Design and Evaluation of a Teleoperated Robotic System for ERCP Cannulation

ABSTRACT

Background

Endoscopic retrograde cholangiopancreatography (ERCP) cannulation requires precise manipulation under fluoroscopic guidance. This study proposes a teleoperated robotic system for ERCP cannulation and evaluates its simulator‐based performance.

Methods

A teleoperated robotic system incorporating a Chebyshev‐based mechanism and a ring‐type master interface was developed. Manipulation performance was assessed using Fitts' law‐based insertion experiments, catheter‐tip trajectory analysis, and a simulator‐based cannulation task.

Results

Robotic manipulation showed a Fitts' law relationship comparable to conventional manual operation. Trajectory analysis showed higher path efficiency and lower jerk in the robotic condition, indicating improved trajectory‐related motion quality. In the simulator‐based cannulation task, the robotic condition showed slightly longer initial task completion times, whereas both robotic and conventional conditions showed comparable changes in task completion time across repeated trials.

Conclusions

This study demonstrates the feasibility of a teleoperated robotic system for simulator‐based ERCP cannulation tasks. Further validation with expert endoscopists and more realistic ERCP environments is required.

1. Introduction

Endoscopic retrograde cholangiopancreatography (ERCP) is an endoscopic procedure that utilises both endoscopy and fluoroscopy to diagnose and treat biliary and pancreatic diseases. While ERCP was initially developed as a diagnostic modality, it is now predominantly used for therapeutic interventions [1]. Because fluoroscopy is continuously required during ERCP, medical staff are inevitably exposed to ionizing radiation. To mitigate this risk, operators typically wear lead aprons. Although these garments effectively reduce radiation exposure to the body, they provide limited protection to the eyes and hands, which remain vulnerable to scattered radiation [2]. In addition, the use of leaded eyewear is often inconsistent in clinical practice [3]. Prolonged use of protective equipment also imposes significant physical strain on the musculoskeletal system, potentially leading to discomfort and injury [4]. These clinical challenges highlight the need for teleoperated robotic systems that can physically separate operators from the radiation environment.

In parallel, various robotic approaches have been proposed to improve endoscope manipulability and procedural efficiency. Examples include wearable hands‐free interfaces [5], foot‐controlled systems [6], robotised conventional endoscopes such as easyEndo [7], and multi‐degree‐of‐freedom platforms for NOTES‐ESD such as STRAS [8]. These studies primarily aim to enhance technical performance and reduce operator workload, rather than directly addressing radiation exposure or enabling remote operations. Nevertheless, they provide important technological foundations for the development of teleoperated ERCP systems. More recently, several robotic systems specifically targeting ERCP have been reported. These include motorized control units for conventional duodenoscopes [9], robot‐assisted systems designed for ERCP in MRI environments [10], and studies demonstrating the feasibility of robot‐assisted ERCP in animal models [11] and preliminary clinical settings [12]. While these studies highlight the feasibility of robotic ERCP, systematic evaluations of manipulability and usability from the operator's perspective remain limited. ERCP is characterised by a complex, multi‐degree‐of‐freedom manipulation environment that requires simultaneous control of the endoscope and surgical tools such as catheters and guidewires. However, how these manipulation characteristics are affected when mediated through robotic systems has not been sufficiently investigated.

This study focuses on the cannulation phase, which is a fundamental step in ERCP. Cannulation must be successfully performed before any therapeutic intervention can proceed and represents the stage at which fluoroscopy is first introduced. The process involves repeated insertion–withdrawal and directional adjustments while approaching the papilla [13]. As procedural difficulty increases, fluoroscopy time may also increase, linking operator skill with radiation exposure. Accordingly, cannulation represents a critical stage where technical complexity and radiation risk intersect. Furthermore, the success rate of selective biliary cannulation is widely used as a key indicator of ERCP competency in training guidelines [14].

To address these gaps, we propose a novel teleoperated robotic system for ERCP. The system consists of a slave robot equipped with a Chebyshev lambda mechanism‐based insertion module for stable control of ERCP tools [15], and a master robot featuring a ring‐type interface that enables intuitive insertion‐withdrawal control [16]. Rather than immediately targeting full radiation isolation, this system is designed as a preliminary platform to evaluate the feasibility of teleoperation and the technical capability for precise cannulation‐related tool manipulation. Although the long‐term goal is to reduce radiation exposure through remote procedures, the present study focuses on establishing the fundamental feasibility of precise cannulation under teleoperation. Direct validation under fluoroscopic conditions is beyond the scope of this study and remains for future investigation.

Instead of replicating the entire ERCP procedure, this study focuses on cannulation as a representative technical task for evaluating robotic manipulation performance. To comprehensively assess the proposed system, two complementary experiments were conducted. First, a Fitts' law‐based insertion task was performed to quantitatively analyse basic manipulation characteristics. Second, a simulator‐based cannulation task was conducted to compare performance and changes in task completion time across repeated trials between robotic and conventional approaches. Through this stepwise evaluation, we aim to assess both the fundamental manipulability and the practical feasibility of the proposed teleoperated ERCP system.

2. Materials and Methods

Ethics statement: This study was conducted as a non‐clinical, simulator‐based engineering validation of a teleoperated robotic system for ERCP cannulation. The purpose of the experiment was to evaluate the robotic system rather than to assess clinical outcomes or participant health‐related characteristics. No patients, clinical procedures, invasive interventions, biological specimens, identifiable personal information, or sensitive health information were involved. Participants served only as operators of the robotic system in the simulated task environment. Only anonymised objective task‐performance data were analysed, and all participants provided informed consent before participation.

2.1. DOF Analysis of the Duodenoscope

The duodenoscope used in ERCP is a mechanically complex system with multiple degrees of freedom (DOFs). In general, it includes bending control (up/down and left/right), axial rotation, insertion, and surgical tool manipulation.

In this study, the DOFs of the duodenoscope were functionally analysed and categorised into seven primary DOFs (Figure 1). DOFs 1‐4 correspond to endoscope positioning and orientation, where DOFs 1 and 2 represent insertion and rotation, and DOFs 3 and 4 control the distal tip bending. DOF 5 corresponds to the elevator for tool angulation, while DOFs 6 and 7 represent the insertion and rotation of surgical tools (catheter and guidewire).

FIGURE 1.

DOFs of duodenoscope.

Cannulation in ERCP requires precise insertion of the catheter and guidewire through the narrow papilla. This process demands coordinated control of endoscope positioning and tool advancement. Therefore, the functional integration of endoscope control and tool manipulation is a key consideration in this study.

2.2. Overall System Configuration

The proposed system adopts a master–slave configuration (Figure 2). The slave robot directly actuates the duodenoscope and surgical tools on the patient side, while the master device provides operator input through a remote interface. The slave system is designed to accommodate a conventional duodenoscope without structural modification, allowing compatibility with existing clinical devices. Endoscope control and surgical tool manipulation are functionally separated to ensure stable and independent actuation. The master interface incorporates a ring‐type controller that mimics natural insertion–withdrawal motion, enabling intuitive control of catheter and guidewire advancement. A joystick is used for endoscope manipulation, providing multi‐DOF control within a compact interface. The system was implemented using an STM32‐based controller and Dynamixel actuators (XM430‐W210 Robotis Co., Seoul, South Korea) for master–slave communication. The control loop operated at approximately 47 Hz, which was sufficient to ensure responsive teleoperation in the local experimental setup.

FIGURE 2.

Overall system.

2.2.1. Endoscope Control Module

The endoscope control module consists of two components: the endoscope manipulation unit, which fixes and controls the endoscope knobs, and the endoscope insertion unit, which grips and advances the endoscope tube.

The endoscope manipulation unit (Figure 3) includes an endoscope support for fixing the position of the endoscope, a knob drive mechanism, an elevator drive, a support roller with a dual‐locking mechanism for stable fixation during procedures, and a roll motor for rotational control. The knob drive is designed to match the geometry of the endoscope knobs and enables independent control of the coaxial knobs. The elevator mechanism is actuated using a parallelogram linkage, allowing synchronized control between the motor input and elevator angle.

FIGURE 3.

Endoscope manipulation unit, (a) overall structure, (b) top view of drives, (c) exploded view of knob drive assembly.

The endoscope insertion unit employs a modified Chebyshev lambda mechanism combined with a parallelogram linkage (Figure 4a). The Chebyshev lambda mechanism satisfies the Grashof condition and can therefore operate as a crank–rocker mechanism. Because the sum of the shortest and longest links is smaller than the sum of the remaining two links, the input crank $r_1$ can complete continuous rotation while the output link undergoes oscillatory motion. By introducing additional links to the conventional Chebyshev configuration, parallel motion is achieved. Two mirrored mechanisms form a gripper structure capable of securely holding the endoscope (Figure 4b). The gripper spacing is controlled via a ball screw mechanism, allowing installation even when the endoscope is already inserted into the body. The mechanism, inspired by Chebyshev's plantigrade machine [17], adopts a mirror‐symmetric linkage configuration that generates alternating motion, thereby enabling stable gripping and continuous insertion using a single motor. Rollers aligned with the endoscope axis ensure smooth transmission of rotational motion (Figure 5). Previous studies have reported that roller‐based insertion mechanisms may induce uneven contact forces, potentially leading to deformation of the endoscope [18]. In contrast, the proposed gripper employs multiple rollers aligned with the endoscope axis, increasing the contact area and distributing the applied forces. This configuration may help reduce localised loading and mitigate deformation during insertion.

FIGURE 4.

Chebyshev with parallelogram mechanism, (a) mechanism design, (b) base design of gripper.

FIGURE 5.

Endoscope insertion unit, (a) overall structure, (b) front view, (c) base mechanism.

2.2.2. Surgical Tool Control Module

The surgical tool control module is mounted on the endoscope support and consists of a catheter control unit and a guidewire control unit (Figure 6). Both units are based on the Chebyshev lambda mechanism. The catheter control unit provides one DOF for linear insertion and withdrawal, utilising a Chebyshev‐based gripper with an extended contact surface for stable handling (Figure 6a). The guidewire control unit enables two DOFs: linear motion and rotation. Linear motion is achieved through coordinated external gripper actuation with the Chebyshev lambda mechanism, while rotational motion is transmitted via a pulley system connected to the internal gripper (Figure 6b), as demonstrated in previous work [15]. Due to spatial constraints, the front and rear grippers are not mechanically linked but are instead synchronized through control with a 180° phase offset. This configuration allows simultaneous opening of the grippers for tool insertion and stable manipulation during operation.

FIGURE 6.

Surgical tool control module, (a) catheter control unit, (b) guidewire control unit.

2.2.3. Kinematic and Quasi‐Static Analysis of the Chebyshev‐Based Driving Mechanism

The driving mechanism used in the proposed system is based on a Chebyshev‐type crank–rocker four‐bar linkage, in which the rotation of the input crank ($r_1)$ is converted into an approximate straight‐line motion of the coupler point. This characteristic was used to generate the insertion and withdrawal motion required for manipulating the endoscope, catheter, and guidewire. In the endoscope‐driving module, the endoscope is positioned between the upper and lower Chebyshev‐based driving structures, and the driving motion is transmitted to the endoscope through static friction at the contact interfaces. Because the endoscope has the largest mass and is expected to impose the highest mechanical load among the manipulated tools, the endoscope‐driving module was selected as the representative worst‐case module for the kinematic and quasi‐static torque analysis.

For the kinematic analysis, the output position ($P$ = ($x,y)$) of the Chebyshev mechanism was formulated as a function of the input crank angle $\left(\theta \right)$. The auxiliary geometric parameters used to describe the linkage configuration are defined in Equation (1), and the corresponding position of the output point is expressed in vector form in Equation (2). In this formulation, the extension ratio of the output point along the coupler direction is represented by (k). Because the implemented mechanism was designed with (L = $r_2$), the extension ratio became (k = 2). This position formulation was used to calculate the output trajectory of the driving point and served as the basis for the subsequent velocity and quasi‐static torque analysis.

$$\begin{array}{c}\gamma \left(\theta \right)=\sqrt{{{4r}_2^2–r}_1^2–\left(d^2{–2\mathit{\text{dr}}}_1\cos \theta \right)},\phi \left(\theta \right)={\tan }^{-1}\left(\frac{r_1+\sin \theta }{r_1\cos \theta –d}\right)\end{array}$$ (1)

$$\begin{array}{c}P\left(\theta \right)=\left[\begin{array}{l}x\left(\theta \right)\\ y\left(\theta \right)\end{array}\right]=\left[\begin{array}{l}2r_1\cos \theta +d\\ 2r_1\sin \theta \end{array}\right]+k\gamma \left(\theta \right)\left[\begin{array}{l}–\sin \phi \left(\theta \right)\\ \cos \phi \left(\theta \right)\end{array}\right],k=\frac{r_2+L}{r_2}\end{array}$$ (2)

Since the Chebyshev‐based mechanism was used to generate insertion and withdrawal motions, the x‐direction displacement was considered the primary output of interest. Therefore, the velocity analysis focused on the x‐direction component of the output point rather than the full two‐dimensional velocity vector. The x‐direction velocity transmission ratio was obtained by differentiating the x‐position of the output point with respect to the input crank angle, as expressed in Equation (3). The actual x‐direction output velocity was then calculated by multiplying this transmission ratio by the motor angular velocity, as shown in Equation (4).

$$\begin{array}{c}{J}_x\left(\theta \right)=\frac{d{x}_H\left(\theta \right)}{d\theta }\end{array}$$ (3)

$$\begin{array}{c}{v}_x\left(\theta \right)=J_x\left(\theta \right)\dot{\theta }\end{array}$$ (4)

Since ERCP‐specific endoscope insertion speeds have not been well standardized, previously reported endoscope motion‐tracking and phantom‐validation studies were used to define a representative reference speed range for the endoscope‐driving module. In a real‐time endoscope motion‐tracking study, slow insertion/removal translations were evaluated within approximately ± 10 mm/s, whereas faster insertion and withdrawal motions were tested up to approximately 40 mm/s [19]. In addition, a colonoscopy tracking phantom validation study used insertion speeds of 10, 15, and 20 mm/s to simulate colonoscope motion during procedures [20]. Based on these reports, 40 mm/s was selected in the present analysis as a conservative upper‐bound reference speed for endoscope insertion/removal motion, rather than as an ERCP‐specific clinical standard.

The speed capability of the proposed Chebyshev‐based driving mechanism was then evaluated using the implemented linkage geometry of the robot. In this design, the crank length ($r_1$) was 10 mm, and the output trajectory was calculated using the corresponding geometric parameters of the fabricated mechanism. The calculated trajectory of the output point (P) is shown in Figure 7. Although the crank can rotate over a full cycle, only the effective near‐linear region of the trajectory was used to generate the insertion and withdrawal motion. This region corresponds to approximately 180° of input crank rotation and is highlighted in Figure 7a. The x‐direction velocity transmission ratio, $\left(\left|\frac{dx}{d\theta }\right|\right)$, was then calculated over the same effective operating region, as shown in Figure 7b. Based on this geometry, the minimum x‐direction velocity transmission ratio in the effective near‐linear operating region was approximately 10.0 mm/rad. Therefore, achieving a conservative reference speed of 40 mm/s requires an input motor speed of approximately 4.0 rad/s, corresponding to 38.2 rpm. This required speed is lower than the no‐load speed of the selected Dynamixel XM430‐W210 actuator, which is 77 rpm at 12 V. Thus, even the conservative upper‐bound reference speed can be generated within the speed capability of the selected actuator.

FIGURE 7.

Kinematic analysis of the Chebyshev lambda linkage, (a) Calculated trajectory of the output point (P) with ($r_1=10$) mm, (b) X‐direction velocity transmission ratio in effective near‐linear region.

This reference speed was also used to support the quasi‐static assumption for the subsequent torque analysis. Because the intended endoscope insertion and withdrawal motions are slow and controlled, the inertial force associated with the motion is expected to be small compared with the conservative gravitational load assumed for the 1 kg endoscope. For example, even if the endoscope reaches the 40 mm/s reference speed within 0.2 s, the corresponding acceleration is 0.2 m/s² and the inertial force for a 1 kg endoscope is 0.2 N, which is approximately 2% of the 9.81 N gravitational load. Therefore, inertial effects were considered secondary compared with the conservative load assumption, and a quasi‐static virtual‐work approach was used to estimate the required actuator torque.

For the quasi‐static force and torque analysis, the endoscope‐driving module was selected as the representative worst‐case module because the endoscope has the largest mass among the manipulated components. The endoscope mass was conservatively assumed to be 1.0 kg, corresponding to a gravitational load of 9.81 N. Because the endoscope‐driving module uses two actuators to generate the insertion and withdrawal motions, the load was assumed to be equally shared by the two actuators, resulting in a load of 4.91 N per actuator.

The required actuator torque was estimated to use a quasi‐static virtual‐work relationship between the input motor rotation and the x‐direction output motion. The endoscope was assumed to move together with the driving mechanism under a no‐slip contact condition. Because the mechanical efficiency of the prototype transmission and the contact interface were not directly measured, the torque requirement was first calculated under an ideal quasi‐static condition. A conservative safety factor of 2 was then applied to account for frictional losses, transmission losses, and possible contact‐related losses at the endoscope‐driving interface.

The ideal torque requirement was calculated as Equation (5).

$$\begin{array}{c}{\tau }_{\text{ideal}}=F_{\text{act}}\max \left|\frac{dx}{d\theta }\right|\end{array}$$ (5)

where ($F_{\text{act}}$) is the load per actuator and $\left(\left|\frac{dx}{d\theta }\right|\right)$ is the maximum x‐direction velocity transmission ratio in the effective near‐linear operating region. For the implemented robot geometry with ($r_1=10$) mm, the maximum value of $\left(\left|\frac{dx}{d\theta }\right|\right)$ in this region was approximately 13.36 mm/rad. Therefore, the ideal torque per actuator was estimated as Equation (6).

$$\begin{array}{c}{\tau }_{\text{design}}=\text{SF}\mathit{·}{\tau }_{\text{ideal}}\end{array}$$ (6)

The design torque was then obtained by applying a safety factor of 2 as Equation (7).

$$\begin{array}{c}{\tau }_{\text{design}}=2\times 4.905\times 0.01336=0.131N\mathit{·}m\end{array}$$ (7)

This estimated design torque is substantially lower than the stall torque of the selected Dynamixel XM430‐W210 actuator, which is 3.0 Nm at 12 V. Although the stall torque represents a momentary maximum torque rather than a continuous operating torque, the estimated quasi‐static design torque indicates that the selected actuator provides sufficient torque capacity for the simulator‐based endoscope‐driving task.

For additional conservatism, the torque requirement was also estimated using the maximum x‐direction transmission ratio over the full crank cycle. In this case, $\left(\right|\frac{dx}{d\theta }\left|\right)$ was approximately 48.99 mm/rad, resulting in an ideal torque of 0.240 N m and a design torque of 0.481 N m after applying the safety factor of 2. Even under this full‐cycle worst‐case condition, the estimated torque remained below the rated stall torque of the selected actuator.

2.2.4. Master Interface and Coordinated Operation

The master interface consists of three main components: a user interface (UI) for visual feedback, a joystick for endoscope control, and a ring‐type controller for surgical tool manipulation, which has been reported to improve usability and reduce cognitive load in teleoperation (Figure 8) [16]. The UI provides real‐time visualisation of the endoscopic image and the state of the control knobs. The joystick enables insertion and withdrawal via buttons, controls knob actuation along the x‐ and y‐axes, and adjusts endoscope rotation through z‐axis tilting. An encoder integrated into the handle allows control of the elevator mechanism. The ring‐type controller consists of two crossed rings and a mode‐switching button. The larger ring controls insertion and withdrawal of the surgical tool, while the smaller ring controls rotational motion. The switch enables selection between catheter and guidewire control, allowing intuitive and coordinated operation.

FIGURE 8.

Master interface (a) UI, (b) endoscope controller, (c) surgical tool controller.

2.2.5. Control System Implementation

The control architecture of the proposed system is summarised in Figure 9. The system was implemented as an operator‐in‐the‐loop master–slave position‐command system. The STM32‐based controller served as the high‐level command interface between the master input devices and the slave‐side actuators. Operator inputs from the joystick and ring‐type controller were sampled by the STM32 controller and converted into corresponding actuator commands for each degree of freedom. These commands were transmitted to the Dynamixel XM430‐W210 actuators, which performed low‐level actuator control internally. Therefore, the STM32 controller did not directly control the motor current or implement a separate low‐level feedback controller; instead, it generated high‐level actuator commands for teleoperated manipulation.

FIGURE 9.

Control architecture of the proposed teleoperated robotic system.

The command update rate of the proposed control system was approximately 47 Hz, corresponding to an update interval of approximately 21 ms. This update rate was considered sufficient for the present simulator‐based cannulation task because the intended insertion and withdrawal motions were slow and controlled. Using the conservative reference insertion speed of 40 mm/s, the corresponding displacement per command update is approximately 0.85 mm. At lower insertion speeds of 10–20 mm/s, the displacement per update is approximately 0.21–0.43 mm. These command increments are small relative to the scale of the simulator‐based manipulation task and allow responsive operation in the local experimental setup.

The proposed system was operated as an operator‐in‐the‐loop teleoperation system rather than an autonomous closed‐loop control system. Therefore, backlash, actuator deadband, and friction were not explicitly compensated by an external feedback controller. The STM32 controller generated high‐level actuator commands, while the Dynamixel actuators performed actuator‐level position regulation internally. According to the manufacturer's specification, the Dynamixel XM430‐W210 actuator has a backlash of 15 arcmin, corresponding to 0.25°. Based on the maximum x‐direction transmission ratio of the Chebyshev‐based mechanism in the effective operating region, this actuator backlash corresponds to an estimated x‐direction displacement uncertainty of approximately 0.058 mm. This value is small relative to the scale of the simulator‐based task. In addition, small deviations caused by unmodeled effects such as backlash, deadband, and friction could be accommodated by the operator through visual monitoring and input adjustment during teleoperation. However, quantitative characterisation of backlash, deadband, friction, slip, and external position tracking was not performed in this study and will be addressed in future work.

3. Results

To evaluate the manipulation performance of the proposed robotic system compared with conventional manual operation, two complementary experiments were conducted. The first experiment was a Fitts' law‐based insertion task designed to quantitatively assess fundamental motor control performance under controlled conditions. The second experiment was a simulator‐based cannulation task designed to evaluate task completion and trajectory‐related performance in a procedure‐inspired scenario. The overall experimental setup is shown in Figure 10. A commercially available duodenoscope (JF‐240, Olympus Corporation, Tokyo, Japan), catheter (PR‐V214Q, Olympus Corporation, Tokyo, Japan), and guidewire (Visiglide2, Olympus Corporation, Tokyo, Japan) were used.

FIGURE 10.

Experimental setup of the proposed robotic system with a JF‐240 duodenoscope.

All experiments were conducted with eight volunteer novice operators aged 26–32 years. None had prior endoscopic or ERCP experience, and all were right‐handed. All operators were able to manipulate the master interface without assistance. Each operator was given approximately 5 minutes of practice before the experiments, and practice data were excluded from analysis. A within‐subject design was adopted, where all participants performed both robotic and conventional conditions.

3.1. Fitts' Law‐Based Performance Assessment

3.1.1. Experimental Design

The Fitts' law‐based insertion task was designed to quantitatively evaluate the fundamental manipulation performance of the proposed system. Classical Fitts' law models movement time (MT) as a linear function of the index of difficulty (ID) and has been widely used to compare speed‐accuracy trade‐offs across input systems [21]. Unlike conventional pointing tasks, ERCP cannulation involves inserting a catheter through a narrow opening rather than simply contacting a target. Therefore, a modified Fitts' task was designed to reflect insertion‐based interaction (Figure 11).

FIGURE 11.

(a) Fitts' law test bed, (b) test bed in endoscope camera, (c) test bed with aurora tracking system.

The index of difficulty was defined as

$$\text{ID}={\log }_2\frac{2A}{W}$$

where $A$ represents the distance to the target and $W$ represents the effective target width. In this study, $W$ was defined as the clearance between the hole diameter $\left(D_{\text{hole}}\right)$ and catheter diameter $\left(D_{\text{catheter}}\right)$:

$$W=D_{\text{hole}}-D_{\text{catheter}}$$

This definition reflects the requirement that the catheter must pass through the opening rather than simply touch it.

Nine task conditions with varying levels of difficulty were defined by combinations of $A$ and $W$ (Table 1), with a catheter diameter of 2 mm. Each condition was repeated an equal number of times under both robotic and conventional manipulation. In addition to movement time, catheter tip trajectories were recorded using an Aurora electromagnetic tracking system (Aurora system, Northern Digital Inc., Waterloo, ON, Canada) for selected trials due to hardware constraints.

TABLE 1.

Fitts law task conditions.

Condition	$D_{\text{hole}}$ (mm)	W (mm)	A (mm)	ID
1	5.1	3.1	10	2.69
2	5.1	3.1	20	3.69
3	5.1	3.1	30	4.27
4	3.4	1.4	10	3.84
5	3.4	1.4	20	4.84
6	3.4	1.4	30	5.42
7	2.5	0.5	10	5.32
8	2.5	0.5	20	6.32
9	2.5	0.5	30	6.91

3.1.2. Procedure

Participants performed an insertion task in which the catheter was advanced from a starting position and passed through a target opening. Each trial was defined as the movement from one opening to successfully passing through another. Each participant completed both robotic and conventional conditions, with 10 repetitions per condition for each difficulty level. The order of conditions was counterbalanced across participants to minimise order effects. Movement time (MT) was recorded for all trials. For selected trials, trajectory data were collected to analyse path characteristics, including path length, path efficiency, and motion smoothness.

Motion smoothness was evaluated using a sample‐based jerk metric derived from the catheter‐tip trajectory. This approach was used to compare relative trajectory smoothness between conditions while avoiding errors caused by occasional frame‐index discontinuities during preprocessing. Mean jerk and maximum jerk were calculated from the magnitude of the sample‐based jerk sequence and were interpreted as relative smoothness metrics rather than absolute physical jerk values.

3.1.3. Results

Both robotic and conventional manipulation exhibited a linear relationship between movement time (MT) and index of difficulty (ID) (Figure 12), consistent with Fitts' law.

FIGURE 12.

Fitts law test results.

To enable direct comparison of speed–accuracy characteristics, the slope of the regression was used as the primary metric. The slope was 1.40 for the robotic condition and 1.76 for the conventional condition, indicating comparable performance between the two methods.

Trajectory analysis results are summarised in Table 2, with representative examples shown in Figure 13. Path length was shorter in the robotic condition (52.61 ± 4.96 mm) than in the conventional condition (65.20 ± 16.27 mm), although the difference was not statistically significant (p = 0.067). Path efficiency was significantly higher in the robotic condition (0.63 ± 0.063) compared to the conventional condition (0.41 ± 0.10) (p = 0.001). Smoothness analysis based on jerk demonstrated significantly improved motion quality in the robotic condition. Mean jerk was lower in the robotic condition (0.14 ± 0.04) than in the conventional condition (0.61 ± 0.11) (p < 0.001), and maximum jerk was also reduced (1.32 ± 0.43 vs. 3.07 ± 0.47, p < 0.001).

TABLE 2.

Comparison of trajectory metrics between robotic and conventional manipulation.

Metric	Robot (Mean ± SD)	Conventional (Mean ± SD)	p‐value
Path length (mm)	52.61 ± 4.96	65.20 ± 16.27	0.067
Path efficiency	0.63 ± 0.063	0.41 ± 0.10	0.001
Mean jerk	0.14 ± 0.042	0.61 ± 0.11	< 0.001
Max jerk	1.32 ± 0.43	3.07 ± 0.47	< 0.001

FIGURE 13.

Example trajectories of the catheter tip during the Fitts‐based insertion task.

3.2. Simulator‐Based Cannulation Task

3.2.1. Experimental Design

A simulator‐based cannulation experiment was conducted to evaluate the applicability of the proposed system in a task reflecting a procedure‐inspired workflow. To ensure objectivity and reproducibility, the experiment utilised a standardized, commercially available ERCP training phantom. The phantom provides a simplified representation of the descending duodenum and major duodenal papilla and preserves the basic geometric constraints required for cannulation (Figure 14). Although the phantom does not fully reproduce the anatomical and procedural complexity of clinical ERCP, it provides a controlled and repeatable environment for comparing robotic and conventional manipulation. Because all participants were novices, a simplified cannulation task was designed rather than a full selective biliary cannulation procedure. The task was structured into three key phases: (1) approaching the papilla with the catheter, (2) inserting the catheter into the orifice, and (3) advancing the guidewire to a predefined target location within the biliary duct. This setup captures the core manipulation steps of ERCP cannulation while maintaining feasibility for novice participants (Figure 15).

FIGURE 14.

ERCP simulator.

FIGURE 15.

Procedure of simulator experiment, (a) starting position, (b) approach to papilla, (c) insertion.

3.2.2. Procedure

Participants performed both robotic and conventional manipulation conditions, with the order counterbalanced across participants. Each participant performed 10 trials per condition. Task completion time was measured from the start of endoscope manipulation to successful guidewire placement at the target location. Changes in task completion time across repeated trials were analysed based on performance changes across repeated trials.

3.2.3. Results

The average task completion time was 32.2 s for conventional manipulation and 37.2 s for robotic manipulation. Both conditions demonstrated decreasing task completion times with repeated trials (Figure 16), indicating a trial‐order effect. To quantify behaviour, linear regression slopes were analysed. The slopes were −2.88 for the robotic condition and −2.95 for the conventional condition, indicating comparable changes in task completion time across repeated trials.

FIGURE 16.

Trial‐order effect of task completion time for robotic and conventional manipulations. Linear trend lines are shown for visualisation of performance improvement across trials.

4. Discussion

In this study, we evaluated the performance of a teleoperated robotic system for ERCP cannulation through Fitts' law‐based experiments, trajectory analysis, and a simulator‐based task. The results indicate that the proposed system achieves manipulation performance comparable to conventional manual operation, while improving trajectory‐related motion quality. The use of novice operators allowed evaluation of baseline system operation under controlled conditions without the influence of prior ERCP experience.

The Fitts' law analysis demonstrated a linear relationship between movement time and index of difficulty for both robotic and conventional manipulation, indicating that the fundamental speed‐accuracy characteristics were preserved under teleoperation. The comparable regression slopes further suggest that the proposed system maintains a level of control performance comparable to manual operation, despite the additional mediation introduced by the robotic interface.

Trajectory analysis revealed that robotic manipulation achieved higher path efficiency and lower jerk compared with conventional manipulation. The lower sample‐based jerk values observed in the robotic condition suggest that the catheter‐tip trajectory involved less abrupt sample‐to‐sample changes during insertion. In the context of cannulation‐related manipulation, smoother catheter advancement may be beneficial because excessive abrupt motion can make fine positioning near a confined target more difficult. However, the jerk metric used in this study was a sample‐based smoothness measure derived from selected Aurora trajectory trials and should be interpreted as a relative comparison between robotic and conventional manipulation rather than as an absolute dynamic quantity. These findings indicate that the robotic system enables more direct and smoother movements, reducing unnecessary motion during catheter advancement. Such motion characteristics may be advantageous in ERCP cannulation, where precise and stable manipulation within a confined anatomical structure is required.

In the simulator‐based experiment, robotic manipulation showed slightly longer initial task completion times; however, both conditions exhibited comparable changes in task completion time across repeated trials, as reflected by similar regression slopes. This suggests that the task‐completion trend across repeated trials was comparable between the robotic and conventional conditions, despite the initial performance gap.

This study has several limitations.

First, all operators were novices, and therefore, the results primarily reflect baseline operation of the proposed system under controlled simulator conditions rather than expert‐level performance. This choice was appropriate for the preliminary engineering objective of the present study because novice operators performed the task according to a standardized instruction protocol, with less influence from individualised expert strategies or prior procedural habits. Therefore, the obtained data provide baseline information on fundamental manipulation patterns, trajectory variability, inefficient motion, and performance changes across repeated trials. However, novice‐operator data cannot replace expert clinical data. Future studies should include expert endoscopists to define clinically meaningful performance benchmarks and to determine whether robotic teleoperation can reduce the novice–expert performance gap.

Second, the experiments were conducted in a simplified simulator environment, which does not fully replicate the anatomical and procedural complexity of in vivo ERCP cannulation. The simulator task was intentionally simplified to provide a controlled and repeatable setting in which the fundamental robotic motions required for cannulation, including insertion, withdrawal, rotation, and catheter/guidewire manipulation, could be observed and evaluated. Therefore, the present results should be interpreted as preliminary, non‐clinical engineering validation rather than clinical validation. Future work should include validation using more realistic ERCP phantoms, fluoroscopy‐compatible environments, and expert endoscopist evaluation under an approved study protocol.

Third, the present experiments were conducted in a local setup, and therefore the 47 Hz update rate should not be directly interpreted as the end‐to‐end latency of a remote clinical teleoperation system. Nevertheless, the corresponding command update interval of approximately 21 ms is shorter than the latency values commonly discussed in the telesurgery literature, where communication delays below 100 ms have been considered preferable or acceptable for robotic teleoperation [22]. The reported latency of the transatlantic Lindbergh operation was approximately 155 ms [23]. However, future remote implementation of the proposed system will require quantitative evaluation of end‐to‐end latency, jitter, packet loss, and tracking error under realistic network conditions. Beyond these limitations, translation towards a clinical‐grade ERCP robotic system will require additional development steps, including refinement of the mechanical design for sterilisation, biocompatibility, fail‐safe operation, emergency release, and compatibility with the clinical ERCP environment. Quantitative safety and reliability testing, more realistic ERCP phantoms, fluoroscopy‐compatible experimental setups, and approved preclinical or clinical study protocols will also be necessary before considering clinical translation. Future integration with AI may further extend the proposed platform by enabling image‐based tool and papilla detection, motion‐quality assessment, abnormal‐motion detection, safety monitoring, and shared‐control assistance. The proposed simulator‐based platform may support this direction by enabling standardized collection of trajectory, operator‐command, image, and task‐performance data under controlled conditions. Future work should investigate the impact of communication delays on teleoperated performance, particularly in remote or networked settings. Despite these limitations, this study provides quantitative evidence that a teleoperated robotic system can maintain comparable manipulation performance by improving trajectory‐related motion quality and reducing unnecessary motion in ERCP cannulation tasks. These findings support teleoperated approaches as a viable option as a potential means to enhance operator safety and procedural stability in radiation‐exposed environments.

5. Conclusion

This study proposed a teleoperated robotic system for ERCP cannulation and evaluated its performance through Fitts' law‐based experiments, trajectory analysis, and a simulator‐based cannulation task. The results demonstrated that the proposed system achieved manipulation performance comparable to conventional manual operation while improving trajectory‐related motion quality, as evidenced by higher path efficiency and lower sample‐based jerk. Although slightly longer initial task completion times were observed in the robotic condition, both robotic and conventional conditions showed comparable changes in task completion time across repeated trials. These findings suggest that the proposed system may serve as a preliminary engineering platform for supporting precise cannulation‐related manipulation in a controlled simulator environment. However, the present study should be interpreted as simulator‐based engineering validation rather than clinical validation. Future work should include validation with expert endoscopists, comparison between novice and expert users, more realistic ERCP phantoms, fluoroscopy‐compatible environments, quantitative safety and reliability testing, and approved preclinical or clinical study protocols. In addition, the proposed platform may provide a foundation for future data‐driven and AI‐assisted ERCP robotic systems by enabling standardized collection of trajectory, operator‐command, image, and task‐performance data under controlled conditions.

Funding

This work was supported by the Technology development Program (RS‐2023‐00321839) funded by the Ministry of SMEs and Startups (MSS, Korea).

Ethics Statement

This study was conducted as a non‐clinical, simulator‐based engineering validation of a teleoperated robotic system for ERCP cannulation. The purpose of the experiment was to evaluate the robotic system rather than to assess clinical outcomes or participant health‐related characteristics. No patients, clinical procedures, invasive interventions, biological specimens, identifiable personal information, or sensitive health information were involved. Participants served only as operators of the robotic system in the simulated task environment. Only anonymised objective task‐performance data were analysed, and all participants provided informed consent before participation.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.