{"id":"f6032a14-0470-48b6-922a-565b0d0b531e","arxiv_id":"2506.21689","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"The optimal motion scaling for delayed telesurgery depends on the individual user and the latency level, and per-user Bayesian regression can predict it.","lead":"This paper tested whether slowing down a surgeon's hand movements, called motion scaling, helps when robotic surgery is delayed by network lag. It found that the best slowing amount differs by person and by delay, and built a personalized model to choose it.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed user-specific optima are extracted from an unvalidated quadratic surrogate; no out-of-sample test shows the recommended scaling factor actually outperforms a fixed scale.","rationale":"The reader's weakest assumption correctly identifies the quadratic iid-Gaussian response surface as a pivotal unvalidated assumption. My reading reinforces that concern: the paper's optimization pipeline (Eq. 11) converts the fitted surface into a recommended action, and no experiment evaluates whether that recommended action actually improves user performance. A model can have acceptable predictive MSE while its argmin is wrong, especially with a small grid of scale values and no model comparison. The absence of a user-factor statistical test for the simulated study further undermines the strong wording of the abstract. These are correctness risks, not internal inconsistencies or disagreements with consensus. The work remains a plausible proof-of-concept, and the physical-experiment trends are useful, but the headline claim is overreached. I agree with the CONDITIONAL verdict and would keep it: the paper should be published only if the authors add decision-level out-of-sample validation, model-checking, a user-effect test, and ideally code/data release.","tokens_in":821,"tokens_out":894,"duration_ms":78145,"concrete_test":"Perform per-user leave-one-cell-out cross-validation over the 24 condition cells. For each held-out delay d, fit the BPR model without that cell, compute s_hat*(d) from Eq. 11, and compare the realized held-out weighted performance at the nearest implemented scale against (a) the nominal scale s = 1 and (b) the empirical best scale at that delay, using paired tests. In parallel, fit a more flexible model (e.g., cubic terms or a Gaussian process) and compare predictive log-likelihood or WAIC; if the recommended s* does not improve performance out-of-sample, or if the alternative model shifts the argmin, the user-specific optimum claim is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that s*(d) = argmin_s P_o(s,d) (Eq. 11) is user-specific and predictable from the quadratic Bayesian polynomial regression in Eq. 2. The load-bearing step is therefore the location of the minimizer of that fitted surface, not just the predictive MSE reported in Fig. 6. The paper never validates the decision: it does not hold out a delay level or scale, fit the model, compute the recommended s*, and then check whether that s* outperforms the nominal s = 1 or the empirical best scale on the held-out trials. With only six scale levels and a fixed quadratic-in-(s,d) form, the argmin can be an artifact of the assumed polynomial shape rather than a real user-specific optimum. The abstract's claim of a statistically significant difference between users is also not backed by any user-factor test in the 2D study; the two-way ANOVA in Table I is for the physical study and includes latency and scale but not user. Finally, the physical experiment does not delay the stereo camera, so its connection to actual round-trip-delayed telesurgery is weaker than stated. If the quadratic surface is misspecified, or if the per-user optima are within estimation noise, the central personalized-prediction conclusion does not follow.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper investigates how motion scaling affects teleoperation performance under time delay, using both a simulated 2D Fitts'-law-style task and a physical da Vinci Research Kit peg-transfer task. The authors fit user-specific Bayesian polynomial regression models of performance (throughput, error, and a weighted combination) as a function of scaling factor and latency, then define the optimal scaling factor for a user and delay as the argmin of that fitted surface (Eq. 11). They claim that the optimal scaling factor is user-specific and varies with delay, and that an informative prior learned from other users improves predictive accuracy in terms of mean squared error. The abstract and introduction further claim a statistically significant difference in performance between users, motivating personalized scaling.","tokens_in":13085,"tokens_out":2952,"duration_ms":33613,"significance":"If the central claim is correct, the paper provides a practical, low-cost approach to personalizing motion scaling for telesurgery under delay, with a clear path to deployment using only a small amount of per-user data. The Bayesian prior transfer idea is sensible and the reported reduction in prediction MSE with an informed prior is a useful empirical result. However, the main scientific claim—that a user-specific, delay-dependent optimal scaling factor exists and can be predicted—is not actually validated at the decision level. The optimization is performed on the fitted quadratic surface, but no experiment or held-out analysis shows that the recommended scaling factor improves real performance relative to a fixed or nominal scaling factor. The user-specificity claim also lacks a direct statistical test, and the physical study delays only the command stream, not visual feedback. These gaps are substantial but addressable within the manuscript's scope.","major_comments":[{"comment":"The load-bearing claim is that s*(d) from Eq. (11), obtained by minimizing the fitted performance surface, is the optimal scaling factor for a user. Yet the model is evaluated only by predictive mean squared error (Fig. 6), not by whether the recommended s* actually improves measured performance on held-out trials. Without a decision-level validation (for example, holding out one latency level or one scale, fitting the model, computing s*, and comparing the achieved performance against the nominal scale and the empirical best scale on those held-out trials), the computed optima may be artifacts of the assumed quadratic polynomial shape rather than genuine user-specific optima.","section":"§II-C and §III-D"},{"comment":"The abstract claims 'a statistically significant difference in performance between users,' but no user-factor test is reported for the 2D study; Fig. 2 shows only two example users, and the statistical analysis in Fig. 4 compares scaling factors within the pooled data, not between users. The two-way ANOVA in Table I for the physical study includes latency and scale but not user as a factor. The user-specificity conclusion therefore currently rests on visual inspection of heatmaps rather than a formal statistical test, and the abstract overstates the evidence.","section":"Abstract and §III-C"},{"comment":"The physical experiment delays only the commands from the master to the patient-side manipulator; the manuscript explicitly states that 'the stereo camera information is not delayed.' The text then asserts that the buffer delay 'can be considered the round trip delay under the assumption delay is equal in both directions,' but this is an untested assumption. Because visual feedback is not delayed, the physical task does not reproduce the full sensorimotor delay loop of actual telesurgery, and the claim in §V that the physical experiments 'validate the findings from simulated environments' for round-trip-delayed telesurgery is overstated.","section":"§IV-A"},{"comment":"The performance model in Eq. (2) assumes P_o(s,d) is a second-degree polynomial surface in (s,d) with iid Gaussian noise, with the justification that 'the data seemed to follow a quadratic trend.' No goodness-of-fit diagnostics, residual analysis, or comparison against alternative functional forms (e.g., cubic terms, nonparametric regression) are provided. Since the optimum in Eq. (11) is determined by the curvature of this fitted surface, a misspecified mean function could produce an artificial minimum; this is a load-bearing modeling assumption that needs direct support.","section":"§II-A and §III-D"}],"minor_comments":[{"comment":"The word 'Additionaly' should be 'Additionally'.","section":"§I-B"},{"comment":"In the list of scaling factors, '04' should be '0.4'.","section":"§III-A"},{"comment":"The word 'adusted' should be 'adjusted' in the definition of the weighted performance metric.","section":"§III-B"},{"comment":"The text has 'FIg. 1' and 'defualt'; these should be 'Fig. 1' and 'default'.","section":"§IV-A"},{"comment":"Equation (9) appears to be missing a closing parenthesis and possibly an inverse; the predictive distribution for Bayesian polynomial regression is not written correctly, and this should be fixed for reproducibility.","section":"§II-A, Eq. (9)"},{"comment":"Figure 6 reports mean squared error without error bars or a description of how the mean and variance are computed across users; adding this information would clarify whether the improvement from the informed prior is consistent.","section":"§III-D"}],"recommendation":"major_revision","confidential_remarks":"To the editor: The paper has a practically relevant topic and the raw experimental data appear to show sensible latency/scale effects, but the central claim of user-specific optimal scaling is not yet supported by a decision-level validation or a direct user-factor statistical test. The missing validation is not a minor omission: it is the difference between 'we fit a curve' and 'the recommended scaling factor actually works.' I believe the authors can address this within the scope of the paper by adding a held-out evaluation of the recommended scaling factor, reporting a user-factor ANOVA or mixed-effects model, and tempering the claims about the physical study in light of the undelayed camera. I therefore recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The useful thing here is the question and the data: the authors run a clean 2D Fitts-style study and a dVRK peg-transfer study measuring how motion scaling and delay interact, and they show convincingly that lower scaling reduces error under delay and that there is a real speed/safety trade-off. The genuinely new piece is treating the optimal scaling factor as a per-user, delay-dependent quantity and fitting Bayesian polynomial regression with priors borrowed from other users. That is a legitimate extension of the earlier motion-scaling work by Orosco, Richter, and Yip, and the informed-prior MSE improvement in Fig. 6 is real evidence the regression wrapper has value.\n\nWhere the paper gets soft is the gap between the abstract and the analysis. The abstract says there is a \"statistically significant difference in performance between users,\" but no user-factor test appears anywhere. The paired t-tests compare scaling factors against 1.0, and the two-way ANOVA in the physical study only includes latency and scale. So the headline user-specificity claim rests on eyeballing heatmaps and the mean optimal-scale plot, not on an actual inference. That is a fixable omission, but it is currently a load-bearing one.\n\nThe second soft spot is the optimization loop. The optimal scaling factor s*(d) is defined as the argmin of the fitted second-degree polynomial surface, and the model is evaluated by predictive MSE, not by whether the recommended s* actually outperforms the nominal scale on held-out trials. With six scale levels and a fixed quadratic form, the per-user optimum can be an artifact of the assumed surface. A cross-validation that holds out a delay or scale, fits the model, computes s*, and checks it against a baseline would settle this. As written, the decision rule is unvalidated.\n\nThe physical study is a reasonable demonstration, but the stereo video is not delayed, so it is not a true round-trip-delayed telesurgery test. That limitation is acknowledged implicitly, but it should be stated plainly. Minor issues: the weighted performance metric depends on a free weight w, and the multiple t-tests are not corrected for multiple comparisons.\n\nThe citation pattern looks fine; the relevant motion-scaling literature is cited, including the authors' own prior work, which is appropriate here. This is not a case of missing key references.\n\nWho is this for? People working on teleoperation, surgical robotics, and human-robot interaction, especially anyone designing adaptive scaling or other human-in-the-loop latency mitigations. The paper deserves a serious referee, but it needs revision: add a user-factor test or soften the claim, run a proper out-of-sample decision evaluation, and report data/code. I would accept it as a conditional, with those as the conditions.","headline":"A useful proof-of-concept for per-user adaptive motion scaling under delay, but the headline user-specificity claim is not backed by a statistical test and the optimal scaling factor is never validated out-of-sample.","tokens_in":13651,"tokens_out":1977,"would_cite":true,"duration_ms":23233,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper argues that the optimal motion scaling factor for delayed telesurgery is different for every user, and that a Bayesian regression model trained with other users' data can predict each person's ideal scaling at any level of…","keywords":["telesurgery","motion scaling","communication delay","Bayesian polynomial regression","personalized modeling","teleoperation","latency","user study"],"falsifier":"Run a validation study where a new user's model is fit on a few trials and then compared on a full grid of scale and delay values: if the fitted quadratic's recommended scaling factor is no better than the default scale of 1.0 or the best observed grid point in held-out trials, the claimed prediction is falsified. Additionally, fit the same data with a cubic or spline surface and check whether the predicted optimum moves substantially; if it does, the quadratic assumption is driving the result rather than the data.","tokens_in":12617,"feed_emoji":"🩺","tokens_out":5767,"duration_ms":58043,"temperature":0.7,"pith_summary":"Long-distance telesurgery suffers from network delay, and one cheap remedy is motion scaling: reduce the ratio between what the surgeon moves and what the robot moves. The open question this paper attacks is what scaling factor to use at any given delay. Drawing on user studies in a simulated pointing task and a physical robotic peg-transfer task, the paper claims that the best scaling factor is different for each user and tends to decrease as latency rises. It then proposes a personalized Bayesian regression model, seeded with a prior learned from other users' data, that predicts a given user's optimal scaling factor for any delay. If the claim holds, telesurgical systems could automatically tune scaling per surgeon and per latency level at almost no cost, improving both speed and safety.","feed_headline":"Optimal motion scaling is personal and delay-dependent","feed_subtitle":"Bayesian models with priors from other users predict each surgeon's ideal scale for a given latency.","key_machinery":"The load-bearing object is the personalized Bayesian Polynomial Regression surface $\\hat{P}_o(s,d) = \\varphi(s,d)^\\top \\beta$, a second-degree polynomial in scaling factor $s$ and delay $d$ with Gaussian observation noise $\\varepsilon \\sim N(0,\\sigma^2)$ and a Normal-Inverse-Gamma prior on $(\\beta,\\sigma^2)$. The prior for a new user is obtained by maximising the likelihood of all other users' datasets, so the model's predictions for a new user are Student-$t$ distributions rather than point estimates; the recommended scale is the minimizer of the predictive mean found by discrete search over scale values. The mechanism that carries the argument is the combination of a response surface and a cross-user prior, which converts a few trials from a new operator into a delay-dependent scaling recommendation.","core_discovery":"The paper's central claim is that the optimal motion scaling factor is a per-user, delay-dependent quantity: for each operator there exists a value $s^\\ast(d) = \\operatorname{argmin}_s \\hat{P}_o(s,d)$ that balances speed against safety better than a fixed scale, and this value shifts downward as latency grows, but at a user-specific rate. The claim is supported by two user studies: a 2D Fitts-style pointing task with ten participants and a peg-transfer task on the da Vinci Research Kit with simulated round-trip delays. Throughput served as the speed metric, and overshoot and force measures served as error, combined into a weighted performance score $WP = (1-w)\\,TP - w\\,(OSD+\\Delta D)$. The paper then argues that a personalized Bayesian polynomial regression surface can capture this relationship, and that when the regression prior is learned by maximum likelihood from other users' data, the model predicts a new user's optimal scale with less data than an uninformative prior.","pith_inferences":["Beyond the paper: because the weighted performance metric encodes a clinical preference, the same fitted model can answer counterfactuals, such as what scaling schedule a safety-first department should use, without additional user studies.","Beyond the paper: a natural next experiment is a head-to-head out-of-sample comparison of the BPR-recommended scale versus the nominal default and versus a fixed low scale, measuring whether personalized recommendation actually improves composite performance.","Beyond the paper: the reported user-to-user variance suggests that a population-level model with user-specific random effects could estimate optima for an unseen user with even fewer trials than the MLE-prior approach."],"forward_implications":["A new user's optimal scaling schedule can be estimated from relatively few calibration trials, because the informed prior transfers information from previously observed users.","At higher delays, the model will generally recommend lower scaling factors, matching the observed trend that overshoot and error dominate throughput losses.","Telesurgical systems could adjust scaling continuously as measured round-trip delay changes, without hardware changes, since scaling is already implemented in software.","The recommended scaling depends on the speed/safety weight $w$; a setting that prioritizes safety will receive a different, lower scaling schedule than one prioritizing speed."],"supporting_citations":[{"why":"Established that motion scaling reduces error in high-delay teleoperation without substantially increasing completion time, providing the baseline effect this paper extends.","marker":"[21]"},{"why":"Investigated scaling factor effects under different delays, reporting that lower scaling improved error and completion time at 750 ms delay.","marker":"[23]"},{"why":"Demonstrated motion scaling benefits on the da Vinci Research Kit in live telesurgery, motivating the physical experiments here.","marker":"[22]"},{"why":"Showed that latency impacts surgical precision and task completion, supporting the paper's premise that delay effects are user-specific.","marker":"[14]"},{"why":"Supplies the derivation of the Bayesian polynomial regression equations used to model the performance surface.","marker":"[36]"},{"why":"Defines the Fitts-law throughput metric used to measure the speed component of performance.","marker":"[38]"},{"why":"Provides the open-source da Vinci Research Kit platform used in the physical user study.","marker":"[39]"}],"fun_headline_variants":["Delay dictates each surgeon's ideal motion scale","Personalized scaling beats fixed for delayed telesurgery","Surgeon-specific scale is key under network lag","Optimal telesurgery scale varies by user and latency","Bayesian model finds per-user scale for delayed surgery"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole optimization depends on assuming that for each person, performance as a function of scaling factor and delay is a smooth bowl-shaped surface plus random scatter, and the paper does not test whether that assumed shape is actually right.","fun_headline_variants_meta":{"raw":{"variants":["Delay dictates each surgeon's ideal motion scale","Personalized scaling beats fixed for delayed telesurgery","Surgeon-specific scale is key under network lag","Optimal telesurgery scale varies by user and latency","Bayesian model finds per-user scale for delayed surgery"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000184,"raw_usage":{"total_tokens":1299,"prompt_tokens":904,"completion_tokens":395,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":520,"completion_tokens_details":{"reasoning_tokens":321}},"tokens_in":520,"tokens_out":395,"duration_ms":4410,"temperature":1.0,"reasoning_tokens":321,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T22:21:35.484510+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a validation study where a new user's model is fit on a few trials and then compared on a full grid of scale and delay values: if the fitted quadratic's recommended scaling factor is no better than the default scale of 1.0 or the best observed grid point in held-out trials, the claimed prediction is falsified. Additionally, fit the same data with a cubic or spline surface and check whether the predicted optimum moves substantially; if it does, the quadratic assumption is driving the result rather than the data.","supporting_citations":[{"cited_title":"Motion scaling solutions for improved performance in high delay surgical teleoperation,","cited_arxiv_id":null,"evidence_quote":"Established that motion scaling reduces error in high-delay teleoperation without substantially increasing completion time, providing the baseline effect this paper extends."},{"cited_title":"Compensatory motion scaling for time-delayed robotic surgery,","cited_arxiv_id":null,"evidence_quote":"Investigated scaling factor effects under different delays, reporting that lower scaling improved error and completion time at 750 ms delay."},{"cited_title":"From bench to bedside: The first live robotic surgery on the dvrk to enable remote telesurgery with motion scaling,","cited_arxiv_id":null,"evidence_quote":"Demonstrated motion scaling benefits on the da Vinci Research Kit in live telesurgery, motivating the physical experiments here."},{"cited_title":"The impact of latency on surgical precision and task completion during robotic-assisted remote telepresence surgery,","cited_arxiv_id":null,"evidence_quote":"Showed that latency impacts surgical precision and task completion, supporting the paper's premise that delay effects are user-specific."},{"cited_title":"O’Hagan, J","cited_arxiv_id":null,"evidence_quote":"Supplies the derivation of the Bayesian polynomial regression equations used to model the performance surface."},{"cited_title":"Towards a standard for pointing device evaluation, perspectives on 27 years of fitts’ law research in hci,","cited_arxiv_id":null,"evidence_quote":"Defines the Fitts-law throughput metric used to measure the speed component of performance."},{"cited_title":"An open-source research kit for the da vinci® surgical system,","cited_arxiv_id":null,"evidence_quote":"Provides the open-source da Vinci Research Kit platform used in the physical user study."}],"review_version":1}