Caveats on the first-generation da Vinci Research Kit latent technical constraints and essential calibrations Zejian Cui12 Joao Cartucho1 Stamatia Giannarou1 and Ferdinando Rodriguez y Baena12

2025-04-30 0 0 1.89MB 15 页 10玖币

侵权投诉

Caveats on the ﬁrst-generation da Vinci Research Kit: latent technical

constraints and essential calibrations

Zejian Cui∗,1,2, Jo˜

ao Cartucho∗,1, Stamatia Giannarou1, and Ferdinando Rodriguez y Baena1,2

Abstract— Telesurgical robotic systems provide a well estab-

lished form of assistance in the operating theater, with evidence

of growing uptake in recent years. Until now, the da Vinci

surgical system (Intuitive Surgical Inc, Sunnyvale, California)

has been the most widely adopted robot of this kind, with more

than 6,700 systems in current clinical use worldwide [1]. To

accelerate research on robotic-assisted surgery, the retired ﬁrst-

generation da Vinci robots have been redeployed for research

use as “da Vinci Research Kits” (dVRKs), which have been

distributed to research institutions around the world to support

both training and research in the sector. In the past ten years,

a great amount of research on the dVRK has been carried out

across a vast range of research topics. During this extensive

and distributed process, common technical issues have been

identiﬁed that are buried deep within the dVRK research and

development architecture, and were found to be common among

dVRK user feedback, regardless of the breadth and disparity of

research directions identiﬁed. This paper gathers and analyzes

the most signiﬁcant of these, with a focus on the technical

constraints of the ﬁrst-generation dVRK, which both existing

and prospective users should be aware of before embarking

onto dVRK-related research. The hope is that this review will

aid users in identifying and addressing common limitations of

the systems promptly, thus helping to accelerate progress in the

ﬁeld.

I. INTRODUCTION

Robotic-Assisted Minimally Invasive Surgery (RMIS) has

gained signiﬁcant popularity in recent years due to its

advantage of causing less tissue trauma and reducing hospi-

talization time for patients. Of all surgical robotic platforms

available in the market, the da Vinci robot (Intuitive Surgical

Inc, Sunnyvale, California) has dominated RMIS, with more

than 10M operations having been performed on this platform

in the past twenty years [2]. To further propel RMIS research

on the da Vinci surgical platform, in 2012, an initiative was

launched to repurpose retired ﬁrst-generation da Vinci robots

by converting them into da Vinci Research Kits (dVRKs)

[3], which consist of both a software package and hardware

controllers. These dVRKs have been distributed to research

institutions around the world. Until now, more than 40

research groups across 10 countries have beneﬁted from

the initiative, forming a thriving dVRK research community,

with more than 250 peer-reviewed research papers on the

dVRK having been published.

∗These authors contributed equally to the work.

Zejian Cui: zejian.cui19@imperial.ac.uk

Jo˜

ao Cartucho: j.cartucho19@imperial.ac.uk

1The Hamlyn Centre for Robotic Surgery, Imperial College London,

London SW7 2AZ, UK

2Mechatronics in Medicine Lab, Department of Mechanical Engineering,

Imperial College London, London SW7 2AZ, UK

Fig. 1. The ﬁrst-generation dVRK with two surgeon-side manipulators

(SSMs), three patient-side manipulators (PSMs) and one endoscopic camera

manipulator (ECM)

Although research conducted on the dVRK branches into

different topics, almost all of these works deal with data

directly provided by the dVRK, containing kinematics, image

and system information, as described in [2]. These data are

arguably the best source of information available to users

who wish to keep abreast of the latest state of the robot as

they develop novel research on the dVRK. However, blind

reliance on the ﬁdelity of these data can be risky. The inac-

curacy of dVRK data partly arises from the intrinsic design

of the telesurgical platform, and partly from limitations of

the factory calibration of the hardware, which has only been

improved upon in recent years thanks to a growing interest in

the ﬁeld and the many contributions of researchers working

with the dVRK. Awareness of these seminal works should be

heightened within the dVRK community, while thus far, there

only exists one review paper about the dVRK [2], which

focuses on categorizing research in terms of relevant high-

level application areas.

To ﬁll in this gap, in this paper we aim to analyze the

technical constraints identiﬁed within the ﬁrst-generation

dVRK so that users can be aware of these shortcomings

and take corrective actions to overcome these, in pursuit of

better research outcomes. We have based this review paper

on existing literature and the feedback from researchers of

the dVRK community, as detailed in Sec. II.

According to our initial research, there is a general con-

sensus among dVRK users about the inaccuracies within the

dVRK kinematics and image data, which necessitates studies

on essential kinematics and camera calibration. Allowing for

arXiv:2210.13598v2 [cs.RO] 12 Jul 2023

the fact that hand-eye calibration is an indispensable part of

teleoperation, the accuracy of which can have a profound

impact on subsequent task implementations, we also stress

technical concerns about hand-eye calibration. Additionally,

we address potentiometer calibration after weighing up its

substantial impact on the performance of autonomous task

execution, even though, thus far, to the best of our knowl-

edge, there is no literature about this topic. In the follow-

ing sections, kinematics calibration, hand-eye calibration,

potentiometer calibration and camera calibration problems

are detailed in Sec.III, VI, IV and V, respectively. For each

of these, we ﬁrst identify the nature of the problem, then

distill prevailing schools of thought concerned on how to

tackle them, and ﬁnally we point out potential paths for better

addressing these problems, placed in the dVRK framework.

A discussion is presented in Sec.VII, where we summarize

the main ﬁndings, and other miscellaneous dVRK technical

issues uncovered during the survey.

II. SEARCH METHODOLOGY

A. Literature review

To identify the most signiﬁcant technical constraints that

have hampered research progress with the dVRK, we ﬁrst

looked into all papers on the dVRK across different re-

search application areas and gathered common issues that

researchers claimed to have affected task performance, taking

advantage of the latest review paper on the subject, which

encompasses papers on the dVRK from year 2014 to 2021

[2]. We also employed the keywords “dVRK” paired with

“error”, “calibration” and “task performance” individually on

Google Scholar, ScienceDirect and RefWorks and included

all relevant ﬁndings in this survey.

B. Consultation with researchers

1) Leading researchers: Having gathered a list of tech-

nical issues reported by dVRK users in the literature, we

consulted with leading researchers of the dVRK community

(Mr. Anton Deguet and Dr. Simon DiMaio) about whether

these issues have been discovered under the ﬁrst-generation

dVRK and thus can be regarded as universal.

2) Peer researchers: We designed a questionnaire fea-

turing selected technical issues, which we deemed to be

“universal”, and had it distributed within the dVRK com-

munity1. The design of the questionnaire and a summary of

user responses are detailed in Appendix I.

III. KINEMATICS CALIBRATION

A well-acknowledged technical issue within the dVRK

is its time-variant inaccurate forward kinematics, which

induces a poor surgical tool tip pose estimation, reﬂected

by a discrepancy in readings of up to 1.02mm [4], [5]

between the alleged tip pose streamed from the dVRK

and the ground truth observed by an external means. This

discrepancy, termed positioning error in this paper, has

exceeded the sub-millimetre positioning accuracy required by

1More info at: https://jhudvrk.slack.com

RMIS [6], having repercussions on diverse research projects

implemented on the dVRK. Note that kinematics errors

exist in all individual joints of the dVRK; however, it is

in the end effector position that the discrepancy caused

by the kinematics errors is most evident, because of the

accumulation of errors across all the joints.

A. Need for kinematics calibration

For research on surgical subtask automation, accurate

positioning is required, such as in debridement [7], suturing

[8], [9], needle extraction [10], and peg transfer [11]–[13],

where a surgical instrument is required to move to the target

grasping point. For research on the development of Active

Constraints (AC, also known as virtual ﬁxtures) via the

dVRK, positioning accuracy will affect the proximity query

result, and therefore the entire enforcement stage of the AC

implementation [14]–[17]. Even for research on image-based

hand-eye calibration [18], inaccurate surgical tool kinematic

data leads to an inaccurate 3D pose estimation and thereby

affects the accuracy of its back-projected 2D pixel position,

which serves as a criterion for evaluating different hand-eye

calibration strategies.

B. Challenges in accurate positioning

1) Cable-driven effects: The dVRK surgical platform,

which is a cable-driven robotic system [19], has an inherent

issue in obtaining accurate joint readings from the built-

in encoders. This is because all encoders are mounted

adjacent to actuators but distant from joints, where the actual

joint angles are estimated through transmission kinematics

[20]. However, unknown parameters such as pulley-cable

friction and nonlinearities [21] contribute to inaccuracies in

joint angle estimation, which result in dVRK joint encoder

readings not reﬂecting actual joint positions. These errors

in joint space subsequently affect the estimated end effector

tip pose through Denavit-Hartenberg (DH) parameters and

forward kinematics. The DH parameters here only concern

active joints.

2) Inaccurate kinematic parameters: The wear and tear of

the dVRK not only induces cable slack [22], which affects

the parameters in the transmission kinematics, resulting

inaccurate joint encoder readings, but also entails instrument

damage which leads to additional inaccuracies in the DH

parameters of the forward kinematics, causing the estimated

tool-tip pose to stray from the actual one. Another signiﬁcant

source of error is due to the potentiometers, as explained in

detail in IV.

3) Other non-kinematic error sources: Other than kine-

matic factors, non-kinematic factors such as external forces

exerted onto the tool shaft and backlash between the tool

shaft and cannula, illustrated in Fig. 2, would also con-

tribute to an overall erroneous end-tip pose estimation [23],

[24]. These external forces will affect the tension of cables

associated to tool tips, leading to an error in transmission

kinematics. In addition, they will also cause a displacement

of the tool shaft because of compliance, as illustrated in Fig.

3; this displacement is not detectable by the encoders [23],

and thus positioning error accumulates.

Fig. 2. As illustrated in red (d), there is a gap between the surgical

instrument’s shaft and the cannulae. Therefore, the point of contact between

this shaft and the cannulae can change. This phenomenon is known as

backlash. Backlash causes the tool to move without being reﬂected by the

encoder readings. Backlash is evident in the ﬁgure, illustrated in green,

where the “S” character of Intuitive’s logo has moved to a different location,

given that the camera is static.

Fig. 3. Compliance of the surgical instrument. Similarly to backlash, this

compliance is not reﬂected on encoder readings, and hence not reﬂected in

the joint position estimation.

C. Methods for kinematics calibration

In this subsection, we gather methods that are available in

the current literature to compensate for dVRK positioning er-

rors, and we classify these in term of the speciﬁc error source

they are aiming to address, namely encoder readings, kine-

matics parameters and other non-kinematic factors. Methods

within the former category aim to estimate the current joint

positions, whereas those within the latter category aim to

force the system to reach the desired position.

1) Compensation for encoder readings: For methods that

aim to compensate encoder reading inaccuracies, we further

divide these into two classes: encoder reading calibration and

tool-tip pose estimation. Both classes recognize the nature of

the erroneous encoder readings; however, the former focuses

on ﬁnding actual joint positions in joint space, while the

latter places an emphasis on ﬁnding the actual tool-tip pose.

a) Encoder reading calibration: Methods for calibrat-

ing encoder readings follow a standardized pipeline that

involves modeling, measurement and parameter ﬁtting [25].

To start with, a model parameterized by η, which maps

encoder readings obtained from transducers to actual joint

positions, is conceived. In the subsequent measurement stage,

robot joints are commanded to move to designated positions

while ground truth values of joint positions are gathered

via an external measurement system. Finally, different ﬁtting

models, which can be either linear or nonlinear, are adopted

to estimate the ηthat best ﬁts designated joint positions to

ground truth measurements.

Major differences between methods in this class are the

selection of calibration models and approaches to measuring

the ground truth. Following the calibration paradigm, Huang

et al. [24] started off using a linear model to map encoder

readings to ground truth values gathered by tracking infrared

markers mounted on the end effector with an optical camera

system, namely the NDI Polaris [26]. To better estimate and

account for the nonlinear nature of cable stretch and friction,

Hwang et al. adopted a data-driven approach to training deep

learning models that map collected readings to actual joint

positions gathered in a similar way [11].

With access to the forward kinematics of a system, which

maps encoder readings to actual joint positions, one can ﬁnd

an inverse calibration model that generates encoder com-

mands that drive a robot to reach desired joint positions. Such

methods lend themselves to implementing automatic surgical

tasks [7], [11], [13], where a surgical tool is requested

to accurately move to a given pose. However, there are

several drawbacks in model-based calibration methods. First,

as pointed out in [27], model-based calibration methods are

hard to implement outside of a lab setting because of the need

for additional sensors, calibration objects, and tedious data

collection procedures. Second, an invariant calibration model

fails to accurately reﬂect a dynamic mapping relationship as,

e.g., the cable wears and tears over time. Third, adopting

model-based approaches requires greater efforts to ﬁnd the

ground truth values of the joint positions for the 7-DoF

da Vinci robot. Although there exists numerical inverse

kinematics solvers for the SSMs and the PSMs, running

at 1.5 KHz 2, there is a lack of kinematics parameters

identiﬁcation procedures. Hence the ﬁdelity of the nominal

kinematics parameters for the disposable surgical instruments

relies on the quality of manufacture, which, however, can

be imprecise. Furthermore, the accuracy of the ground truth

measurement provided by an external sensor, a depth-sensing

camera for example, is also inevitably affected by the inaccu-

racy in camera parameters and the hand-eye transformation

matrix, as analyzed in V and VI, respectively.

b) Tool-tip pose estimation: In addition to generating

joint commands that drive a robot to reach the desired

state, there are times when the actual tool-tip pose in the

2One numerical solver is provided at https://github.com/jhu-cisst/cisst

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

Caveatsonthefirst-generationdaVinciResearchKit:latenttechnicalconstraintsandessentialcalibrationsZejianCui∗,1,2,Jo˜aoCartucho∗,1,StamatiaGiannarou1,andFerdinandoRodriguezyBaena1,2Abstract—Telesurgicalroboticsystemsprovideawellestab-lishedformofassistanceintheoperatingtheater,withevidenceofgrowingupt...

展开>> 收起<<

Caveats on the first-generation da Vinci Research Kit latent technical constraints and essential calibrations Zejian Cui12 Joao Cartucho1 Stamatia Giannarou1 and Ferdinando Rodriguez y Baena12.pdf

共15页,预览3页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Caveats on the first-generation da Vinci Research Kit latent technical constraints and essential calibrations Zejian Cui12 Joao Cartucho1 Stamatia Giannarou1 and Ferdinando Rodriguez y Baena12

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: