Motor learning with force feedback instruments among laparoscopic surgeons and residents Authors: Alec G. Moore, Elliot L. Servais, James Wall, Shreya Purohit, Mark Garibaldi Published: 2026 ABSTRACT Robotic-assisted surgery (RAS) enhances precision and ergonomics but lacks haptic feedback, limiting surgeons’ ability to regulate force and develop efficient sensorimotor control. This study evaluated whether force feedback improves motor learning underlying robotic control among robotically naïve surgeons. Twenty-eight robot-naïve participants (14 residents, 14 laparoscopic surgeons) completed four simulated surgical tasks with ex-vivo porcine models on the da Vinci 5 system. Participants were assigned to Force Feedback (FFB) or non-Force Feedback (NOF) groups and completed repeated trials. Outcomes included applied force, technical performance, system control, cognitive workload, gaze, and operational time. FFB significantly reduced applied force across tasks over time by 24% (2.6 N vs. 3.4 N, p = .002) with greater reductions during force-sensitive tasks and among experienced surgeons. Participants using FFB demonstrated greater improvements in robotic control (GEARS; p = .037) with significant gains over time (p < .001), whereas NOF showed minimal change. Residents with NOF exhibited higher clutch frequency compared to FFB (p = .001), indicating less efficient system control over time. FFB reduced cognitive load based on smaller pupil diameter (3.7 mm vs. 4.1 mm, p = .021). Task efficiency improved with FFB (− 130 s, p = .030) without evidence of a speed–accuracy tradeoff. Gaze analyses indicated that brief fixation on force feedback displays was associated with lower applied force (p = .001). Findings suggest that Force Feedback instruments accelerate early skill acquisition as part of the learning curve associated with robotic control among surgeons with no prior robotic experience. Further research is needed to determine whether, how, and why the behavioral advantages observed with FFB translate to meaningful improvements in surgical performance and patient outcomes. Introduction Robotic assisted surgery (RAS) has transformed minimally invasive surgery by enhancing dexterity, precision, and visualization across multiple surgical specialties [1, 2]. Advances such as improved depth perception, intuitive user interfaces, and wristed instrumentation have been shown to reduce physical and cognitive workload while improving surgeon performance [3, 4]. Despite these benefits, a persistent limitation of RAS has been the absence of haptic feedback [5]. Haptic feedback can encompass a sense of touch (vibrotactile) or resistance (force feedback), both of which are integral to sensorimotor control in open and laparoscopic surgery [6, 7]. Force feedback (or kinesthetic feedback) provides proprioceptive information that is intended to subconsciously enable the development of internal models that are neural representations enabling the prediction and regulation of human motion including tool use [8, 9]. The sensation afforded by force feedback may reduce the learning curve with control of robotic surgical systems. Most systems employ a leader-follower teleoperation architecture in which hand movements with the controllers are translated into movements of robotic instruments and the endoscope. Mastery of robotic control involves the efficient use of the surgeon console workspace, which includes surgeons’ ability to maintain a small working volume of the instruments while coordinating the position of the hand controllers [10]. This coordination enables surgeons to grasp, dissect, suture, tie knots, and manipulate tissue. Positioning the hand controllers involves clutching such that the instruments are frozen while the surgeon can independently move the controllers into a new, ideal position that may enable greater range of motion. Effective clutch use is considered a marker of advanced robotic skill. In the absence of force feedback that has historically been unavailable through robotic surgical systems, surgeons may rely on compensatory strategies such as visual estimations of force also known as “visuohaptics” [11, 12] or interactions with artificial motion boundaries, which may prolong motor learning and contribute to excessive force application. High levels of applied forces have been associated with increased risk of tissue damage and adverse outcomes [3, 4, 13, 14]. Robotically naïve surgeons may therefore benefit disproportionately from force feedback with learning and performing to control robotic instruments intended for surgery. Early studies of force feedback in RAS have demonstrated reductions in applied force during pre-clinical and clinical use among expert surgeons [14], as well as indications of accelerated early learning with suturing among fellows [13]. Behavioral studies focused on motor learning have shown the relative influence of haptic guidance and haptic error amplification with force feedback on skill acquisition among novice surgeons [15]. There is an opportunity to evaluate the effects of force feedback on robotic naïve trainees including residents earlier in the training pathway, as well as experienced laparoscopic surgeons, neither of whom have prior exposure to kinesthetic feedback rendered through Force Feedback instruments. In addition to improving control of robotic instruments, force feedback may reduce extraneous cognitive load, a known barrier to skill acquisition in robotic surgery; thereby facilitating more efficient learning and performance with RAS [5, 15, 16]. The objective of this study was to evaluate the effect of force feedback on motor learning underlying robotic control among robotically naïve surgeons. Motor learning denotes the development of neural pathways associated with precise and efficient control of movements which encompass force modulation [17]. In this study, motor learning refers to the learning curve that is characterized by performance variables as a function of time, including mean force applied, surgical technical skills, system control, cognitive load, and operational time. The authors hypothesized that (1) force feedback would enhance motor learning underlying robotic control compared to no force feedback, and (2) residents and laparoscopic surgeons using force feedback would demonstrate better robotic surgical performance comparable to those without force feedback within their cohorts. This research uniquely investigates the influence of novel haptic technology intended for robotic surgery on the learning curve among residents and laparoscopic surgeons without prior experience using robotic surgical systems nor prior exposure to kinesthetic feedback as rendered through Force Feedback instruments. Materials and methods Study subjects Twenty-eight participants without prior robotic surgery experience were enrolled. Fourteen were post graduate year 3 (PGY3) surgical residents, and fourteen were experienced laparoscopic surgeons. Participant sample distribution is summarized in Table 1. Table 1Summary of sample distributionFFBNOFPGY3 Residents77 20–29 Years Old33 30–39 Years Old44Laparoscopic Surgeons77 30–49 Years Old23 50–59 Years Old33 60–69 Years Old21Right hand dominant1213Ambidextrous21Average Years of Clinical Experience11.210.5 Table 1Summary of sample distributionFFBNOFPGY3 Residents77 20–29 Years Old33 30–39 Years Old44Laparoscopic Surgeons77 30–49 Years Old23 50–59 Years Old33 60–69 Years Old21Right hand dominant1213Ambidextrous21Average Years of Clinical Experience11.210.5 Summary of sample distribution Study design This was a 2 × 2 × 4 × 5 mixed model, repeated measures study. Participants were stratified by surgical expertise and assigned to either a Force Feedback (FFB, n = 14) or non-Force Feedback (NOF, n = 14) cohort. All participants used robotic instruments equipped with integrated force sensors. Four simulated clinical tasks were performed using ex vivo porcine tissue models. Participants completed a baseline trial with force feedback disabled (Trial 0), followed by four experimental trials (Trials 1–4) performed with force feedback enabled (FFB cohort) or disabled (NOF cohort). Study variables are summarized in Table 2. Before the baseline trial, participants were given standardized product training. This product training and Trials 0–2 were completed on one day. After a 12-hour decay period, participants completed Trials 3–4 the following day. Potential interference from system and instrument use with real-world cases was controlled by requiring adherence to a two-day, in-person study protocol. The study protocol was approved by the WIRB-Copernicus Group (WCG) institutional review board, and written informed consent was obtained from all participants. Table 2Summary of study variablesFactorLevelsSurgical Expertise2 groups (Novice or Expert)Force Feedback Cohort2 instrument types (FFB or NOF)Task4 simulated clinical tasksTrial5 trials (Baseline Trial 0 + Trials 1–4) Table 2Summary of study variablesFactorLevelsSurgical Expertise2 groups (Novice or Expert)Force Feedback Cohort2 instrument types (FFB or NOF)Task4 simulated clinical tasksTrial5 trials (Baseline Trial 0 + Trials 1–4) Summary of study variables System setup and instruments Simulated clinical tasks were performed using the da Vinci 5 surgical system (Intuitive Surgical, Sunnyvale, CA, USA). The da Vinci 5 surgical system introduces force feedback that transmits measured instrument tip force to the participant’s hand controllers. Force feedback instruments (ProGrasp Forceps, Large Needle Driver, Fenestrated Bipolar Forceps, Maryland Bipolar Forceps, Monopolar Curved Scissors) were similar in form and function to standard robotic instruments with the addition of tip mounted force sensors measuring push and pull forces along the X, Y, and Z axes. Force feedback was disabled for all participants during baseline trials and enabled only for the FFB cohort during Trials 1–4. The system features three levels of force feedback sensitivity that could be separately rendered: Low, Medium, High. The force relayed to the hand controls is scaled by this sensitivity setting at a portion of the actual forces detected by the instrument. For the purposes of the study, the force sensitivity level was preset at the default sensitivity setting (Medium) for all trials within the FFB cohort. The system includes a computer-controlled feature that translates large hand movements at the console into smaller, highly precise micro-movements referred to as motion scaling. For the purposes of this study, motion scaling was preset at Normal (2:1) for both cohorts. Force feedback was blinded to NOF participants while using identical instruments with force feedback disabled. A Monopolar Curved Scissors instrument did not render force feedback but the instrument was available to both groups. A 30-degree endoscope with optional Firefly fluorescence imaging was used in both cohorts to identify critical structures in the cholecystectomy task. Monopolar Energy was set to 20 Watts spray on coag and 0 Watts on cut, and Bipolar Energy was turned off. Participants were instructed not to change any of these settings. Tissue models Clinical tasks were conducted using ex vivo porcine tissue models (Fig. 1). The models included a porcine biliary system, stomach, and kidneys (Animal Technologies, Tyler, Texas). Fig. 1Clinical tasks and corresponding ex vivo porcine models Clinical tasks and corresponding ex vivo porcine models Training All participants received standardized product training that covered system operations, ergonomics, instrument control, camera navigation, energy use, and Firefly imaging without exposure to force feedback functionality. Task specific procedural training was provided through standardized video walkthroughs that included pre-recorded videos displaying the endoscopic view of an expert robotic surgeon completing the tasks. Participants randomly assigned to the FFB cohort received additional product training on Force Feedback instruments following baseline testing, which included practice dissecting and suturing tissue. The NOF cohort received matched practice without force feedback following baseline testing. Both cohorts completed only one repetition with practice dissecting and suturing tissue. Study procedure Participants repeated four simulated clinical tasks: small bowel anastomosis, stomach defect suturing, cholecystectomy, and renal vessel dissection (Table 3). Each task was completed once in each trial with task order counterbalanced using a Latin square design. Table 3Description of simulated clinical tasks, sub-tasks, and instrumentsSimulated Clinical TaskDescription of tasks and sub-tasksInstrumentsPartial Small Bowel Anastomosis TaskCompleting this task required a half circumference enterotomy using a running 3 − 0 Vicryl RB 1 suture within 25 minSub-tasks:• Suspend the bowel vertically and identify the resection point• Suture with a running stitch to reanastomoses the bowel• Arm 1: Force Feedback Large Needle Drivers• Arm 2: 30-degree (down) endoscope• Arm 3: Force Feedback Large Needle Drivers• Arm 4: Force Feedback ProGraspStomach Suturing TaskCompleting this task required one-handed suturing with interrupted stitch across a 1.5 cm defect in the stomach with 3 − 0 Vicryl SH sutures and a running stitch across a 4 cm defect with 2 –0 V Loc 180 suture in the stomach within 15 min.• Arm 1: Force Feedback Large Needle Driver• Arm 2: 30-degree (down) endoscope• Arm 3; Force Feedback Large Needle Driver• Arm 4: Force Feedback ProGraspCholecystectomy TaskCompleting this task required a complete gallbladder removal within 25 min. The cystic structures were clipped by the researcher at the request of the participant; monopolar energy was set to 20 W spray mode, and Firefly imaging was available following Indocyanine green injection.Sub-tasks:• Locate the gallbladder and retract to expose the cystic duct and artery• Dissect out the cystic duct and artery• Clip and transect the cystic duct and artery• Dissect the gallbladder off the liver bed• Arm 1: Force Feedback Fenestrated Bipolar• Arm 2: 30-degree (down) endoscope• Arm 3: Monopolar Curved Scissors• Arm 4: Force Feedback ProGraspRenal Vessel Dissection TaskCompleting this task required dissection on kidneys to visualize the vessels within 5 min using limited monopolar energy at 20 W spray mode.• Arm 1: Force Feedback Fenestrated Bipolar• Arm 2: 30-degree (down) endoscope• Arm 3: Monopolar Curved Scissors• Arm 4: Force Feedback ProGrasp Table 3Description of simulated clinical tasks, sub-tasks, and instrumentsSimulated Clinical TaskDescription of tasks and sub-tasksInstrumentsPartial Small Bowel Anastomosis TaskCompleting this task required a half circumference enterotomy using a running 3 − 0 Vicryl RB 1 suture within 25 minSub-tasks:• Suspend the bowel vertically and identify the resection point• Suture with a running stitch to reanastomoses the bowel• Arm 1: Force Feedback Large Needle Drivers• Arm 2: 30-degree (down) endoscope• Arm 3: Force Feedback Large Needle Drivers• Arm 4: Force Feedback ProGraspStomach Suturing TaskCompleting this task required one-handed suturing with interrupted stitch across a 1.5 cm defect in the stomach with 3 − 0 Vicryl SH sutures and a running stitch across a 4 cm defect with 2 –0 V Loc 180 suture in the stomach within 15 min.• Arm 1: Force Feedback Large Needle Driver• Arm 2: 30-degree (down) endoscope• Arm 3; Force Feedback Large Needle Driver• Arm 4: Force Feedback ProGraspCholecystectomy TaskCompleting this task required a complete gallbladder removal within 25 min. The cystic structures were clipped by the researcher at the request of the participant; monopolar energy was set to 20 W spray mode, and Firefly imaging was available following Indocyanine green injection.Sub-tasks:• Locate the gallbladder and retract to expose the cystic duct and artery• Dissect out the cystic duct and artery• Clip and transect the cystic duct and artery• Dissect the gallbladder off the liver bed• Arm 1: Force Feedback Fenestrated Bipolar• Arm 2: 30-degree (down) endoscope• Arm 3: Monopolar Curved Scissors• Arm 4: Force Feedback ProGraspRenal Vessel Dissection TaskCompleting this task required dissection on kidneys to visualize the vessels within 5 min using limited monopolar energy at 20 W spray mode.• Arm 1: Force Feedback Fenestrated Bipolar• Arm 2: 30-degree (down) endoscope• Arm 3: Monopolar Curved Scissors• Arm 4: Force Feedback ProGrasp Description of simulated clinical tasks, sub-tasks, and instruments Completing this task required a half circumference enterotomy using a running 3 − 0 Vicryl RB 1 suture within 25 min Sub-tasks: • Suspend the bowel vertically and identify the resection point • Suture with a running stitch to reanastomoses the bowel • Arm 1: Force Feedback Large Needle Drivers • Arm 2: 30-degree (down) endoscope • Arm 3: Force Feedback Large Needle Drivers • Arm 4: Force Feedback ProGrasp • Arm 1: Force Feedback Large Needle Driver • Arm 2: 30-degree (down) endoscope • Arm 3; Force Feedback Large Needle Driver • Arm 4: Force Feedback ProGrasp Completing this task required a complete gallbladder removal within 25 min. The cystic structures were clipped by the researcher at the request of the participant; monopolar energy was set to 20 W spray mode, and Firefly imaging was available following Indocyanine green injection. Sub-tasks: • Locate the gallbladder and retract to expose the cystic duct and artery • Dissect out the cystic duct and artery • Clip and transect the cystic duct and artery • Dissect the gallbladder off the liver bed • Arm 1: Force Feedback Fenestrated Bipolar • Arm 2: 30-degree (down) endoscope • Arm 3: Monopolar Curved Scissors • Arm 4: Force Feedback ProGrasp • Arm 1: Force Feedback Fenestrated Bipolar • Arm 2: 30-degree (down) endoscope • Arm 3: Monopolar Curved Scissors • Arm 4: Force Feedback ProGrasp (A) Metrics Across each trial the force applied, technical skills, system control, cognitive load, gaze, and operational time were measured to assess differences in surgeon learning. Force applied: The forces applied on tissues by the Force Feedback instruments during each task per trial were measured. The magnitude of forces was directly collected and stored in the system data log of the robotic system. Data included the force measures for the instrument tip on arm 1, sampled at 1 Hz during critical task segments. Computing the maximum forces per instrument was not possible due to the temporal aliasing effect of the sampling rate, thus we examined the 95th percentile to compare upper limits of force applied. Forces less than 0.5 N were excluded as 0.5 N is the typical minimum force threshold at which users of the robotic system can detect forces at the system hand controllers.Technical skills assessment: Global Evaluative Assessment of Robotic Skills (GEARS) is a validated instrument that evaluates technical proficiency in robotic-assisted surgery across six domains: Depth Perception, Bimanual Dexterity, Efficiency, Tissue Handling, Camera Control, and Robotic Control. This 5-point Likert scale distinguishes skill levels ranging from novice to expert [18]. Measurement of technical skills was based on all clinical tasks per trial.System control: The dV5 system internally logs system telemetry, such as the position and coordinates, speed and velocity, acceleration, direction and orientation, rotation rates, as well as the start and stop of actions and system errors. These data enable the characterization of a participant’s control with motions and actions associated with the input devices (controllers) and end effectors (instruments and endoscope).Cognitive load: Eye tracking was used to capture pupillometric data to assess cognitive load based on Mean Pupil Diameter (MPD). Pupillometric data capture is made possible with custom designed brackets that enable a camera mounted to the console to measure pupillometry while participants look through the eye holes. MPD offers continuous, objective measurement of cognitive load while avoiding hindsight bias and single-point limits of self-reported measures like the NASA-TLX [19]. MPD has been corroborated with Index of Pupillary Activity algorithm [20]. Ambient lighting conditions were controlled with fixed illumination within the room and endoscope.Gaze: Eye tracking can capture the Area of Interest (AOI) associated with the force gauges displayed from the arm pods from the viewer. AOI is a measure of visual attention that was based on the total amount of time spent looking inside the AOI boundary. The reticle was configured with a median angular accuracy of 2.6°. Data synchronization enabled gaze measurement by linking the participant’s gaze position, represented as a 2D pixel location in each video frame with its corresponding timestamp. This allowed the calculation of the three degrees of visual angle for each frame, represented as a circular area centered on the target gaze pixel.Operational time: The total amount of time a participant’s head was positioned in the surgeon console while actively controlling the instruments to finish the task. If a participant lifted their head from the console, the event was annotated then later removed from the total time on task. Measurement of time was recorded across each exercise. Force applied: The forces applied on tissues by the Force Feedback instruments during each task per trial were measured. The magnitude of forces was directly collected and stored in the system data log of the robotic system. Data included the force measures for the instrument tip on arm 1, sampled at 1 Hz during critical task segments. Computing the maximum forces per instrument was not possible due to the temporal aliasing effect of the sampling rate, thus we examined the 95th percentile to compare upper limits of force applied. Forces less than 0.5 N were excluded as 0.5 N is the typical minimum force threshold at which users of the robotic system can detect forces at the system hand controllers. Force applied: The forces applied on tissues by the Force Feedback instruments during each task per trial were measured. The magnitude of forces was directly collected and stored in the system data log of the robotic system. Data included the force measures for the instrument tip on arm 1, sampled at 1 Hz during critical task segments. Computing the maximum forces per instrument was not possible due to the temporal aliasing effect of the sampling rate, thus we examined the 95th percentile to compare upper limits of force applied. Forces less than 0.5 N were excluded as 0.5 N is the typical minimum force threshold at which users of the robotic system can detect forces at the system hand controllers. Technical skills assessment: Global Evaluative Assessment of Robotic Skills (GEARS) is a validated instrument that evaluates technical proficiency in robotic-assisted surgery across six domains: Depth Perception, Bimanual Dexterity, Efficiency, Tissue Handling, Camera Control, and Robotic Control. This 5-point Likert scale distinguishes skill levels ranging from novice to expert [18]. Measurement of technical skills was based on all clinical tasks per trial. Technical skills assessment: Global Evaluative Assessment of Robotic Skills (GEARS) is a validated instrument that evaluates technical proficiency in robotic-assisted surgery across six domains: Depth Perception, Bimanual Dexterity, Efficiency, Tissue Handling, Camera Control, and Robotic Control. This 5-point Likert scale distinguishes skill levels ranging from novice to expert [18]. Measurement of technical skills was based on all clinical tasks per trial. System control: The dV5 system internally logs system telemetry, such as the position and coordinates, speed and velocity, acceleration, direction and orientation, rotation rates, as well as the start and stop of actions and system errors. These data enable the characterization of a participant’s control with motions and actions associated with the input devices (controllers) and end effectors (instruments and endoscope). System control: The dV5 system internally logs system telemetry, such as the position and coordinates, speed and velocity, acceleration, direction and orientation, rotation rates, as well as the start and stop of actions and system errors. These data enable the characterization of a participant’s control with motions and actions associated with the input devices (controllers) and end effectors (instruments and endoscope). Cognitive load: Eye tracking was used to capture pupillometric data to assess cognitive load based on Mean Pupil Diameter (MPD). Pupillometric data capture is made possible with custom designed brackets that enable a camera mounted to the console to measure pupillometry while participants look through the eye holes. MPD offers continuous, objective measurement of cognitive load while avoiding hindsight bias and single-point limits of self-reported measures like the NASA-TLX [19]. MPD has been corroborated with Index of Pupillary Activity algorithm [20]. Ambient lighting conditions were controlled with fixed illumination within the room and endoscope. Cognitive load: Eye tracking was used to capture pupillometric data to assess cognitive load based on Mean Pupil Diameter (MPD). Pupillometric data capture is made possible with custom designed brackets that enable a camera mounted to the console to measure pupillometry while participants look through the eye holes. MPD offers continuous, objective measurement of cognitive load while avoiding hindsight bias and single-point limits of self-reported measures like the NASA-TLX [19]. MPD has been corroborated with Index of Pupillary Activity algorithm [20]. Ambient lighting conditions were controlled with fixed illumination within the room and endoscope. Gaze: Eye tracking can capture the Area of Interest (AOI) associated with the force gauges displayed from the arm pods from the viewer. AOI is a measure of visual attention that was based on the total amount of time spent looking inside the AOI boundary. The reticle was configured with a median angular accuracy of 2.6°. Data synchronization enabled gaze measurement by linking the participant’s gaze position, represented as a 2D pixel location in each video frame with its corresponding timestamp. This allowed the calculation of the three degrees of visual angle for each frame, represented as a circular area centered on the target gaze pixel. Gaze: Eye tracking can capture the Area of Interest (AOI) associated with the force gauges displayed from the arm pods from the viewer. AOI is a measure of visual attention that was based on the total amount of time spent looking inside the AOI boundary. The reticle was configured with a median angular accuracy of 2.6°. Data synchronization enabled gaze measurement by linking the participant’s gaze position, represented as a 2D pixel location in each video frame with its corresponding timestamp. This allowed the calculation of the three degrees of visual angle for each frame, represented as a circular area centered on the target gaze pixel. Operational time: The total amount of time a participant’s head was positioned in the surgeon console while actively controlling the instruments to finish the task. If a participant lifted their head from the console, the event was annotated then later removed from the total time on task. Measurement of time was recorded across each exercise. Operational time: The total amount of time a participant’s head was positioned in the surgeon console while actively controlling the instruments to finish the task. If a participant lifted their head from the console, the event was annotated then later removed from the total time on task. Measurement of time was recorded across each exercise. (B) Data Capture For the purposes of this study, all data related to force applied, system control, and operational time derive from the robotic system. All da Vinci® robotic systems contain a recording device (dVLogger) that continuously log telemetry and endoscopic video. All data related to cognitive load and gaze derive from an eye tracker (Neon Eye Tracking and Bare Metal Nest PCB, Pupil Labs GmbH, Berlin, Germany). The eye tracking camera was made possible with a custom designed bracket that is mounted to the viewer without interfering with the endoscopic view through the eye holes. All data related to the technical skills assessment derive from post-hoc scoring of surgeon performance presented from the endoscopic video of the system. All real-time annotations were made possible with a custom application. To enable post-processing, robotic system data were synchronized with the eye tracker and annotator. (C) Statistical Analysis All statistical analyses were completed using NCSS statistical software (NCSS 24.0.2 / 25.0.2 / 26.0.3. NCSS, LLC. Kaysville, Utah, USA. www.ncss.com). Omnibus mixed-model analysis of variance (ANOVA) was performed (condition x task x trial x experience) for each metric. Additional omnibus two-way mixed-model ANOVAs (condition x task, condition x experience), applicable post hoc tests or independent-samples t tests (for FFB v/s NOF condition), and paired sample t test were performed for processing data of baseline (Trial 0) and Trials 1–4. Where violations of normality occurred or data are categorically not normally distributed, appropriate nonparametric methods were leveraged: Friedman test for multivariate comparisons, and Aspin-Welch unequal-variance t tests. Nonparametric tests yielded identical significant comparisons as parametric tests. One-sided p values were calculated for all comparisons based on prior work by Servais et al. [13] indicating improvements in performance. For overall comparisons, a p value < 0.05 was considered statistically significant. For comparisons across exercises and error types, Bonferroni corrections were applied to the p values to account for Type 1 errors. There was a 14% data loss of video footage distributed among 16 participants across trials that was intended for scoring with GEARS. A Random Forest multiple imputation procedure was conducted using custom Python code and scikit learn (v.1.6.1) package to replace missing values associated with the GEARS dataset with realistic, predicted values as informed by known trends in peer reviewed literature [23]. Predictors included the participant, experimentatal conditions, trial, clinical task, and subscale. System Control analyses focused on clutch frequency per minute (based on system time-on-task) and the 95th percentile of workspace volume (interval ellipsoid) with left and right hands analyzed separately due to differences in instrument use; workspace was converted to cubic inches for interpretability. Force analyses examined the 95th percentile of applied force per trial, while Operational Time was analyzed for all tasks with variable duration. Cognitive Load analyses included only participants with valid pupillometric data across all attempts and were restricted to data collected before task time limits. Examination of baseline differences Before fully analyzing the data, applicable measures collected during Trial 0 (baseline) for each participant in the FFB and NOF cohorts were compared to examine whether pre-existing differences existed across the experimental and control groups. Initial comparisons revealed that preexisting differences existed with two subscales associated with GEARS (Robotic Control and Depth Perception) across cohorts and conditions during the Trial 0 baseline clinical tasks. The analysis of GEARS subscales exclusively focused on change scores. Inter-rater reliability of technical skill ratings Four independent raters with surgical expertise were blinded to cohort assignment through cropped and relabeled video recordings. Videos were randomly distributed, and each was independently scored by two experts against GEARS. The mean scores were used for analysis. The Interclass Correlation Coefficients indicated strong interrater reliability across dimensions and clinical tasks (ICC = 0.781). Thus, GEARS ratings were averaged across raters, and overall scores were computed by averaging across dimensions. Results Force applied The 95th percentile of force across tasks for each trial applied by residents and experienced laparoscopic surgeons was significantly lower with FFB including a 24% reduction in mean force (M = 2.6 N, SD = 2.0 with FFB vs. M = 3.4 N, SD = 2.6 with NOF, p = .002) as shown in Fig, 2 A. We did not observe an interaction effect in reduced force with FFB by task: 26% less force with FFB during the anastomosis task (M = 0.9 N, SD = 0.4 with FFB vs. M = 1.2 N, SD = 0.6 with NOF), 23% less force with FFB during the cholecystectomy task (M = 5.1 N, SD = 2.2 with FFB vs. M = 6.6 N, SD = 2.7 with NOF), 18% less force with FFB during the renal vessel dissection task (M = 2.7 N, SD = 0.8 with FFB vs. M = 3.3 N, SD = 1.8 with NOF), and 29% less force with FFB during the stomach suturing task (M = 1.6 N, SD = 0.5 with FFB vs. M = 2.3 N, SD = 0.8 with NOF). Technical skills scores For all trials and tasks combined, residents and experienced laparoscopic surgeons demonstrated significant improvement in performance with subscales over trials 1–4 (p < .008). These differences were stable across both groups due to practice effects. Specifically, residents and experienced laparoscopic surgeons demonstrated significant improvement in ‘Robotic Control’ over time when utilizing FFB (p < .001). Also, all participants with FFB demonstrated significant improvement in Robotic Control compared to NOF (p = .037) as shown in Fig. 2B (M = + 0.37, SD = 0.9 with FFB vs. M = + 0.02, SD = 0.7 relative change in scores from Trial 1). After collapsing all trials and tasks, participants did not significantly improve in Robotic Control unless they were using FFB. These enhancements were predominantly observed during the Anastomosis and Suturing tasks with FFB, where performance showed marginal or significant differences relative to the NOF group (p < .024). System control Clutching frequency among residents and expert laparoscopic surgeons was significantly, and positively, and moderately correlated to a larger workspace volume for the left (ρ = 0.47, p < .001) and right (ρ = 0.42, p < .001) hands. Residents with NOF clutched significantly more frequently (M = 1.8 CPM, SD = 1.4) than their resident counterparts with FFB (M = 1.0 CPM, SD = 0.9) or laparoscopic surgeons with FFB (M = 1.3 CPM, SD = 1.5) or without FFB (M = 1.3 CPM, SD = 1.2) particularly with the anastomosis and cholecystectomy tasks (p = .001) as shown in Fig. 2C, D, and E. Surgeon console workspace utilization was not significantly different between residents with FFB (LH: M=1878cm3, SD = 1354; RH: M=2154cm3, SD = 1330) and residents without FFB (LH: M=1899cm3, SD = 1580; RH: M=2555cm3, SD = 2111). Likewise for laparoscopic surgeons, workspace utilization with FFB (LH: M=2196cm3, SD = 2428; RH: M=3022cm3, SD = 3281) and without FFB (LH: M=2814cm3, SD = 2800; RH: M=3108cm3, SD = 3106) were not significantly different. Laparoscopic surgeons did utilize significantly more workspace volume than residents in both their left (p = .002) and right (p = .004) hands. Fig. 2A Average applied force across trials, tasks, and experience for FFB and NOF conditions (p < .050). B Technical skill (GEARS) ratings for Robotic Control for all ex vivo clinical task, trial and experience for FFB and NOF conditions (p < .050). C Average clutching frequency for all tasks, trials, and experience for FFB and NOF conditions (p < .050). D and E Total left hand and right hand workspace volume (cm3 95th percentile) for residents and experienced laparoscopic surgeons across trials and tasks for FFB and NOF conditions with Bonferroni-corrected paired-samples t-tests (p’s < 0.008) A Average applied force across trials, tasks, and experience for FFB and NOF conditions (p < .050). B Technical skill (GEARS) ratings for Robotic Control for all ex vivo clinical task, trial and experience for FFB and NOF conditions (p < .050). C Average clutching frequency for all tasks, trials, and experience for FFB and NOF conditions (p < .050). D and E Total left hand and right hand workspace volume (cm3 95th percentile) for residents and experienced laparoscopic surgeons across trials and tasks for FFB and NOF conditions with Bonferroni-corrected paired-samples t-tests (p’s < 0.008) Operational time Residents and experienced laparoscopic surgeons demonstrated no statistical difference in operational time across trials and a subset of tasks (Suturing, Cholecystectomy). Because Renal Vessel dissection had a fixed task duration, it was excluded from comparisons. The rate of improvement was stable across both groups due to practice effects. However, both groups with FFB completed the Anastomosis task approximately 2.1 min faster in Trials 1 to 4 compared to those in the NOF cohort (MD = 130 s, p = .030) as shown in Fig. 3. While the authors did not observe an interaction effect between the presence of FFB and experience (p = .193), the influence of FFB on operational time may be associated with clinical experience. Among expert laparoscopic surgeons with FFB, operational time was significantly lower across trials compared to those with NOF by a magnitude of 3.4 min with robotic anastomosis tasks (MD = 206 s, p = .020). The authors did not observe the same effect associated with residents. Improvements in each GEARS subscale significantly and negatively correlated with shorter anastomosis operative time suggesting no speed-to-accuracy tradeoff with improvements in performance with FFB (Bimanual Dexterity: ρ = − 0.43, Depth Perception: ρ = − 0.36, Efficiency: ρ = − 0.49, Force Sensitivity: ρ = − 0.51, Robotic Control: ρ = − 0.40, p < .001). Fig. 3A Total time taken to complete for each task across trial and experience level for FFB and NOF conditions (p < .050) A Total time taken to complete for each task across trial and experience level for FFB and NOF conditions (p < .050) Cognitive load Residents and experienced laparoscopic surgeons with FFB demonstrated significantly lower cognitive load across trials and tasks compared to those with NOF, as indicated by mean pupil diameter measurements (M = 3.7 mm, SD = 0.70 with FFB vs. M = 4.1 mm, SD=0.85 without FFB, p = .021) shown in Fig. 4A and B. This stable difference was observed with a fixed display brightness within the console. Experienced laparoscopic surgeons exhibited significantly lower cognitive load than residents regardless of trial and task (M = 3.6 mm, SD = 0.69 for laparoscopic surgeons vs. M = 4.3 mm, SD = 0.74 for residents, p < .001). Fig. 4A Average pupil diameter across each trial, task and experience level for FFB and NOF conditions (p < .050). B Average pupil diameter based on experience across each trial and task for FFB and NOF conditions (p < .050) A Average pupil diameter across each trial, task and experience level for FFB and NOF conditions (p < .050). B Average pupil diameter based on experience across each trial and task for FFB and NOF conditions (p < .050) Gaze Residents and experienced laparoscopic surgeons with FFB who dwelled at the force gauges in the arm pods (M = 2.3 s, SD = 3.2) applied 0.6 N less force than those who did not dwell on the gauges (p = .001). Also, mean force applied was significantly and negatively correlated with total duration of fixation on the gauges (ρ = − 0.16, p < .050) as shown in Fig. 5A and B. This suggests that participants who fixated on the gauges for longer applied less force. Fig. 5A Average applied force and total fixation duration on the force gauges across trials, task, and experience among participants with FFB (p < .050). B Image of eye tracking reticle representing the position of the participant dwell on the force gauge in the arm pod of the console viewer A Average applied force and total fixation duration on the force gauges across trials, task, and experience among participants with FFB (p < .050). B Image of eye tracking reticle representing the position of the participant dwell on the force gauge in the arm pod of the console viewer Discussion This study demonstrates that force feedback facilitates motor learning and improves robotic control in residents and experienced laparoscopic surgeons without prior robotic experience. Across GEARS subscales, both residents and experienced laparoscopic surgeons show greater improvement over time when using FFB compared to those with NOF. Enhanced robotic control with FFB is most consistently observed during the anastomosis task, while task specific effects are observed for suturing (residents) and cholecystectomy (laparoscopic surgeons). Importantly, improvements across trials are observed among participants who used FFB, indicating that force feedback may meaningfully accelerate skill acquisition in robotic teleoperation. Related to system control, residents with FFB maintain the smallest average workspace volume of groups, while residents without FFB leverage significantly more clutches to maintain a similar workspace volume. These findings suggest that force feedback influences the instrument motion characteristics particularly among residents with no prior robotic experience; thereby, allowing them to complete clinical tasks with less clutching associated with workspace management. This finding suggests that FFB may be facilitating motor learning related to robotic control of tools with multiple internal degrees of freedom [15]. In contrast, laparoscopic surgeons demonstrate no significant differences in clutching. This is attributed to habituation with larger hand-workspace volumes related to the fulcrum-effect of manipulating laparoscopic instruments [6]. Force feedback facilitates significantly lower forces applied at the instrument tips across tasks regardless of experience. This effect was most pronounced during tasks that require force modulation such as suturing. Relatedly, robotic control coincided with mean force applied further suggesting that force feedback may be processed subconsciously without interfering in task execution. This interpretation aligns with the concept that haptic feedback more directly supports implicit motor adaptation with tool use rather than conscious strategizing [21]. Visual interactions with FFB force gauges further contribute to force modulation. Participants who fixate for a brief duration on the force gauges apply significantly lower forces, while dwell time is negatively correlated with applied force. This suggests that even momentary visual reinforcement of force information can be integrated with haptic feedback to improve control with tool use [22]. Force feedback was associated with lower mean pupil diameter, suggesting a possible reduction in cognitive workload. While lighting conditions were consistent, these findings should be interpreted cautiously given the known physiologic and technical influences on pupillometry. Overall, reduced cognitive load is associated with improved robotic control. This finding supports the interpretation that force feedback can reduce extraneous cognitive demand which may be related to reduction in attentional demands associated with skill acquisition [5]. Force feedback was associated with modest improvements in efficiency, particularly among experienced laparoscopic surgeons. Participants completed anastomosis tasks faster with FFB, independent of training effects. These gains were not accompanied by declines in the quality of performance, suggesting no speed-accuracy tradeoff. While time on task alone was relatively similar across conditions, improvements in efficiency during the anastomosis task suggest that force feedback may be optimizing procedural flow. Limitations Although significant differences in pupil diameter were observed, the authors cannot necessarily attribute this uniquely and entirely to changes in cognitive load. The decision to use pupil-related measures was supported by the ability to objectively measure physiological responses without invasive instrumentation of each participant (e.g., encephalography), as well as the experimental control afforded in task consistency, and lighting conditions in the room at the immersive console. Future work could further explore new methodologies of objectively measuring cognitive load while performing clinical tasks from the robotic console. Additionally, while most video samples were scored against GEARS, technical challenges with data capture reduced access from 2,800 videos to 2,410 videos. The missing data impacted 14% of samples across 16 participants and each trial-task. Removal of incomplete data would prohibit balanced statistical analyses. To allow for comparisons to be drawn from the existing data, the authors performed a multiple imputation procedure informed by the existing data, following methodology from peer reviewed literature [23]. While this procedure preserved descriptive statistics of the existing data with an average deviation in mean by category of 0.01 on GEARS 5-point Likert-scales, it nevertheless introduces an opportunity for variance that may have been different than measured outcomes associated with a full dataset. Statistical analyses in this paper report one-sided p-values. While analyses are based on hypotheses that draw from prior works indicating the direction of changes of various measures (e.g. [13]), it nevertheless introduces the opportunity for missing statistical significance in the opposite direction, and where possible, could increase the likelihood for type I error. To ensure a feasible study while permitting the analysis of all reported measures, a priori power analysis was conducted with the anticipated Bonferroni corrections, and a one-sided p-value was determined to be acceptable based on the directional nature of the hypotheses. Future work could reduce the scope to focus on fewer metrics to improve the corrected power while allowing for two-sided p-value analysis without greatly increasing the required sample size. This study used simulated clinical tasks involving ex vivo porcine tissue that allowed for repeatable assessments of surgical performance. These tasks enabled the authors to measure differences in learning with FFB instruments. The findings cannot be used to infer how observed behavioral changes translate to clinical outcomes as the study did not involve measuring performance in surgical procedures. Future research should answer questions related to how and why the behavioral advantages associated with force feedback contribute to meaningful improvements in clinical outcomes. Conclusion The findings suggest that Force Feedback instruments accelerate early skill acquisition by enhancing motor learning underlying robotic control among residents and laparoscopic surgeons with no prior robotic experience. The findings are limited to a simulated training environment and should not be generalized to clinical settings. Further research is needed to determine whether, how, and why the behavioral advantages observed with force feedback translate to meaningful improvements in surgical performance and patient outcomes.