 have no clear physical interpretation, this method may be useless in predicting stability of virtual xtures. Even though the ZOH is included explicitly, it appears that this model does not capture the nuances of the discrete system.
2704

State-Space Method

The rst thing required by this method is that all of the eigenvalues lie inside the unit circle in the z-plane. This is an obvious necessary condition for stability, and every simulation that was run veried this condition. But we require a stricter condition than this because we desire no oscillations against the virtual xture. Using Equation 43, we convert the discrete nonzero eigenvalues to their equivalent s-plane locations, and then use Equation 44 to nd the damping ratio of the eigenvalues. Unlike the two-port method, the state-space method requires knowing Zh and Ze , not just upper bounds on their possible values. We again consider the unstable system from Section 5.1, but we now use the algorithm from [2] to compute the discrete state-space model. For this system, four of the eigenvalues are identically equal to zero (in the z-plane), with the other six eigenvalues having computed values of 1 = 2 = 0, 3,4 = 0.9863 0.2090i, and 5,6 = 0.9909 0.0841i. 3,4 lie outside the unit-circle in the z-plane, so this method correctly predicts the system is unstable. We now consider a system similar to the one just discussed, but with Kpm = Kps = 2000, Kvm = Kvs = 10, and KsV F = 2000. By reducing the position gains and increasing the velocity-feedback gains, we intend to stabilize the system. The discrete statespace model of this system has eigenvalues of 1 = 0.0093, 2 = 0.0101, 3,4 = 0.9820 0.1427i, and 5,6 = 0.9828 0.0557i. All of these eigenvalues lie inside the unit circle, but the damping ratio of 3,4 is only = 0.054. Figure 6 shows the simulation of this system. The signals are bounded, but there is some undesirable transient behavior that lasts longer than we would like. If we change system parameters with the goal of

0.14

0.12

0.1

0.08

0.06

continuous-contact model of Section 4. As we mentioned previously, boundedness will not be enough to ensure that the slave manipulator does not vibrate against the virtual xture. We have found that across many simulations, eigenvalues with 0.2 result in desirable behavior of the slave manipulator against the FRVF, but this condition can be tightened or relaxed as the designer sees t.

0.04

6
0.02 Xh X
e

Conclusions

0.1

0.2

0.3

0.4

0.5 Time

0.6

0.7

0.8

0.9

Figure 6: Simulation of system with = 0.054.

0.08

0.07

0.06

0.05

0.04

0.03

0.02

0.01 Xh X
e

0.1

0.2

0.3

0.4

0.5 Time

0.6

0.7

0.8

0.9

Figure 7: Simulation of system with = 0.257.

increasing the damping ratio of the most-oscillatory eigenvalues, we can design a system that will be well behaved when interacting with the virtual xture. Consider the same system discussed previously, but now increase the velocity-feedback gains to Kvm = Kvs = 50. We can even increase the FRVF stiness to KsV F = 5000. The predicted eigenvalues of this system are 1 = 0.0504, 2 = 0.0549, 3,4 = 0.9387 0.1688i, and 5,6 = 0.9420 0.0480i, which all lie inside the unit circle. The damping ratio of 3,4 has increased to = 0.257. Figure 7 shows the simulation of this system. It has the desired behavior of the slave manipulator touching the virtual xture with a constant force. We have found across many simulations that requiring all eigenvalues to have a damping ratio 0.05 can be used as a good rule-of-thumb for bounded stability. Eigenvalues with a damping ratio less than 0.05 sometimes correspond to instability in the Simulink simulation. This is most likely caused by violating the
2705

This work presents a tool to help predict and prevent unstable vibrations of the slave manipulator of a telemanipulation system against a FRVF through appropriate choice of control system gains and virtual xture stiness. The model used in Section 2 assumes the human, environment, master, and slave can be represented accurately by mass-spring-damper models. It also assumed perfect measurement of forces and positions. The validity of these assumptions are all open to debate, and should be kept in mind when applying the results of this research to a physical system. The equilibrium positions found by Equations 23-25 in Section 3 assume that Fh is large enough to move the slave manipulator at least to the depth needed to engage the FRVF. Those equations will not hold true if this condition is not satised. The purpose of this paper was to predict and prevent the slave from vibrating against its virtual xture, so that is the only case we considered. We also found in Section 3 that a telemanipulator with a FRVF located below the surface of an environment can be rewritten around an equilibrium position, turning the original system into one that can be analyzed using methods previously developed in the telemanipulation literature. In Section 4 we considered two methods of analyzing the stability of the system around an equilibrium point. The rst method used Llewellyns absolute stability criteria for two-port networks to determine if there is any passive human/environment combination that will make the system go unstable. The second method converted the system to its discrete state-space form, and then used the position of the eigenvalues to determine if the system was stable. In Section 5 we introduced a Simulink simulation that accurately simulates an idealized physical system. We found that using Llewellyns stability criteria did not give stability predictions that were consistent with simulations, but that the eigenvalues of the discrete state-space formulation could be used to evaluate contact stability of the virtual xture. We determined that a good rule-of-thumb for ensuring a wellbehaved virtual xture is that the most-oscillatory eigenvalues should have a damping ratio 0.2. The problem with using the discrete state-space form is that the computed eigenvalues only apply to

100 80 60 40 20 Im[s] 0 20 40 60 80 100 150 120

0.72

0.58

0.44 0.32 0.2 0.1

Acknowledgements
This material is based on work supported by the National Science Foundation, grant #ITR-0205318.

0.84

0.95

References
[1] J. J. Abbott and A. M. Okamura. Virtual Fixture Architectures for Telemanipulation. IEEE Intl. Conf. Robotics and Automation, 2003, In press. [2] J. J. Abbott and A. M. Okamura. Generating a StateSpace Formulation of a General Class of Telemanipulator. Haptic Exploration Lab Technical Report 03-1, The Johns Hopkins University, June 1, 2003. [3] R. J. Adams and B. Hannaford. Stable Haptic Interaction with Virtual Environments. IEEE Trans. Robotics and Automation, 15(3):465-474, 1999. [4] A. Bettini, S. Lang, A. Okamura, and G. Hager. Vision Assisted Control for Manipulation Using Virtual Fixtures. IEEE/RSJ Intl. Conf. Intelligent Robots and Systems, 1171-1176, 2001. [5] J. E. Colgate and J. M. Brown. Factors Aecting the Z-Width of a Haptic Display. IEEE Intl. Conf. Robotics and Automation, 3205-3210, 1994. [6] R. C. Dorf and R. H. Bishop. Modern Control Systems, chapter 13, p. 758. Addison Wesley, California, 1998. [7] R. B. Gillespie and M. R. Cutkosky. Stable UserSpecic Haptic Rendering of the Virtual Wall. ASME Dynamic Systems and Control, 58:397-406, 1996. [8] B. Hannaford. A Design Framework for Teleoperators with Kinesthetic Feedback. IEEE Trans. Robotics and Automation, 5(4):426-434, 1989. [9] K. Hashtrudi-Zaad and S. E. Salcudean. Analysis of Control Architectures for Teleoperation Systems with Impedance/Admittance Master and Slave Manipulators. Intl. J. Robotics Research, 20(6):419-445, 2001. [10] D. A. Lawrence. Stability and Transparency in Bilateral Teleoperation. IEEE Trans. Robotics and Automation, 9(5):624-637, 1993. [11] P. Marayong, A. Bettini, and A. Okamura. Eect of virtual xture compliance on human-machine cooperative manipulation. IEEE/RSJ Intl. Conf. Intelligent Robots and Systems, 1089-1095, 2002. [12] S. Park, R. D. Howe, and D. F. Torchiana. Virtual xtures for robotic cardiac surgery. Fourth Intl. Conf. on Medical Image Computing and Computer-Assisted Intervention, 1419-1420, 2001. [13] S. Payandeh and Z. Stanisic. On Application of Virtual Fixtures as an Aid for Telemanipulation and Training. Proc. 10th Symp. On Haptic Interfaces For Virtual Envir. and Teleoperator Systs., 18-23, 2002. [14] L. Rosenberg. Virtual Fixtures: Perceptual Tools for Telerobotic Manipulation. Proc. IEEE Virtual Reality International Symposium, 76-82, 1993. [15] Y. Yokokohji and T. Yoshikawa. Bilateral Control of Master-Slave Manipulators for Ideal Kinesthetic Coupling Formulation and Experiment. IEEE Trans. Robotics and Automation, 10(5):605-620, 1994.

100

80

60

40

20

0.95

0.84

0.72 100

0.58 Re[s]

0.44 0.32 0.2 0.1 50 0

Figure 8: Sensitivity of eigenvalues to changes in human and environment parameters (very fast eigenvalues not shown).

the human and environment explicitly considered. We would like the stability results to span all possible valid human and environmental impedances that the telemanipulator may encounter. This could possibly be done numerically, by varying the human and environment mass, damping, and stiness parameters and observing how the eigenvalues move. Figure 8 shows how the eigenvalues of the nal system discussed in Section 5.2 will vary for 4096 human/environment combinations spanning the valid impedances that were used in Section 5.1. This plot indicates that it may be possible to use the discrete state-space model robustly by incorporating this type of analysis of parameter sensitivity. In the future, we will investigate how to incorporate the discrete state-space model into a more robust method for determining FRVF stability. In the future, we will continue this work with the experimental setup introduced in [1], and also with a telemanipulator where the master and/or slave is of the admittance type [9]. We will also revisit the two-port method, because Llewellyns absolute stability criteria accounts for the robustness issues mentioned above. We will see if dierent modeling assumptions lead to better predictive power of the method. A method using passivity for robust stability analysis in conjuction with an eigenvalue method for design of transient behavior may provide the best overall system.
2706

You might also like