A Suspended-Configuration Endoscopic Robotic Platform with Dual-Module Actuation for Enhanced Gastrointestinal Interventions

Abstract

Robot-assisted gastrointestinal endoscopic surgery (RAGES) has emerged as a critical approach for minimally invasive procedures, including Endoscopic Submucosal Dissection, polypectomy, hemostasis, and therapeutic interventions. However, manual manipulation of flexible endoscopes remains technically demanding and limits surgical precision. To enhance surgical efficiency, this paper proposes a joystick-controlled Endoscopic Robotic System (ERS) comprising a Flexible Arm Delivery Device (FDD) and Main Body Driving Device (MDD). Through systematic structural design and validation, the ERS achieves stable four-DOF control: 360° circumferential rotation, 180° bending, 280–330 mm axial delivery, and ±15° rotational compensation. Key innovations include dual-encoder slippage detection in the FDD and a gripper assembly addressing circumferential angle loss through real-time torque sensing. Experimental results confirm compatibility with dual-channel endoscopes without custom modifications. The system demonstrates proof-of-concept for comprehensive endoscopic control while maintaining compact design (<0.2 m² per module) and low manufacturing cost. Although force-sensing capabilities require further refinement, this work establishes a foundation for accessible robotic platforms in gastrointestinal endoscopy, with identified limitations guiding future development toward clinical viability.

1. Introduction

Minimally Invasive Surgery (MIS) has revolutionized early-stage gastrointestinal tumor treatment, with flexible endoscopy as a cornerstone of organ-preserving interventions. Early detection of mucosal-layer tumors—enabled by widespread screening—eliminates the need for invasive laparoscopic resection, allowing complete tumor removal via natural orifices.

Endoscopic Submucosal Dissection (ESD) represents the pinnacle of this evolution, offering en bloc resection to preserve organ integrity [1,2]. Inserted orally (upper tract: esophagus, stomach, duodenum) or transanally (lower tract: colon, rectum), ESD achieves true “incisionless” surgery. However, its adoption lags in Western healthcare systems, creating disparities in access for patients with early-stage gastrointestinal malignancies [3].

The primary barrier to broader ESD use is technical complexity: the procedure demands exceptional manual dexterity to manipulate slender, limited-DOF instruments in confined spaces, requiring extensive training for proficiency [4,5,6]. These challenges have driven the development of robotic-assisted gastrointestinal endoscopic surgery (RAGES) systems, which aim to augment human capabilities and reduce the learning curve.

Current RAGES platforms address these issues through diverse designs, yet each has distinct limitations. The University of Twente’s system retrofits conventional endoscopes with motorized tip control but requires manual axial insertion. It is compatible with at least one type of commercial endoscope, though difficult to expand to others [7]. Yanpei et al.’s foot-controlled system offers pitch, yaw, rotation, and telescoping DOFs but has axial space limitations and difficult assembly [8]. The STRAS v2 system from the University of Strasbourg achieves four degrees of freedom (pitch, yaw, rotation, and telescoping) with comprehensive functionality but suffers from a substantial footprint (at least > 0.5 m²) and relies on trolley-based movement for axial positioning—resulting in cumbersome human–robot collaboration. Notably, this system—like Sivananthan et al.’s design (which struggles to achieve compatibility with commercial endoscopes [9])—is not compatible with such devices, further limiting its clinical application [10,11]. Basha et al.’s universal drive system enables rapid assembly (∼1 min) and 5-DOF control but depends on external robotic arms (e.g., UR5e), limiting clinical practicality [12]. Nakadate’s standalone system supports multi-DOF control and efficient assembly, yet axial insertion remains manual [13].

A thorough analysis of existing systems identifies fundamental design trade-offs: many platforms sacrifice full DOF automation (requiring manual control of axial translation or circumferential rotation), which complicates human–robot interaction and increases operational difficulty [7,14,15,16]. Systems with comprehensive DOF control often use rigid endoscopes or auxiliary robotic arms, limiting applicability in standard clinical settings [12,17,18,19]. While some modular designs address multi-DOF needs [9,20,21,22], few balance compactness, assembly efficiency, and universal compatibility. Horizontal endoscope mounting fundamentally constrains miniaturization [10,23]; suspended architectures offer a solution by decoupling axial movement from system translation but require dedicated delivery modules [11,13]. Alternative systems like Medrobotics’ Flex—using continuous bending segments—gain flexibility but lose reach, making them unsuitable for deep gastrointestinal procedures [24,25,26,27].

To address these challenges—operational complexity, compatibility constraints, footprint limitations, and assembly inefficiency—this study presents a novel Endoscopic Robotic System (ERS) with a suspended endoscope architecture (key comparison with existing systems in Table 1). The ERS comprises the two following modules. A Main Body Driving Device (MDD): Suspended configuration for 3-DOF control (pitching, yawing, and primary body rotation). A Flexible Arm Delivery Device (FDD): Friction wheel assemblies for axial delivery and gripper assemblies for rotational compensation.

Table 1. Comparison of flexible endoscope robotic systems.

System Name (Research Team, Year)	DOF	Floor Area	Compatibility with Commercial Endoscopes	Quick Assembly/Disassembly Feasibility
Ruiter et al. (2012) System	Pitch, Yaw	Approx. 0.5 m2	Compatible with at least one type, difficult to expand	Easy 1
Yanpei et al. (2021) Foot-Controlled System	Pitch, Yaw, Rotation, Telescoping	Axial floor area > 500 mm	Compatible with at least one type, difficult to expand	Difficult, involving key driving components
Sivananthan et al. (2021) Gaze-Controlled System	Pitch, Yaw, Rotation	Requires UR5 Robotic Arm	Compatible with at least one type, difficult to expand	Difficult, involving robotic arm and key driving components
Zorn et al. (2018) STRAS v2 System	Pitch, Yaw, Rotation, Telescoping	At least >0.5 m2	Not compatible	Difficult, involving key driving components
Basha et al. (2024) Generic Actuation System	Pitch, Yaw, Rotation, Telescoping	Requires UR5 Robotic Arm + Additional Guide Rail	Compatible with at least one type, difficult to expand	Relatively easy, multiple buckles
Nakadate et al. (2020) CBS-ESD Robot	Pitch, Yaw, Rotation	Hanging type, single module estimated < 0.2 m2	Compatible with at least one type, relatively easy to expand	Easy
ERS (This Study)	Pitch, Yaw, Rotation, Telescoping	Hanging type, two single modules each < 0.2 m2	Compatible with at least one type, easiest to expand	Easy

¹ Due to space constraints, further evaluation details of the metrics can be found in the Supplementary File.

Design priorities include the following: (1) full 4-DOF automation (no manual intervention); (2) compatibility with standard dual-channel endoscopes (e.g., GS-600DQ) without modifications; (3) compact footprint (<0.2 m² per module); and (4) tool-free rapid assembly via snap-fit mechanisms. Integrating adaptive clamping, dual-encoder slip detection, and torque-sensing compensation, the ERS addresses critical gaps in RAGES—aiming to lower barriers to high-quality ESD and democratize access across healthcare settings.

2. Materials and Method

2.1. System Architecture and Design Requirements

For RAGES procedures, the system design must address diverse procedural requirements across endoscopic interventions, enabling researchers and engineers to meet clinical needs. A formal requirements specification report will be drafted upon completion of this phase, with key considerations including the following:

1

The system shall enable the precise control of endoscope tip pitching (up–down) and yawing (left–right), ensuring accurate adjustment of the surgical field during interventions like ESD, polypectomy, and hemostasis.

2

The system shall provide axial insertion/retraction and circumferential rotation, enabling stable access to the operative site and full four-DOF manipulation across anatomical regions (esophagus, stomach, duodenum, colon, and rectum).

A complete list of all abbreviated terms used in this manuscript is provided in Table 2. To meet these requirements, this work introduces a novel Endoscopic Robotic System (ERS), as shown in Figure 1. The system consists of the Main Body Driving Device (MDD) and Flexible Arm Delivery Device (FDD), which coordinate to achieve full four-DOF control. The MDD—designed on a pulley-equipped hanger—fixes the endoscope main body and occupies < 0.2 m². The FDD, mounted on an adjustable-inclination support, handles axial delivery and rotational compensation (functions the MDD cannot fully cover). This system is compatible with standard dual-channel electronic endoscopes (e.g., GS-600DQ, Beijing Lepu Zhiying Technology Co., Ltd., Beijing, China) without customization.

Table 2. Abbreviations and their full forms.

Acronym	Explanation
RAGES	Robot-Assisted Gastrointestinal Endoscopic Surgery
ERS	Endoscopic Robotic System
FDD	Flexible Arm Delivery Device
MDD	Main Body Driving Device
FA	Friction Wheel Assembly
GA	Gripper Assembly
FP	Force Sensing Platform
MIS	Minimally Invasive Surgery
ESD	Endoscopic Submucosal Dissection
PWM	Pulse Width Modulation
ADC	Analog-to-Digital Converter
DOF	Degrees of Freedom

Figure 1. Overview of the proposed ERS and FA details: (a) flexible clamping details; (b) FDD overall structure; (c) MDD diagram.

2.2. System Specification Summary

The key specifications and performance metrics of the ERS are summarized in Table 3, with parameters based on this study’s laboratory data.

Table 3. Summary of ERS design specifications and performance.

Parameter	Specification	Module
Degrees of Freedom (DOF)
Pitching (Up/Down)	Full Range via Knobs	MDD
Yawing (Left/Right)	Full Range via Knobs	MDD
Primary Body Rotation	360° Continuous	MDD
Axial Delivery	280–330 mm Stroke	FDD (FA)
Rotational Compensation	Torque-sensing and ±15° Fine-tuning	FDD (GA)
Performance
Distal Bending Range	180°	MDD
Clamping Spring Stroke	12 mm	FDD (FA)
Axial force sensing accuracy	±1.3 N (within valid tilt range)	FDD (FP)
Torque sensing value	Theoretical torque: equal to flexible arm’s input.	FDD (GA)
Physical Properties
Module Footprint	<0.2 m2 per module	MDD and FDD
Endoscope Compatibility	GS-600DQ (Standard)	ERS
Platform Height @ 0°	680 mm	FDD (FP)
Platform Angle Adjust.	−23° to +20°	FDD (FP)

2.3. Main Body Driving Device (MDD)

Conventionally, endoscopists manually control the patient-side endoscope via built-in knobs while assisting surgeons with robotic systems—an approach adopted by systems like MASTER [28] and Via-Cath [29]. In contrast, the proposed MDD drives two dial wheels and the endoscope body via three motors, controlling three DOFs: pitching, yawing, and primary body rotation.

Suspending the endoscope main body reduces the system’s footprint. Horizontally arranged endoscopes require large strokes or manual intervention for axial delivery, but vertically suspended endoscopes allow a single lateral device to handle axial delivery.

The current device stably performs all basic functions. This study focuses on introducing the suspended endoscope as a delivery system auxiliary and demonstrating FDD design feasibility. Interchangeable fitting ferrules in the MDD accommodate diverse endoscopy wheel profiles (see Figure S1). Refined MDD design and optimization will be explored in future work.

2.4. Flexible Arm Delivery Device (FDD)

The FDD is divided into two functional assemblies—Friction Wheel Assembly (FA) and Gripper Assembly (GA)—based on its control of two DOFs: axial delivery and circumferential rotation. It also includes a Force Sensing Platform (FP) for collision detection, force sensing, and clinical platform compatibility.

2.4.1. Friction Wheel Assembly (FA)

The FA uses the friction wheel drive principle: two wheels radially compress the flexible arm to generate axial friction. Both wheels employ a cantilever beam-type single-side shaft fixing scheme, enabling rapid assembly/disassembly—critical for clinical efficiency and sterilization. Its key innovation is the adaptive clamping mechanism (Figure 1a).

Axial slippage during delivery (exacerbated by biological fluids) reduces operational accuracy. The FA detects slippage via real-time encoder value comparison. The driven wheel uses a lead screw to compress a spring, controlling clamping force linearly (this balances maximum static friction against the risk of clamping damage to the flexible arm). The wheels’ clamping contour distributes force evenly, reducing endoscope damage.

2.4.2. Gripper Assembly (GA)

A key challenge in endoscopic rotation is angle loss: under ideal conditions, the endoscope main body’s rotation transmits to the arm’s end at about 1:1 [30], but in actual surgical conditions, the necessary tortuosity of the flexible arm will compromise this (Figure 2).

Figure 2. Comparison of three bending postures of the endoscopic flexible arm.

The GA is actuated by two motors. One motor controls the rotation of the hollow screw. The resulting linear motion of the nut drives three sets of coaxial parallel four-bar linkages (grippers), enabling radial clamping of the flexible arm. The contact ends of the grippers are U-grooved pulleys. The clamping action of these three grooved pulleys not only achieves circumferential positioning of the flexible arm but also preserves its degree of freedom for axial advancement. The other motor drives the rotation of the gripper assembly via a gear system. Since this motor is mounted solely on a torque sensor, with the flange bearing providing only support (as shown in Figure 3b,c), the actual operational torque $T_1$ can be directly measured in real time as torque $T_4$ by the sensor.

Figure 3. GA structural diagram: (a) gripping details; (b) GA overall structure; (c) Rotation motor structure.

Rotational angle loss is an inherent issue that requires compensation. The GA operates under two primary scenarios: (1) when the flexible arm is extended only a short distance and undergoes sharp bending—a situation rarely encountered in practice but directly addressable by GA rotation (as shown in the third condition of Figure 2); and (2) when the flexible arm is inserted deeply with moderate bending—major rotation operations are managed via torque detection, whereas minor ones are handled directly by the GA.

Under scenario (2), when the MDD rotates, the middle segment of the flexible arm first undergoes an “idle rotation angle”—i.e., an ineffective rotational interval. Upon completion of this rotation, a sharp torque peak occurs at the GA clamping point, which triggers a system alert or control response, thereby improving the control efficiency of the rotational degree of freedom. If a positional deviation remains after the idle rotation, the GA can independently drive the distal end of the flexible arm to adjust by up to ±15°, a compensation action unaffected by the idle rotation occurring in the middle segment.

2.4.3. Force Sensing Platform (FP)

During axial delivery, two risks must be addressed: (1) accidental mucosal scraping/collision, and (2) overstretching the flexible arm. The FP mitigates these via real-time axial force detection.

To overcome static friction, the FP includes a slide rail and adjustable-angle fixed platform. As shown in Figure 4, a sliding clearance between the platforms allows the movable platforms (FA and GA) to slide and contact the fixed platform, thereby activating the tension–compression sensors. Within the stable tilt range—where the movable platform can consistently overcome static friction and initiate sliding—the sensors can reliably detect trends in axial external forces. Theoretically, even a minimal axial force applied to the flexible arm is distributed among three components: the sliding force of the movable platform, friction, and the tension/compression measured by the sensors. This distribution prevents the axial force from being completely masked by static friction when the system is near horizontal or at insufficient tilt angles, thereby avoiding undetectable scenarios. As a result, the sensor readings or their variation trends can alert the surgeon to potential collisions or overstretching.

Figure 4. FP structural diagram and axial force experiment role: (a) FP details; (b) FP angle base structure; (c) Axial force detection experiment schematic.

The overall workflow of the complete ERS is illustrated in Figure 5.

Figure 5. ERS surgical workflow and DOF control logic diagram.

3. Results and Discussion

To meet the functional verification needs of each structural part of the FDD, the four motors of this device are all low-cost brushed DC gear motors. This selection helps reduce the device’s debugging cycle and manufacturing cost, allowing rapid judgment of the design’s rationality and feasibility only through encoder readings and the intuitive working effect of driving the flexible arm. However, this approach introduces known limitations, as the performance of these components may affect the quantitative rigor of certain verification experiments. A comprehensive performance evaluation with higher-precision hardware is planned for subsequent studies.

The endoscope used in this experiment is a dual-channel electronic endoscope for the digestive tract (GS-600DQ). The parameters of the key electrical components in the FDD are listed in Table 4.

Table 4. The parameter table of key electrical hardware.

Name	Model	Key Specifications	Function in Charge
Motor 1 1	BDC	60 RPM (i = 168)	Delivery
Motor 2	BDC	37 RPM (i = 270)	Clamping
Motor 3	BDC	178 RPM (i = 56)	Gripping
Motor 4	BDC	60 RPM (i = 168)	Rotation
Encoder 5	AB-Phase	12 pulses	Slippage Detection
Sensor 1	Tension 2	0–5 kg	Collision Detection
Sensor 2	Torque	0–2 NM	Angle Detection 3
Control System	STM32	F407zgt6	Main Control

¹ The type of motors are JGB37-520 type, manufactured by szcmmotor in Huizhou, Guangzhou, China., operating at 12 V and equipped with a built-in AB-phase encoder (11 pulses), the encoder numbers are the same as the motor numbers; for example, motor1 corresponds to encoder1. ² Tension–Compression Sensor (HYLY-007, from HYchuangan, Bengbu, Anhui, China). ³ Idle Rotation Angle Detection (S4001, from STBD, Shenzhen, China).

Details of the experiment implementation and on-site layout can be viewed in the Supplementary Video attached to this article.

3.1. System Motion Capability Experiment

The overall motion capability reflects the robot system’s ability to control the motion of the endoscope’s flexible arm. Since the FDD can only control the flexible arm’s delivery and small-range circumferential rotation in terms of DOF, we integrate the MDD into this part of the experiment to ensure system integrity; by installing the endoscope, we test the stable motion range of each DOF using the drive motor.

For measuring instruments: the axial stroke of the flexible arm is measured with a tape measure (JC-06, DELIXI ELECTRIC, Linyi, Shandong, China); the compression amount of the pressure spring and the nut stroke of the GA are measured with a vernier caliper (YBKC-01, DELIXI ELECTRIC, China); the tilt angle of the FP and the roll angle of the GA are read directly from the designed 3D model; the pitch/yaw angles of the flexible arm are read by comparing with a common 360° protractor pattern; and the force magnitude is read in real time using the host computer provided by the sensor manufacturer (HYchuangan, China). The resulting motion parameter diagram, which provides an overview of the ERS 4-DOF control range, is shown in Figure 6. The detailed profiles of individual parameters (e.g., axial stroke–tilt angle relationship and rotational compensation range) are presented in Figure 7.

Figure 6. The overview of the endoscope driving parameters for the ERS.

Figure 7. Detailed diagram of the driving parameters: (a) performance details Figure 1; (b) performance details Figure 2.

Through experiments, it is confirmed that the system can stably achieve four DOFs of the flexible arm: pitching, yawing, telescoping, and rotation. The FP allows the FDD to have a sufficiently large elevation angle, enabling the tension–compression sensor to sense the initial contact force (the range obtained through continuous testing is 15 N–19 N; therefore, the sensing range is approximately 30 N).

A primary limitation identified in this experiment is the effective axial stroke. Although the FDD module has the ability to independently deliver the flexible arm with an axial stroke of 700 mm, after being integrated into the ERS system, the effective axial stroke (with GS-600DQ) is only about 280 mm (at a 23° depression angle) to 330 mm (at a 20° elevation angle). This is the result of using an endoscope with an effective length of only 1000 mm. To meet the operational requirements of ESD, the available length achieved by the robotic system should reach 900 mm. Replacing it with a longer endoscope, such as the ESM-370 (Weishikang Medical Electronics Co., Ltd., Shanghai, China) (1680 mm effective length), could potentially achieve an available length of 960 mm to 1010 mm in this system. Alternatively, we can further optimize the structural design, module usage mode, and layout of the robotic system.

3.2. Operational Performance Experiment

3.2.1. Basic Surgical Operation Experiment

The operation mode of ERS directly affects its usability and operational experience. Based on the ADC peripheral of the STM32 control system (F407ZGT6, from Your Cee, Shenzhen, China), a simple wired remote controller is developed. The two toggle directions of a single joystick are respectively set as the forward and reverse control directions for two motors (e.g., the forward and backward toggle directions of the left joystick are configured to control the forward and reverse rotation of Motor 1 with 0–30% PWM, enabling teleoperation of Motor 1). In turn, open-loop control of the four motors of the FDD is realized via two joysticks. The toggle directions of the joysticks are set to correspond to the associated motion directions of the flexible arm, which helps improve operational intuitiveness. For the MDD, Bluetooth control via the HC-05 module is adopted to enable convenient debugging, and it assists the FDD in completing the experiments.

The setup of this surgical operation experiment is shown in Figure 8, and a working environment for the flexible arm is established: an open box with a delivery port is used, and 7 target points are placed at different positions inside the box to simulate the locations of target sites for various endoscopic procedures. The experimenter controls the joystick handle and Bluetooth software simultaneously by a single person to drive the system, guiding the end of the flexible arm to reach the simulated target positions. The video of the physical object’s motion effect can be found in Appendix A [Single-Person Operation of ERS].

Figure 8. The experiment setup diagram for the surgical operation experiment.

During the experiment, it is observed that the control of the flexible arm’s delivery, pitching, and yawing is extremely sensitive. However, since the control of pitching and yawing has not yet been implemented through a mapping method similar to that of the joystick handle, controlling these two DOFs proved challenging, indicating a need for a more intuitive mapping strategy to improve usability.

Additionally, a key finding was the inferior clamping performance of the GA compared to that of the FA for rotational tasks: only when Motor 4 is stalled under 50% PWM can the generated friction force be sufficient to drive the entire endoscope (including the flexible arm) to rotate. Three main reasons are identified for this phenomenon: first, the GA lacks a flexible clamping structure, making it difficult to control the clamping force; second, the tangential friction force required for rotation is significantly greater than the axial friction force required for delivery; third, the U-shaped grooved wheels of the GA are made of smooth nylon without anti-slip patterns, resulting in a lower friction coefficient compared to the FA friction wheels (3D-printed with ABS).

Subsequently, to validate this hypothesis, the surface of the grooved wheels is manually modified to increase its coefficient of friction. Upon repeating the experiment, it is found that the anti-slip effect for rotation is indeed improved (with only approximately 45% PWM required to drive rotation). Overall, the GA can achieve a certain degree of rotation control: within a rotation angle range of 30°, it can independently control the end of the flexible arm to reach the simulated target areas.

3.2.2. Tilt Angle Force Experiment

To characterize the force sensing capabilities of the FP, an experiment was conducted to investigate the influence of static and sliding friction. Specifically, this experiment only involves adjusting the tilt angle to observe the variation characteristics of the force. The setup of the tissue collision force experiment is depicted in Figure 9.

Figure 9. The tissue collision force experiment setup diagram.

In this experiment, the force magnitude is read in real time using the host computer provided by the sensor manufacturer (HYchuangan, Bengbu, Anhui, China). Considering that the sensor has a detection accuracy of only 0.1 N and the lack of a calibration compensation algorithm, no zero-calibration is performed on the tension/compression force data. This approach facilitates the judgment of the actual force-bearing condition and also makes it easy to determine whether the absolute measuring range (50 N) is exceeded.

In the experiment, when the movable platform of the FDD stably shows a downward sliding trend, the absolute value of the initial tension/compression force detected by the tension–compression sensor is approximately 17.3 N (in practice, this value still fluctuates by about ±1.3 N due to sliding friction), corresponding to an inclination angle of roughly 8°. Therefore, when the downward or upward tilt angle exceeds 8°, the movable platform maintains a stable downward sliding trend, enabling the most accurate detection of external forces—this serves as the minimum tilt threshold for all subsequent force measurement experiments.

To preliminarily verify the real-time performance and accuracy of the sensor after installation on the device, a downward tilt scenario is tested: starting from an initial downward tilt angle of approximately 13°, an initial force of 28.5 N is recorded. With the sensor kept powered on, as the downward tilt angle is further increased, the force is observed to increase controllably at the sensor’s precision (0.1 N) as shown in Figure 10. However, when the force returns to around 17.3 N, the tilt angle of the FP is close to horizontal (approximately 1° upward tilt). When the tilt angle is 0° (horizontal state), the sensor force is approximately 17.5 N, which indicates that the horizontal static friction force is at least greater than 17.5 N.This significant static friction may be attributed to the use of standard drawer slides in the FP assembly.

Figure 10. The tilt angle experiment setup diagram.

This experiment already demonstrates that the influence of friction is significant enough to greatly reduce the accuracy of tissue collision force experiments. Nevertheless, since the sensor can still provide real-time feedback with a minimum precision of 0.1 N on the device, the subsequent experiment aimed to determine if the trend of force variation, rather than its absolute value, could reliably indicate different operational states. This approach, if viable, would hold reference value even with low-cost components.

We quickly conducted experiments under various working conditions, and basically obtained relatively obvious and distinguishable force curves. Taking Figure 11 as an example, the experiment was carried out at an 8° depression angle. Phase 2 (Step 1) to Phase 4 (Step 3) correspond respectively to the no-load delivery stage of the GA (without clamping), the simulated friction stage under normal friction of the human esophagus (using holes for simulation), and the collision stage where the end of the flexible arm breaks through the wet tissue while maintaining the friction state.

Figure 11. Force curve diagram of the experiment under three working condition stages.

Although the sensed force value of this curve has a repeatability deviation of approximately ±1 N under different working conditions, the characteristics under each working condition can still be clearly distinguished through the force change rate. However, the reliability of this method is compromised by several factors. First, the experimental conditions were not exhaustive. Second, obvious vibrations generated during the operation of the FDD, or vibrations generated during the GA clamping process controlled by Motor 3, will make the characteristics of the force curve extremely irregular. These findings suggest that while trend-based force sensing is a promising concept, its current implementation is not sufficiently robust for reliable application due to mechanical noise, and the video of the intuitive response effect of the tension–compression force sensor can be found in Appendix A [Elevation Angle and Human Tissue Collision Test].

3.2.3. Discussion

First, in terms of four DOFs control performance (pitching, yawing, axial telescoping, and circumferential rotation), the achieved stable manipulation ranges (e.g., 360° circumferential rotation and 180° bending) confirm the mechanical design’s adequacy for macro-scale positioning of the flexible arm during various gastrointestinal endoscopic procedures. As summarized in Table 3 (Section 2.2), the system achieves 280–330 mm axial delivery stroke and ±15° rotational compensation capability through the FDD. However, the operation mode still needs to be improved in terms of intuitiveness and simplicity.

Second, the roles of the GA design in both mitigating circumferential angle loss and reducing the flexible arm’s usable stroke require further discussion. Future designs must therefore focus on miniaturizing the GA or better balancing this functional trade-off. For instance, it is feasible to reduce the height of the MDD, thereby shortening the middle section of the flexible arm and increasing the axial telescopic stroke of the ERS. These design efforts are all aimed at better fulfilling the essential requirements of surgical procedures.

Third, the low-cost force-sensing method proved inadequate for reliable collision detection due to its qualitative nature and noise susceptibility. Should this function be retained, any future implementation must incorporate higher-precision sensors and effective vibration damping.

Finally, the manufacturing cost is remarkably low, which highlights the system’s potential for low-cost production. Using low-performance motors and sensors was a deliberate choice to prioritize rapid functional verification, but future work should balance necessary hardware upgrades—essential for translating laboratory prototypes into commercial products.

4. Conclusions

This study designed, fabricated, and validated an Endoscopic Robotic System (ERS) for robot-assisted gastrointestinal endoscopic surgery (RAGES). Through systematic structural design and experimental validation, the ERS achieved stable control over four degrees of freedom of the endoscopic flexible arm, including 360° circumferential rotation, 180° bending, 280–330 mm axial delivery, and ±15° rotational compensation—capabilities that address the macro-positioning requirements of various gastrointestinal endoscopic procedures including ESD, polypectomy, and therapeutic interventions.

The main innovations include the following: a high-precision delivery control module (FDD) with flexible clamping and dual-encoder slip detection; a gripper assembly (GA) that mitigates circumferential angle loss—common in suspended endoscope configurations—through real-time torque sensing and ±15° fine-tuning; and compatibility with standard dual-channel endoscopes (e.g., GS-600DQ) without the need for custom modifications. The low-cost manufacturing approach demonstrates feasibility for rapid prototyping and accessible production.

Key limitations pave the way for future research. First, miniaturization of the GA is necessary to maximize the endoscope’s usable stroke—currently reduced from 700 mm (standalone FDD) to 280–330 mm (integrated system). Integration of longer endoscopes or redesign of the GA could address this constraint. Second, the force-sensing approach requires validation with higher-precision sensors and vibration damping mechanisms before clinical translation. The current implementation’s susceptibility to mechanical vibrations and reliance on qualitative trend analysis necessitate substantial refinement. Finally, the hybrid control interface (joystick and Bluetooth) should be unified to streamline the surgical workflow and reduce cognitive load on operators.

This research offers a design reference for endoscopic robotic systems targeting RAGES applications and serves as a foundational step toward clinically viable, accessible robotic platforms for advanced endoscopic procedures. By demonstrating that a compact, low-cost system can achieve comprehensive four-DOF control while addressing key challenges such as angle loss and slippage detection, this work establishes proof-of-concept for democratizing robotic assistance in gastrointestinal endoscopy.