Pivot-constrained robotics for personalized ophthalmic surgery: a narrative review with theoretical synthesis and illustrative simulations

Introduction

Artificial intelligence (AI), advanced sensing, and robotic assistance are increasingly recognized as major drivers of innovation in ophthalmology (1,2). Existing reviews have also summarized ophthalmic robotics and AI from the perspectives of surgical vitreoretinal disease, optical coherence tomography (OCT)-aided robotic systems, robotic assistance, imaging-guided intervention, and emerging applications in surgical retina and ophthalmic microsurgery (1,3). However, they have generally focused on robotics, imaging, sensing, and AI as enabling technologies, rather than examining the shared physical constraint that governs instrument motion in ocular surgery. In particular, the relationship among ocular entry-site geometry, remote center of motion (RCM) enforcement, constrained inverse kinematics, task-specific motion structure, and personalized robotic execution has not been explicitly emphasized. In ophthalmic microsurgery, the instrument does not move freely in Cartesian space. Rather, it typically enters the eye through a corneal, limbal, or scleral opening that functions as a pivot-like constraint (4). As a result, shaft orientation, insertion depth, tip motion, and tissue interaction become tightly coupled. More broadly, the medical robotics literature has consistently emphasized the importance of RCM technologies, constrained inverse kinematics, and safety-aware control in minimally invasive procedures (5). The central challenge, therefore, is not simply to improve perception, automate subtasks, or generate better predictions, but to produce motion that remains geometrically feasible, anatomically consistent, and safe under a strict entry-site constraint.

RCM constraints are a central issue in ophthalmic robotics (5), because most modern ocular procedures are performed through ports or small incisions (Figure 1). In ophthalmology, the workspace is extremely small, tissue tolerances are narrow, permissible forces are low, and clinically meaningful maneuvers are often highly structured (6). Movements that violate these restrictions may injure delicate ocular tissues or induce globe rotation, potentially compromising the procedure (7). A desired trajectory may appear straightforward when expressed as a Cartesian tip path, yet become partially infeasible or poorly conditioned once the pivot constraint, tool length, and robot kinematics are taken into account. Similarly, a controller may achieve acceptable average tracking error while still producing excessive port deviation, damping-induced slowing, singularity-related instability, or unsafe transmission of force (5). These are not minor implementation details. Rather, they are direct consequences of how ophthalmic robotic surgery is formulated at the motion-control level.

Figure 1 Pivot/port constraints in ophthalmic robotic surgery. Ophthalmic robotic manipulation can be framed as a tip-tracking problem under a shaft-through-port constraint. (A) Surgical geometry is defined by the relationship between the intended tip motion and the requirement that the instrument shaft pass through a fixed incision or port. (B) This requirement can be expressed as a line-through-hole constraint and enforced using structural, kinematic, or optimization-based approaches. (C) Personalization is achieved by mapping patient-, setup-, and phase-specific information to explicit constraint parameters, enabling individualized constrained motion planning and execution. RCM, remote center of motion.

In many discussions, personalized robotic surgery is implicitly reduced to prediction, such as identifying anatomy, estimating risk, or recommending an operative plan (8,9). Yet predictive AI alone does not constitute meaningful robotic personalization unless it changes how the robot actually moves. In ophthalmic robotic surgery, personalization should instead be defined as the translation of patient-specific geometry, surgical phase, and safety requirements into explicit motion constraints, reference-generation strategies, and execution architectures. In practical terms, anatomy, entry-site geometry, tissue fragility, phase-specific risk, and uncertainty should modify the admissible workspace, the generated trajectory, the control hierarchy, and the safety envelope (10).

Accordingly, this narrative review proposes a theory-grounded translational framework for ophthalmic robotic surgery as a pivot-constrained manipulation problem, focusing on the physical and control principles of pivot- and RCM-constrained motion, related failure modes, and the conditions under which AI can meaningfully shape robotic execution. We present this article in accordance with the Narrative Review reporting checklist (available at https://aes.amegroups.com/article/view/10.21037/aes-2026-0028/rc).

Methods

This study was designed as a narrative review aimed at developing a theory-grounded and translational framework for understanding ophthalmic robotic surgery as a pivot-constrained manipulation problem. Rather than providing a broad survey of AI and robotics in ophthalmology, the review was structured to integrate four complementary components: theoretical foundations, narrative literature review, illustrative simulations, and future perspectives.

First, we examined the theoretical foundations of pivot- and RCM-constrained robotic manipulation using relevant literature from robotics, minimally invasive surgical robotics, and ophthalmic robotic surgery. The theoretical framework was developed through a targeted search of PubMed, Embase, Web of Science, IEEE Xplore, and Google Scholar using ophthalmology- and robotics-related terms, including “remote center of motion”, “RCM”, “pivot constraint”, “entry-site constraint”, “constrained inverse kinematics”, “task-priority control”, “null-space control”, “damped least squares”, “virtual fixtures”, and “ophthalmic robot”. Reference lists of relevant articles and reviews were also screened. Literature was selected when it directly informed pivot/RCM geometry, constrained kinematic mapping, control strategy, execution stabilization, safety metrics, or ophthalmic task formulation. Particular attention was given to the geometric, kinematic, and control principles that determine pivot-constrained execution, including constrained inverse kinematics, motion feasibility, workspace conditioning, singularity, and safety-aware control. This component was intended to establish why pivot and RCM constraints should be regarded as central, rather than peripheral, to the formulation of robotic motion in ophthalmic surgery.

Second, a narrative literature review was conducted to examine published studies in ophthalmic robotic surgery, with the search strategy summarized in Table 1. The literature search was performed between 1 and 21 March, 2026, using Google Scholar, PubMed, Embase, Web of Science, and IEEE Xplore. The search timeframe covered studies published from January 2000 to March 2026 in order to capture the development of modern ophthalmic robotic surgery together with related advances in RCM mechanisms, constrained inverse kinematics, surgical robot control, and safety-aware manipulation. Search terms consisted of combinations of controlled vocabulary and free-text keywords related to ophthalmic robotics, retinal surgery, cataract surgery, subretinal injection, retinal vein cannulation, membrane peeling, RCM, pivot constraint, constrained inverse kinematics, surgical robot control, geometric decoupling, shared control, OCT guidance, and robotic manipulation. The review included peer-reviewed articles and conference papers published in English that addressed robotic applications in ophthalmology, including anterior-segment surgery, posterior-segment injection or cannulation, and posterior-segment non-injection or general manipulation. Non-English publications and studies outside the scope of ophthalmic robotic surgery were excluded. Because this study was designed as a narrative review, formal systematic review eligibility criteria, independent dual screening, and meta-analytic procedures were not applied. Instead, eligible studies were identified and reviewed descriptively according to their relevance to the theoretical foundations, translational applications, and representative ophthalmic robotic tasks addressed in this study (Figure S1). The included studies were subsequently organized into three broad categories: anterior-segment surgery, posterior-segment injection or cannulation, and posterior-segment non-injection or general manipulation.

Third, representative simulations of anterior- and posterior-segment maneuvers were incorporated to examine whether the theoretical principles discussed in this review could be consistently instantiated at the level of ophthalmic surgical tasks. The simulations were selected as theory-informed representative examples based on clinical recognizability to ophthalmic surgeons, dependence on a fixed ocular entry point, diversity of task-level motion geometry, and suitability for simplified geometric implementation. Continuous curvilinear capsulorhexis (CCC) was selected to represent anterior-segment circular motion through a corneal or limbal entry site; epiretinal membrane (ERM) peeling was selected to represent posterior-segment surface-oriented tangential manipulation near fragile retinal tissue; and small incision lenticule extraction (SMILE) lenticule dissection was selected to represent planar intrastromal swing motion through a small corneal side-cut tunnel. Rather than serving as formal experimental validation, these simulations were intended as theory-informed demonstrations to assess the practical coherence of pivot-constrained task formulation, including task geometry, pivot compatibility, and motion structure, in representative ophthalmic settings.

Finally, based on the theoretical synthesis, literature review, and illustrative simulations, future directions in ophthalmic robotic surgery were discussed, with particular emphasis on emerging technologies such as multimodal large language models (MLLMs), physical AI, digital twins, and hybrid translational architectures.

Theoretical background of pivot-constrained robotic surgery

Figure 2 illustrates the two representative strategies for enforcing this constraint in ophthalmic robotic surgery, namely a mechanically enforced RCM and a virtual RCM realized by a high-degree-of-freedom serial robot, together with their main translational trade-offs (11). The core translational logic derived from this framework is summarized in Table 2, while detailed mathematical formulations and extended theoretical comparisons are provided in Appendix 1.

Figure 2 Representative approaches for enforcing pivot constraints in ophthalmic robotic surgery. (A) A parallel or dedicated RCM mechanism provides a mechanically enforced pivot at the port, offering intrinsic safety and predictable motion but with limited workspace and task specificity. (B) A high-degree-of-freedom serial robot can instead realize a virtual RCM through kinematic control, enabling greater workspace flexibility and patient-specific adaptation, but requiring calibration and continuous monitoring. Together, these approaches illustrate the trade-off between hardware-level constraint enforcement and software-defined constrained autonomy. (+) indicates an advantage or favorable characteristic, whereas (−) indicates a limitation or disadvantage. DoF, degree of freedom; RCM, remote center of motion.

The first conceptual requirement is the definition of the surgical task (12,13). In ophthalmic robotic surgery, the robot is rarely asked to execute a generic point-to-point movement. Instead, it must preserve a clinically meaningful motion class, such as circular capsular motion, retinal surface following, retinal vessel approach, membrane peeling, or depth-limited injection. Robotic quality in ophthalmology depends on preserving clinically meaningful motion structure rather than on unconstrained trajectory tracking alone (25). This distinction is important because the intended surgical motion is shaped not only by the target itself, but also by the entry geometry, local anatomy, available intraocular workspace, and the need to protect delicate ocular tissues. A path that appears acceptable in Cartesian space may therefore become impractical, poorly conditioned, or unsafe once pivot location, tool length, angular limits, and robot kinematics are considered. Accordingly, robotic design in ophthalmology should begin from task geometry and surgical intent, not from endpoint tracking alone.

Once the surgical task has been defined in this structured way, it must be embedded within an architecture capable of preserving compatibility with the ocular entry site. One is a dedicated parallel or RCM mechanism that provides a mechanically enforced pivot at the port (14), thereby offering intrinsic geometric consistency, predictable shaft behavior, and potential advantages for tasks requiring stable needle-axis control within a narrow cone of motion (26,27). The other is a high-degree-of-freedom serial robot that realizes a virtual RCM through kinematics and control, enabling greater workspace flexibility, shared autonomy, and adaptation to patient-specific entry geometry (5,28). Motion feasibility depends on both task formulation and the selected RCM architecture. Higher-degree-of-freedom systems may benefit from virtual RCM strategies, particularly when redundancy can be exploited through null-space-based constraint handling (15,18), whereas dedicated mechanisms may provide more intrinsic pivot consistency in task-specific settings (27). The practical point is that ophthalmic robotic feasibility is determined jointly by what the robot is asked to do and by how the system enforces shaft-port compatibility during that task.

This leads directly to the role of inverse kinematics, control, and stabilization. In unconstrained robotics, the principal challenge is often how to map a desired end-effector motion into joint motion. In ophthalmic robotic surgery, however, that mapping must remain compatible with the pivot constraint throughout execution (16,17). As summarized in Table 2, the question is therefore how tip tracking, pivot maintenance, and stability should be balanced under constrained intraocular access. Even when a task is nominally feasible, performance may still degrade near difficult configurations or under uncertainty if the control method is not sufficiently stable and safety-aware. Different strategies address this in different ways. Some methods balance tracking and pivot preservation through weighted optimization (19), whereas others use task-priority or null-space-based approaches so that the main surgical objective is preserved while remaining degrees of freedom are used for secondary correction (20). Geometry-based reference generation may further improve consistency by embedding pivot compatibility into the desired motion before execution begins (21).

A further implication is that failure in ophthalmic robotic surgery should not be interpreted only as loss of positional accuracy. Clinically relevant failure modes also include pivot mismatch (23), calibration error, loss of feasible motion, poor conditioning, singularity-related slowing (24), path deterioration, and unsafe force transmission. A robotic system may therefore appear acceptable in terms of average tip error while still performing poorly with respect to pivot maintenance, structured path preservation, motion smoothness, or safety-related mechanical interaction. For this reason, ophthalmic robotic systems should be evaluated not only by target-tracking accuracy, but also by execution stability, shaft-port compatibility, and other safety-relevant behaviors that more directly reflect translational robustness under constrained intraocular access (22).

How pivot/RCM constraints are instantiated across ophthalmic robotic tasks

The literature on ophthalmic robotic surgery is heterogeneous in platform design, task selection, and translational maturity, yet many of these differences can be understood within a common pivot- or RCM-constrained framework. Across anterior- and posterior-segment applications, the central challenge remains the same: to generate safe and precise intraocular motion through a constrained ocular entry site. On this basis, the literature may be organized into three broad domains: anterior-segment robotic systems related to cataract surgery and other structured anterior-segment maneuvers, posterior-segment injection or cannulation systems, and posterior-segment non-injection or general manipulation systems.

Anterior-segment robotic platforms

Anterior-segment robotic systems remain less mature than posterior-segment systems, but they are increasingly important because cataract-related maneuvers provide clear examples of structured, phase-dependent ophthalmic motion. In this domain, the literature is dominated by semiautomated lens extraction, robotic cataract manipulation, virtual-fixture-based training systems, and task-specific robotic capsulorhexis platforms rather than fully integrated clinical cataract robots (Table 3). A common feature of these studies is that they treat the anterior segment as a workspace in which high precision, geometric regularity, and tissue-aware execution are required, but they differ considerably in how explicitly they formulate pivot or RCM constraints.

One important group of studies combines OCT guidance with robotic cataract manipulation. The OCT-guided robotic cataract surgery system reported by Chen et al. addressed semiautomated lens extraction in postmortem pig eyes and integrated robot-to-eye alignment, anatomical modeling, surgical path planning, and irrigation/aspiration handpiece navigation under real-time OCT feedback (29). This study is notable because it moves beyond generic robotic assistance and attempts to link anterior-segment imaging, geometry, and path execution in a single workflow. Similarly, Shin et al. used an OCT-guided robot with deep-learning segmentation to identify lens material, capsule, cornea, and iris during robotic lens-fragment extraction (31). This work is significant not simply because it adds AI, but because it uses tissue-aware segmentation to guide task execution, suggesting a path toward anatomy-informed robotic action in the anterior segment. At present, however, these OCT-based platforms remain limited to ex vivo porcine or benchtop settings and do not yet represent full cataract-surgery robotic systems spanning all operative phases.

A second group of studies emphasizes control architecture more explicitly. Yang et al. described a master-slave cataract surgery robot with a software-enforced RCM at the incision point, using phase-dependent modes, virtual fixtures, and virtual force feedback for safety-constrained manipulation (30). This work is especially relevant to the present review because it demonstrates that anterior-segment robotics is not simply about positioning accuracy, but about encoding surgical phase and safety logic into the control structure. Lin et al. likewise developed a vision-guided hybrid cataract robot with explicit RCM-constrained control, aimed at improving motion accuracy and smoothness (32). In their preliminary validation, 5-mm circular tracking produced maximum errors of 0.084 mm at the actuator end point and 0.292 mm at the RCM point, and capsule-edge tracking error was lower with the robot than with manual tracking (0.129 vs. 0.764 mm). These studies show that once cataract surgery is viewed as structured intraocular manipulation through a fixed entry site, classical ideas from constrained robotics such as RCM control, geometric guidance, and phase-aware operation become directly relevant. Nonetheless, the validation of these systems remains preliminary, with most evidence still confined to bench-level feasibility rather than clinically integrated use.

A third theme in the anterior-segment literature is structured task guidance rather than complete operative automation. Varga et al. reported a shared-control cataract training robot with geometric virtual fixtures and haptic guidance for novice training (33). In a simulator pilot involving one non-medical participant and 12 incision tasks, the proposed shared-control approach reduced completion time from 11.49 to 8.18 s and near-target time from 2.54 to 0.41 s compared with no haptic support. Although this system is focused on training rather than live surgery, it is highly informative from a theoretical standpoint because it demonstrates how geometry and phase-specific constraints can be used to shape motion in a clinically interpretable way. Chen et al. later extended this structured-task perspective to AI-guided robotic capsulorhexis, using digitally extracted surgical features to guide a robot toward ideal circular motion in porcine-eye proof-of-concept experiments (34). In that study, clinical video-based analysis and lens caliper assistance increased the proportion of ideal capsulorhexis from 16.7% to 64.1%, while robotic capsulorhexis was further demonstrated in porcine eyes. This is especially relevant because capsulorhexis is not a generic point-to-point task but a highly structured circular maneuver, making it a natural example of how pivot-aware geometry, task definition, and robotic control can converge. Taken together, the anterior-segment literature suggests that the field is progressing toward task-structured and sensing-aware robotic assistance, but remains largely at the level of proof-of-concept, training, or phase-specific automation rather than complete clinical robotic cataract surgery.

Posterior-segment injection- or cannulation-related robotic platforms

Posterior-segment injection and cannulation represent the most mature and clinically advanced use cases in ophthalmic robotics. This is not surprising, because these tasks require a combination of micrometer-scale positioning, long-duration stability, safe shaft-port interaction, and highly controlled depth or vascular access, all of which are difficult to achieve manually. The literature in this domain includes intravitreal injection assistance (35), retinal vein cannulation, subretinal injection, and more recent autonomous or semiautonomous injection systems (Table 4). Compared with anterior-segment robotics, this category contains more direct human validation, but it also remains highly task-specific.

The retinal vein cannulation lineage is one of the clearest examples of translational progression. de Smet et al. described a robotically controlled micromanipulator with a parallelogram-based mechanical RCM for venular cannulation in a porcine model (36), while Gonenc et al. approached the same problem through a handheld micromanipulator combined with a force-sensing microneedle (37). These two studies illustrate two different philosophies: arm-based robotic access with explicit mechanical RCM on one hand, and highly refined handheld assistance on the other. Gijbels et al. then reported the Preceyes Surgical System for retinal vein cannulation in a first-in-human setting (38), marking an important clinical milestone for pivot-aware robotic vascular access inside the eye. More recently, Esfandiari et al. approached retinal vein cannulation and bimanual retinal manipulation through adaptive hybrid position-force control with fiber Bragg grating (FBG) sensing and a dynamic entry-site constraint (46). Although this latter study remains limited to phantom and pilot-user testing, it highlights an increasingly important shift in the literature: from purely kinematic pivot enforcement toward force-aware entry-site safety.

Subretinal injections have emerged as perhaps the most active posterior-segment robotic application, likely because gene therapy and cell delivery demand accurate subretinal targeting, stable needle insertion, and controlled bleb formation. Several distinct technical lineages are visible. One line focuses on mechanism and platform development, such as the hybrid parallel-serial subretinal injection robot of Zhou et al. (25), the mechanical-RCM-based robot-assisted subretinal injection system (RASR) platform of Yang et al. (42), and the OCT-guided ophthalmic surgical robot of Li et al. (49). These studies emphasize precision injection through explicit RCM mechanisms and master-slave control, but remain mainly in benchtop or preclinical validation. A second line centers on imaging-guided and AI-supported subretinal targeting. Mach proposed an OCT-guided robotic subretinal injection framework using deep-learning-assisted registration and depth-aware targeting (40), while Zhang et al. introduced an autonomous retinal needle-navigation platform with chance-constrained model predictive control (MPC), convolutional neural network (CNN)-based depth prediction, and scleral-force-aware navigation (44). These studies are important because they move the field beyond pure mechanism design toward geometry estimation, uncertainty handling, and safety-aware autonomy. A third line shows increasing translational maturation, as in the Preceyes clinical study by Cehajic-Kapetanovic et al. (41) and the soft micron accuracy robot (SMAR) system described by Chen et al. (45), which reported live-animal data and preliminary human validation.

More recent work suggests further diversification of injection-related robotics. Charreyron et al. explored magnetic navigation of a flexible microcannula, representing a distinct actuation paradigm from conventional rigid-arm RCM robotics (39). Ladha et al. used simulated subretinal injection in an artificial retina model to compare robotic and manual delivery under microscope and intraoperative OCT (iOCT) support, highlighting the importance of stabilization and imaging in gene-therapy workflows (43). Lian et al. developed HiPOSuR as a semiautomated subretinal injection system validated in rabbits (48), and Bian et al. extended the field toward autonomous subretinal and retinal vascular injections using multiview and multisensor fusion (47). Collectively, these studies show that posterior-segment injection robotics has advanced from assistive stabilization and mechanism design toward clinical translation and early autonomy. At the same time, most platforms remain closely tied to specific delivery tasks, and relatively few have demonstrated generalizable adaptation across broader posterior-segment workflows.

Posterior-segment non-injection and general manipulation platforms

The broader posterior-segment manipulation literature is more heterogeneous than the injection/cannulation literature. It includes general vitreoretinal platforms, membrane dissection and peeling systems, force-safety frameworks, handheld robotic tools, passive support systems, endoscopic assistance, mechanism-design studies, and benchmarking or sensing studies. This category is especially important for understanding the overall state of the field because it reveals how ophthalmic robotics has developed not only through direct clinical task success, but also through enabling technologies in force sensing, visualization, tremor suppression, RCM mechanism design, and entry-site localization (Table 5).

Several foundational platforms shaped this area. The Steady-Hand Eye Robot with micro-force sensing reported by Uneri et al. established an early framework for cooperative robotic retinal manipulation under mechanical RCM constraints (50). The IRISS system, introduced by Rahimy et al. (52) and later expanded in mechanical design and master-slave form by Wilson et al. (55), provided another major platform for general intraocular robotic manipulation. These early systems emphasized feasibility, cooperative control, and platform design rather than narrowly defined task-level optimization. Nasseri et al. reported a hybrid parallel-serial ophthalmic micromanipulator using virtual fixtures, highlighting an intermediate approach in which pivot-aware geometric guidance was applied to assist ophthalmic micromanipulation under experimental benchtop conditions (14). In contrast, handheld systems such as Micron (51) and later Yang’s tremor-canceling and microscope-guided tools (54) pursued enhanced precision without relying on a conventional arm-based RCM. This handheld-assistance lineage was further extended by Yang et al., who investigated automated intraocular laser surgery using a handheld micromanipulator, thereby suggesting a possible transition from tremor suppression toward task-specific automation in handheld ophthalmic robotics (53). These handheld devices are important historically because they show that ophthalmic robotic development did not begin from one unified architectural model. Rather, the field evolved through both full robotic manipulators and refined handheld assistive devices, each addressing different aspects of microsurgical limitations.

The membrane-dissection and peeling literature marks one of the clearest points of clinical relevance in posterior-segment robotics. The Preceyes Surgical System achieved first-in-human feasibility for epiretinal and internal limiting membrane (ILM) dissection over the macula, demonstrating that robotic retinal microsurgery could move beyond laboratory proof-of-concept into human clinical use (56,58). Subsequent work, such as the simulator-based comparison by Maberley et al., focused on comparing robotic and manual ILM peeling under more controlled conditions (58). More recently, Zheng et al. contributed not a robotic platform itself, but a motion-and-force benchmark framework for ILM peeling using FBG sensing, inertial sensing, and microscope-image analysis in rabbits (22). This kind of study is important because it provides quantitative targets for what a robotic peeling system must achieve in terms of motion quality and force control. Together, these works suggest that successful retinal manipulation requires not only robotic steadiness, but also task-specific understanding of structured tissue interaction.

A parallel line of work has focused on safety at the scleral entry site and during retinal manipulation. He et al. proposed a robot-assisted retinal surgery safety framework with scleral-force control (16), and Ebrahimi et al. extended this direction through adaptive scleral-force and insertion-depth control (57). Urias et al. likewise used the Steady-Hand Eye Robot with active scleral-force control or force-based scleral safety control in rabbit models (59). These studies are particularly important for the present review because they show that, in ophthalmic robotics, successful motion cannot be judged only by tip accuracy. The shaft-port interaction itself is a safety-critical variable, and force-aware control may be required even when geometric control appears satisfactory. In this sense, they help connect the literature on clinical retinal manipulation to the more general theoretical discussion of pivot constraints, entry-site compliance, and control failure modes.

Another important subset of the literature addresses enabling technologies rather than direct tissue manipulation. Smits et al. presented a parallel RCM design framework for robotic eye surgery, contributing mechanism-level insight into how pivot-constrained access can be achieved (27). Li et al. later extended this design-oriented direction with a spatial 2R1T RCM mechanism for subretinal surgical robotics, focusing on kinematic design, singularity analysis, and workspace optimization (64). Birch et al. studied trocar localization for robot-assisted vitreoretinal surgery, which is conceptually important because accurate entry-site geometry is a prerequisite for any meaningful pivot-aware execution (23). Zhou et al. developed the Eye Explorer robotic system (60) and later real-time fundus reconstruction and intraocular mapping approaches for endoscopic guidance (61), while Ashikaga et al. reported OQrimo as an endoscope-holding robot validated in a first-in-human comparative study (65). Although these systems do not directly represent fine tissue-manipulation robots in the same way as peeling or cannulation platforms, they broaden the translational picture by showing that posterior-segment robotics also depends on visualization, localization, navigation, and workflow support.

The literature also contains passive or assistive stabilizing systems, such as the passive surgical support robots studied by Yamamoto et al. for general ophthalmic stabilization (62) and ILM peeling tremor suppression (63). These are not active pivot-constrained robotic manipulators in the strict sense, but they reflect a clinically pragmatic branch of ophthalmic robotics in which the goal is not full robotic autonomy or telemanipulation, but rather selective enhancement of human stability and motion quality. This suggests that translational progress in ophthalmic robotics may continue to come not only from fully actuated robots, but also from hybrid ecosystems of robotic assistance, smart sensing, geometric guidance, and selective automation.

Illustrative task reformulations under a pivot-constrained framework

To complement the theoretical framework presented above, including inverse kinematics and control, illustrative simulations were performed for three representative ophthalmic tasks: CCC, ERM peeling, and lenticule dissection in SMILE. All simulations were implemented in MATLAB using the Robotics System Toolbox (66) under the assumption that a commercially available six-degree-of-freedom serial robot, modeled here as a UR5e manipulator (67), performs ophthalmic tasks through a virtual RCM. These simulations were not intended as formal experimental validation of a deployable robotic system. Rather, they were designed to illustrate how pivot-constrained motion planning and execution can be mapped onto ophthalmic task geometries with different anatomical locations, workspace requirements, and motion structures.

CCC

CCC was selected as a representative anterior-segment task because it involves a highly structured circular motion performed within a limited intraocular workspace. From a robotic perspective, this is a useful example of a task in which path geometry is clinically meaningful: the quality of motion depends not only on whether the tip advances, but also on whether it preserves the intended circular trajectory while entering through a fixed ocular port. Figure 3A schematically illustrates this situation, showing an intraocular instrument following a circular capsulotomy path under an entry constraint. Figure 3B provides representative simulation views and trajectory plots, indicating that the circular path can be executed while maintaining pivot-consistent access. Figure 3C shows the corresponding joint-angle profiles, demonstrating coordinated multi-joint motion during the maneuver. The simulated motion is also provided as an animation in Appendix 1. In the CCC simulation, the tool-tip path was defined as a circular trajectory on the anterior capsular plane. Each tip point was converted into a pivot-compatible end-effector pose using the incision-side entry point, tool length, and desired shaft direction, followed by frame-by-frame inverse kinematics to generate UR5e joint motion. The mathematical formulation used for this trajectory generation and pivot-compatible pose construction is provided in Appendix 1.

Figure 3 Illustrative simulation of CCC under pivot-constrained robotic motion. (A) Schematic view of the capsulorhexis task, in which the instrument tip follows a circular capsulotomy path under an ocular entry constraint. (B) Representative simulation snapshots and trajectory visualization showing constrained execution of the circular tip path. (C) Corresponding joint-angle profiles during task execution. The MATLAB code for this simulation is provided in Appendix 1, and the corresponding animated simulation is available in Video S1. CCC, continuous curvilinear capsulorhexis; TIP, tool tip.

ERM peeling

ERM peeling was selected as a representative posterior-segment manipulation task because it requires controlled approach to the retinal surface, delicate tangential manipulation, and stable behavior within a narrow and safety-critical intraocular workspace. In contrast to CCC, which emphasizes circular path geometry, ERM peeling emphasizes surface-oriented motion, local positional stability, and controlled instrument behavior near fragile tissue. Figure 4A presents both a schematic view and a representative surgical view of the peeling task, showing the instrument entering through a fixed ocular port to approach and manipulate the retinal surface. Figure 4B presents representative simulation views and trajectory or displacement plots demonstrating constrained execution during peeling. Figure 4C shows the corresponding joint-angle profiles, again indicating coordinated motion across multiple joints. An animation of the simulated peeling motion is provided in Appendix 1. In the ERM peeling simulation, a short tangential-lifting tip trajectory was defined from a predefined retinal membrane contact point. Each tip point was then converted into a pivot-compatible end-effector pose by aligning the tool shaft with the entry point-tip line, followed by inverse-kinematics-based joint generation. The mathematical formulation is provided in Appendix 1.

Figure 4 Illustrative simulation of ERM peeling under pivot-constrained robotic motion. (A) Schematic and representative surgical view of the membrane-peeling task, in which the intraocular instrument approaches and manipulates the retinal surface through a fixed ocular port. (B) Representative simulation views and trajectory/displacement plots showing constrained execution during peeling. (C) Corresponding joint-angle profiles during task execution. The MATLAB code for this simulation is provided in Appendix 1, and the corresponding animated simulation is available in Video S2. ERM, epiretinal membrane; TIP, tool tip.

Lenticule dissection in small-incision lenticule extraction

Lenticule dissection in SMILE was included as a more complex illustrative example because the critical manual part of the procedure is not simply tissue removal, but controlled identification, separation, and extraction of the lenticule within a confined corneal pocket. In the swing technique (68), the lower lenticule interface is first dissected, after which the dissector is lifted and swung at the margin to reach and separate the upper interface before lenticule extraction.

To clarify the simulated task sequence, the SMILE simulation was divided into six phases. Phase 1 represented incision entry and advancement of the tool tip toward the posterior/lower lenticule interface. Phase 2 represented posterior lenticule dissection, in which the tip followed a smaller arc on the posterior/lower lenticule plane opposite the incision. Phase 3 represented withdrawal of the tip back toward and through the incision. Phase 4 represented re-entry through the same incision and advancement toward the upper/cap interface. Phase 5 represented upper/cap-interface swing dissection, in which the tip followed a larger arc corresponding to the cap-side dissection path. Phase 6 represented return of the tool tip toward the incision and completion of the maneuver.

Figure 5A shows a representative surgical view of the dissection task. Figure 5B presents simulation views of constrained robotic execution together with the planned intra-corneal trajectory inside the surgical region of interest. Figure 5C shows the corresponding joint-angle profiles for the task. An animation of the simulated SMILE motion is provided in Appendix 1. The simulation therefore illustrates that pivot-constrained ophthalmic robotics is not limited to one anatomical region or one class of movement. Rather, the same theoretical framework can, in principle, be extended to different ophthalmic procedures provided that the task is reformulated in terms of anatomy-aware geometry, constrained access, and structured motion execution.

Figure 5 Illustrative SMILE/lenticule-dissection simulation under a pivot-constrained framework. (A) Clinical schematic of the SMILE swing maneuver through a small corneal incision. (B) Updated UR5e-based simulation showing the magnified workspace, fixed corneal entry point, end-effector position, instrument shaft, and tool tip. The right inset shows the top-view trajectory in the corneal plane. The posterior swing follows the lenticule circle, whereas the cap-interface swing follows the cap circle. The black line indicates the instantaneous shaft alignment through the fixed entry point, and the cyan curve indicates the tool-tip trajectory. (C) UR5e joint-angle trajectories during the phase-structured SMILE swing simulation. The simulation is intended as a conceptual demonstration of pivot-compatible task motion rather than experimental validation of a deployable robotic system. The animated simulation is available in Video S3. A.u., arbitrary units; EE, end-effector; SMILE, small incision lenticule extraction; TIP, tool tip.

From pivot-aware geometry to personalized execution: where AI enters the control stack

In a pivot-constrained robotic architecture, AI is clinically meaningful not merely when it improves prediction or interpretation, but when its outputs alter how the robot plans, constrains, and executes motion (1). From this perspective, personalization is best defined as the translation of patient-specific anatomy, surgical phase, and safety requirements into executable robotic constraints, reference trajectories, and control policies. Figure 6 illustrates this concept as a layered framework integrating patient data, multimodal AI, constrained robotic autonomy, and digital twins into a closed-loop architecture for planning, execution, and continuous refinement.

Figure 6 Conceptual framework for personalized ophthalmic robotic surgery integrating multimodal AI, constrained autonomy, and digital twins. (A) Patient-specific multimodal imaging and biometry are translated into executable task constraints, including entry-site selection, approach axis, and safety margins. (B) A multimodal AI layer, including vision-language or MLLMs, provides real-time surgical guidance through phase recognition, intent parsing, visual feedback interpretation, and safety gating. (C) A robotics layer implements constrained autonomy through shared control, port/shaft constraint enforcement, online monitoring, fallback behaviors, and optional interaction limits. (D) A digital twin layer supports preoperative rehearsal, in silico safety verification, intraoperative refinement, and postoperative policy updating, forming a closed loop for personalized planning, execution, and iterative improvement. AI, artificial intelligence; intra-op, intra-operative; MLLM, multimodal large language model; post-op, post-operative; pre-op, pre-operative; VLM, vision-language model.

At the first layer, patient-specific multimodal imaging and biometry are converted into task-relevant constraints, such as entry-site selection, approach axis, allowable workspace, and safety margins, rather than remaining as descriptive information (69). At the second layer, multimodal AI, including vision-language models (VLMs) and MLLMs, functions as a real-time interpretive and supervisory interface that supports phase recognition, intent parsing, visual feedback interpretation, and safety gating, thereby linking perception to execution rather than static image labeling (70,71). At the third layer, constrained robotics translates these patient-specific constraints and AI-derived situational cues into safe execution through shared control, online port/shaft constraint enforcement, monitoring, fallback behaviors, and force-related safety limits (69). At the fourth layer, a digital twin supports rehearsal, verification, adaptation, and policy updating across preoperative, intraoperative, and postoperative stages, linking planning, execution, and learning in a closed-loop translational framework (72,73).

Taken together, Figure 6 suggests a future direction in which ophthalmic robotic surgery moves beyond isolated elements of robotics, AI, or imaging and toward an integrated form of personalized physical AI. In such a system, multimodal data define executable geometric constraints, multimodal AI interprets the operative state, constrained robotics enforces safe execution, and digital twins enable rehearsal and iterative refinement (74). The key conceptual shift is that personalization is no longer limited to predicting patient risk or recognizing anatomy. Instead, personalization becomes operational only when those predictions and interpretations are transformed into changes in motion planning, safety margins, control logic, and autonomy level.

This perspective also helps clarify the role of vision-language-robotics in ophthalmic surgery. The value of VLMs and MLLMs does not lie simply in their ability to describe what is seen, but in their ability to bridge perception and action (75). In future systems, real-time surgical microscope video, iOCT, instrument tracking, and robot-state information could be interpreted jointly to provide feedback on surgical phase, tissue-interface recognition, instrument-tissue proximity, trajectory deviation, and potential safety violations. If linked appropriately to the robotic control stack, they may support context-aware supervision, adaptive constraint updating, procedural checklists, and intelligent safety gating. Likewise, the value of physical AI in ophthalmology lies not in replacing the surgeon, but in enabling structured robot behavior that is anatomically grounded, safety-aware, and responsive to both patient-specific geometry and intraoperative context.

Discussion

This review argues that the central issue in ophthalmic robotic surgery is not simply whether a pivot or RCM can be enforced, but how that constraint is incorporated into task formulation, motion generation, and execution. Across both anterior- and posterior-segment applications, robotic quality depends not only on positional accuracy, but also on whether the executed motion preserves the geometric and directional structure that defines surgical quality. In ophthalmic surgery, this distinction is especially important because the workspace is small, tissue tolerance is limited, and many clinically meaningful maneuvers are highly structured (16,59). The more relevant question is whether the robot preserves the intended path structure, motion smoothness, and safety logic under a strict entry-site constraint (5).

Pivot-constrained control is often framed as a kinematic or inverse-kinematics problem, and this is clearly necessary (69), but the reviewed literature suggests that successful ophthalmic robotic assistance cannot be reduced to constraint satisfaction alone (13,30). Controllers that appear mathematically acceptable may still differ substantially in their ability to preserve circularity, tangential progression, depth regulation, or surface-relative stability. Future comparative studies should therefore move beyond generic tip error and feasibility metrics and include evaluation of task-geometry preservation, motion continuity, and force-related safety.

The review further suggests that the future of ophthalmic robotics is unlikely to lie in unrestricted end-to-end autonomy. A more plausible direction is layered integration, in which physically grounded control, pivot-aware geometry, and task-structured reference generation remain the core of the system, while AI contributes through phase recognition, visual interpretation, semantic guidance, and safety-aware supervision. In this framework, AI becomes meaningful only when it changes robot behavior in a physically interpretable way (1). Digital twins may also become increasingly important by linking preoperative rehearsal, reachability analysis, intraoperative updating, and postoperative refinement within a patient-specific framework (72).

Several limitations remain in the current field. Standardized benchmark tasks are still limited, integrated task-level comparison is uncommon, and force-aware or tissue-aware evaluation remains underdeveloped in many applications. This review also has limitations: it is not a formal meta-analysis, the diversity of platforms and procedures limits direct quantitative synthesis, and the illustrative simulations were intended as conceptual demonstrations rather than experimental validation. Nevertheless, several practical implications emerge. Ophthalmic robotic research should increasingly evaluate preservation of structured task geometry, compare controller behavior under clinically meaningful motion classes, and prioritize physically grounded, interpretable, and safety-bounded assistance over generic autonomy claims.

Conclusions

Pivot-constrained motion is a foundational requirement in ophthalmic robotic surgery, but it does not by itself guarantee high-quality robotic assistance. The key issue is how pivot-aware constraints are incorporated into task definition, reference generation, structured motion preservation, safety enforcement, and surgeon-centered execution. Future progress will likely depend on moving beyond generic constrained tracking toward task-structured, anatomy-aware, and physically grounded robotic assistance. In both anterior- and posterior-segment surgery, clinically relevant performance is better understood through structured motion classes such as circular movement, tangential retinal surface following, controlled lifting, and phase-specific manipulation than through unconstrained trajectories alone. Accordingly, standardized pivot-aware benchmarks, comparative control studies, and explicit evaluation of task-geometry preservation are needed. The next major advance in ophthalmic robotic surgery will likely be a layered hybrid architecture that integrates personalized geometry, physical AI, digital twins, and vision-language-robotics to deliver patient-specific, phase-aware, pivot-conscious assistance that remains interpretable, controllable, and clinically trustworthy.

Acknowledgments

All figures are original works by the authors. Selected detailed illustrations were prepared with the assistance of AI-based image generation tools, and no third-party copyrighted material was used.

Footnote

Reporting Checklist: The authors have completed the Narrative Review reporting checklist. Available at https://aes.amegroups.com/article/view/10.21037/aes-2026-0028/rc

Peer Review File: Available at https://aes.amegroups.com/article/view/10.21037/aes-2026-0028/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://aes.amegroups.com/article/view/10.21037/aes-2026-0028/coif). T.K.Y. received a consulting fee from MediWhale. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Poh SSJ, Sia JT, Yip MYT, et al. Artificial Intelligence, Digital Imaging, and Robotics Technologies for Surgical Vitreoretinal Diseases. Ophthalmol Retina 2024;8:633-45. [Crossref] [PubMed]
2. de Smet MD, Naus GJL, Faridpooya K, et al. Robotic-assisted surgery in ophthalmology. Curr Opin Ophthalmol 2018;29:248-53. [Crossref] [PubMed]
3. Chatzimichail E, Feltgen N, Motta L, et al. Transforming the future of ophthalmology: artificial intelligence and robotics' breakthrough role in surgical and medical retina advances: a mini review. Front Med (Lausanne) 2024;11:1434241. [Crossref] [PubMed]
4. Ahronovich EZ, Simaan N, Joos KM. A Review of Robotic and OCT-Aided Systems for Vitreoretinal Surgery. Adv Ther 2021;38:2114-29. [Crossref] [PubMed]
5. Davila A, Colan J, Hasegawa Y. Real-time inverse kinematics for robotic manipulation under remote center-of-motion constraint using memetic evolution. J Comput Des Eng 2024;11:248-64.
6. Roizenblatt M, Edwards TL, Gehlbach PL. Robot-assisted vitreoretinal surgery: current perspectives. Robot Surg 2018;5:1-11. [Crossref] [PubMed]
7. Patel N, Urias M, Ebrahimi A, et al. Force-based Control for Safe Robot-assisted Retinal Interventions: In Vivo Evaluation in Animal Studies. IEEE Trans Med Robot Bionics 2022;4:578-87. [Crossref] [PubMed]
8. Mishra K, Leng T. Artificial intelligence and ophthalmic surgery. Curr Opin Ophthalmol 2021;32:425-30. [Crossref] [PubMed]
9. Qiu Y, Chen X, Wu X, et al. Embodied artificial intelligence in ophthalmology. NPJ Digit Med 2025;8:351. [Crossref] [PubMed]
10. Sayols N, Hernansanz A, Sozzi A, et al. Dynamic Global/Local multi-layer motion planner architecture for autonomous Cognitive Surgical Robots. Robot Auton Syst 2024;180:104758.
11. Bian Y, Zhao J, Li J, et al. A class of spatial remote center-of-motion mechanisms and its forward kinematics. Robotica 2023;41:885-99.
12. Jian X, Wu B, Song Y, et al. Performing task automation for surgical robot: A spatial-temporal varying primal-dual neural network with guided obstacle avoidance and null space optimization. Expert Syst Appl 2025;273:126780.
13. Colan J, Davila A, Fozilov K, et al. A Concurrent Framework for Constrained Inverse Kinematics of Minimally Invasive Surgical Robots. Sensors (Basel) 2023;23:3328. [Crossref] [PubMed]
14. Nasseri MA, Gschirr P, Eder M, et al. Virtual fixture control of a hybrid parallel-serial robot for assisting ophthalmic surgery: An experimental study. In: 5th IEEE RAS/EMBS International Conference on Biomedical Robotics and Biomechatronics. IEEE; 2014:732-8.
15. Stanev D, Moustakas K. Modeling musculoskeletal kinematic and dynamic redundancy using null space projection. PLoS One 2019;14:e0209171. [Crossref] [PubMed]
16. He C, Patel N, Iordachita I, et al. Enabling Technology for Safe Robot-Assisted Retinal Surgery: Early Warning for Unsafe Scleral Force. IEEE Int Conf Robot Autom 2019;2019:3889-94. [Crossref] [PubMed]
17. Qin W, Zhang H, Fan Z, et al. A shared control method for teleoperated robot using artificial potential field. Adv Intell Syst 2024;6:2300814.
18. Yuan Y, Sun W. Continuously shaping prioritized Jacobian approach for hierarchical optimal control with task priority transition. IEEE Trans Robot 2025;41:1639-56.
19. Wilde N, Smith SL, Alonso-Mora J. Scalarizing multi-objective robot planning problems using weighted maximization. IEEE Robot Autom Lett 2024;9:2503-10.
20. Kuo CH, Dai JS. Kinematics of a fully-decoupled remote center-of-motion parallel manipulator for minimally invasive surgery. J Med Devices 2012;6:021008.
21. Chiaverini S, Siciliano B, Egeland O. Review of the damped least-squares inverse kinematics with experiments on an industrial robot manipulator. IEEE Trans Control Syst Technol 1994;2:123-34.
22. Zheng Y, Yang X, Mo B, et al. Evaluation of the Hand Motion and Peeling Force in Inner Limiting Membrane Peeling. Transl Vis Sci Technol 2023;12:32. [Crossref] [PubMed]
23. Birch J, Da Cruz L, Rhode K, et al. Trocar localisation for robot-assisted vitreoretinal surgery. Int J Comput Assist Radiol Surg 2024;19:191-8. [Crossref] [PubMed]
24. Pisla D, Birlescu I, Crisan N, et al. Singularity analysis and geometric optimization of a 6-DOF parallel robot for SILS. Machines 2022;10:764.
25. Zhou M, Yu Q, Huang K, et al. Towards robotic-assisted subretinal injection: A hybrid parallel-serial robot system design and preliminary evaluation. IEEE Trans Ind Electron 2019;67:6617-28.
26. Zhang W, Wang Z, Ma K, et al. State of the art in movement around a remote point: a review of remote center of motion in robotics. Frontiers of Mechanical Engineering 2024;19:14.
27. Smits J, Reynaerts D, Vander Poorten E. Synthesis and methodology for optimal design of a parallel remote center of motion mechanism: Application to robotic eye surgery. Mech Mach Theory 2020;151:103896.
28. Wang Z, Zhang W, Ding X. A family of RCM mechanisms: Type synthesis and kinematics analysis. Int J Mech Sci 2022;231:107590.
29. Chen CW, Francone AA, Gerber MJ, et al. Semiautomated optical coherence tomography-guided robotic surgery for porcine lens removal. J Cataract Refract Surg 2019;45:1665-9. [Crossref] [PubMed]
30. Yang Y, Jiang Z, Yang Y, et al. Safety control method of robot-assisted cataract surgery with virtual fixture and virtual force feedback. J Intell Robot Syst 2020;97:17-32.
31. Shin C, Gerber MJ, Lee YH, et al. Semi-Automated Extraction of Lens Fragments via a Surgical Robot Using Semantic Segmentation of OCT Images with Deep Learning - Experimental Results in ex vivo Animal Model. IEEE Robot Autom Lett 2021;6:5261-8. [Crossref] [PubMed]
32. Lin C, Guang C, Zheng Y, et al. Preliminary evaluation of a novel vision-guided hybrid robot system for capsulotomy in cataract surgery☆☆. Displays 2022;74:102262.
33. Varga B, Poncelet M. A Shared Control Approach to Robot-Assisted Cataract Surgery Training for Novice Surgeons. Sensors (Basel) 2025;25:5165. [Crossref] [PubMed]
34. Chen X, Liu J, Liang H, et al. Digitalization of surgical features improves surgical accuracy via surgeon guidance and robotization. NPJ Digit Med 2025;8:497. [Crossref] [PubMed]
35. Ullrich F, Michels S, Lehmann D, et al. Assistive Device for Efficient Intravitreal Injections. Ophthalmic Surg Lasers Imaging Retina 2016;47:752-62. [Crossref] [PubMed]
36. de Smet MD, Meenink TC, Janssens T, et al. Robotic Assisted Cannulation of Occluded Retinal Veins. PLoS One 2016;11:e0162037. [Crossref] [PubMed]
37. Gonenc B, Chae J, Gehlbach P, et al. Towards Robot-Assisted Retinal Vein Cannulation: A Motorized Force-Sensing Microneedle Integrated with a Handheld Micromanipulator †. Sensors (Basel) 2017;17:2195. [Crossref] [PubMed]
38. Gijbels A, Smits J, Schoevaerdts L, et al. In-Human Robot-Assisted Retinal Vein Cannulation, A World First. Ann Biomed Eng 2018;46:1676-85. [Crossref] [PubMed]
39. Charreyron SL, Boehler Q, Danun AN, et al. A Magnetically Navigated Microcannula for Subretinal Injections. IEEE Trans Biomed Eng 2021;68:119-29. [Crossref] [PubMed]
40. Mach K, Wei S, Kim JW, et al. OCT-guided Robotic Subretinal Needle Injections: A Deep Learning-Based Registration Approach. Proceedings (IEEE Int Conf Bioinformatics Biomed) 2022;2022:781-6. [Crossref] [PubMed]
41. Cehajic-Kapetanovic J, Xue K, Edwards TL, et al. First-in-Human Robot-Assisted Subretinal Drug Delivery Under Local Anesthesia. Am J Ophthalmol 2022;237:104-13. [Crossref] [PubMed]
42. Yang K, Jin X, Wang Z, et al. Robot-assisted subretinal injection system: development and preliminary verification. BMC Ophthalmol 2022;22:484. [Crossref] [PubMed]
43. Ladha R, Meenink T, Smit J, et al. Advantages of robotic assistance over a manual approach in simulated subretinal injections and its relevance for gene therapy. Gene Ther 2023;30:264-70. [Crossref] [PubMed]
44. Zhang P, Kim JW, Gehlbach P, et al. Autonomous Needle Navigation in Retinal Microsurgery: Evaluation in ex vivo Porcine Eyes. IEEE Int Conf Robot Autom 2023;2023:4661-7. [Crossref] [PubMed]
45. Chen Y, Guo X, Ye X, et al. A soft micron accuracy robot design and clinical validation for retinal surgery. Microsyst Nanoeng 2025;11:170. [Crossref] [PubMed]
46. Esfandiari M, Kim JW, Zhang P, et al. Bimanual Robotic Eye Manipulation Using Adaptive Sclera Force Control: Towards Safe Retinal Vein Cannulation. IEEE Trans Med Robot Bionics 2025;7:1499-512. [Crossref] [PubMed]
47. Bian GB, Deng Y, Li Z, et al. Autonomous robotic intraocular surgery for targeted retinal injections. Sci Robot 2026;11:eadx7359. [Crossref] [PubMed]
48. Lian Z, Xu A, Li Y, et al. Robotic augmentation of surgical skills in subretinal injection: a novel semiautomated system. Int J Surg 2025; [Epub ahead of print]. [Crossref] [PubMed]
49. Li Y, Wu S, Shi G. Research and Realization of an OCT-Guided Robotic System for Subretinal Injections. Actuators 2026;15:53.
50. Uneri A, Balicki MA, Handa J, et al. New Steady-Hand Eye Robot with Micro-Force Sensing for Vitreoretinal Surgery. Proc IEEE RAS EMBS Int Conf Biomed Robot Biomechatron 2010;2010:814-9. [Crossref] [PubMed]
51. Maclachlan RA, Becker BC, Tabarés JC, et al. Micron: an Actively Stabilized Handheld Tool for Microsurgery. IEEE Trans Robot 2012;28:195-212. [Crossref] [PubMed]
52. Rahimy E, Wilson J, Tsao TC, et al. Robot-assisted intraocular surgery: development of the IRISS and feasibility studies in an animal model. Eye (Lond) 2013;27:972-8. [Crossref] [PubMed]
53. Yang S, MacLachlan RA, Riviere CN. Toward Automated Intraocular Laser Surgery Using a Handheld Micromanipulator. Rep U S 2014;2014:1302-7. [Crossref] [PubMed]
54. Yang S, MacLachlan RA, Riviere CN. Manipulator Design and Operation for a Six-Degree-of-Freedom Handheld Tremor-Canceling Microsurgical Instrument. IEEE ASME Trans Mechatron 2015;20:761-72. [Crossref] [PubMed]
55. Wilson JT, Gerber MJ, Prince SW, et al. Intraocular robotic interventional surgical system (IRISS): Mechanical design, evaluation, and master-slave manipulation. Int J Med Robot 2018;14: [Crossref] [PubMed]
56. Edwards TL, Xue K, Meenink HCM, et al. First-in-human study of the safety and viability of intraocular robotic surgery. Nat Biomed Eng 2018;2:649-56. [Crossref] [PubMed]
57. Ebrahimi A, He C, Patel N, et al. Sclera Force Control in Robot-assisted Eye Surgery: Adaptive Force Control vs. Auditory Feedback. Int Symp Med Robot 2019;2019:10. [Crossref] [PubMed]
58. Maberley DAL, Beelen M, Smit J, et al. A comparison of robotic and manual surgery for internal limiting membrane peeling. Graefes Arch Clin Exp Ophthalmol 2020;258:773-8. [Crossref] [PubMed]
59. Urias MG, Patel N, Ebrahimi A, et al. Robotic Retinal Surgery Impacts on Scleral Forces: In Vivo Study. Transl Vis Sci Technol 2020;9:2. [Crossref] [PubMed]
60. Zhou D, Kimura S, Takeyama H, et al. Eye Explorer: A robotic endoscope holder for eye surgery. Int J Med Robot 2021;17:1-13. [Crossref] [PubMed]
61. Zhou D, Takeyama H, Nakao S, et al. Real-time fundus reconstruction and intraocular mapping using an ophthalmic endoscope. Int J Med Robot 2023;19:e2496. [Crossref] [PubMed]
62. Yamamoto S, Kuroki Y, Ide T, et al. Customization of a passive surgical support robot to specifications for ophthalmic surgery and preliminary evaluation. Jpn J Ophthalmol 2023;67:637-44. [Crossref] [PubMed]
63. Yamamoto S, Taniguchi Y, Kuroki Y, et al. Investigation of hand tremor suppression by a customized passive surgical support robot during internal limiting membrane peeling. Retina 2024;44:2091-7. [Crossref] [PubMed]
64. Li Y, Wu S, Fan J, et al. Design and analysis of a spatial 2r1t remote center of motion mechanism for a subretinal surgical robot. Actuators 2024;13:124.
65. Ashikaga K, Sakanishi Y, Hirakata T, et al. Robot-Assisted Peripheral Retinal Visualization during Vitrectomy without Scleral Depression: A First-In-Human Comparative Study. Ophthalmol Sci 2026;6:101105. [Crossref] [PubMed]
66. Calusdian J, Yun X. A simple and highly portable MATLAB interface for learning robotics. SN Appl Sci 2019;1:890.
67. Trochimczuk R, Wolniakowski A, Ostaszewski M, et al. Experimental Evaluation of UR5e Collaborative Robot Force Control in Low-Force Applications. Sensors (Basel) 2026;26:1709. [Crossref] [PubMed]
68. Kim BK, Mun SJ, Lee DG, et al. Chung's swing technique: a new technique for small-incision lenticule extraction. BMC Ophthalmol 2016;16:154. [Crossref] [PubMed]
69. Menoth Mohan D, Shirinzadeh B, Smith J, et al. A review of recent methods and applications of virtual fixtures in robot-assisted surgery. Comput Biol Med 2025;199:111294. [Crossref] [PubMed]
70. Wang G, Bai L, Wang J, et al. EndoChat: Grounded multimodal large language model for endoscopic surgery. Med Image Anal 2026;107:103789. [Crossref] [PubMed]
71. Hao P, Wang H, Yang G, et al. Enhancing Visual Reasoning With LLM-Powered Knowledge Graphs for Visual Question Localized-Answering in Robotic Surgery. IEEE J Biomed Health Inform 2025;29:9027-40. [Crossref] [PubMed]
72. Asciak L, Kyeremeh J, Luo X, et al. Digital twin assisted surgery, concept, opportunities, and challenges. NPJ Digit Med 2025;8:32. [Crossref] [PubMed]
73. Batool Z, Wu Q, Li J, et al. Digital twin-assisted surgery: technological architecture, integration across surgical phases, implementation challenges and future directions. Int J Surg 2026; [Epub ahead of print]. [Crossref] [PubMed]
74. Kyeremeh J, Asciak L, Blackmur JP, et al. Digital twins assisted surgery: A conceptual framework for transforming surgical training and navigation. Surgeon 2025;S1479-666X(25)00143-X. [Epub ahead of print]. [Crossref] [PubMed]
75. Chen F, Chen H, Yu T, et al. AI-Driven Revolution of Medical Robotics Across Surgical Innovation, Rehabilitation Intelligence, and Multimodal Healthcare Delivery. MedComm (2020) 2026;7:e70597. [Crossref] [PubMed]