{"id":"1d5fa58b-6f30-4bbe-b607-be232c36169c","arxiv_id":"2411.12765","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Motion capture and stepwise regression show that wrist rotations and several thumb and index finger joint movements are significantly associated with overshoot in a blindfolded 90-degree knob rotation.","lead":"This paper tracks arm and hand movements while blindfolded people turn a rotary knob, and it identifies which joints are linked to the tendency to turn too far. The results suggest two causes: a hand-centered sense of direction that misses part of the rotation, and fingertips that roll on the knob and add extra turning.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Unsigned max-min joint ranges cannot establish which joints 'contributed' to overshoot; pooling grasp orientations makes the reported wrist/DIP coefficients hard to interpret.","rationale":"The reader's weakest assumption—trial non-independence and uncorrected stepwise selection—is real and important: with 18 participants and repeated trials, OLS standard errors are likely anti-conservative, and backward elimination invalidates the reported p-values. I agree that a mixed-effects model and multiple-testing correction are needed. However, I see an even more load-bearing issue that the reader did not emphasize: the predictor variable itself, an unsigned max-min range, cannot support the directional claim that a given joint 'contributed to overshooting.' The paper's mechanisms (wrist egocentric bias, fingertip rolling) and the signed-error criterion both require knowing whether the joint moved in the rotation-consistent direction. A range discards that information, and pooling across grasping orientations, for which the same joint moves in different directions, can manufacture or obscure significant coefficients. This is not a disagreement with the reader's statistical concern; it is a deeper construct-validity problem that persists even after fixing the repeated-measures structure. I therefore keep the reader's CONDITIONAL verdict unchanged: the paper would need a signed-displacement (or otherwise direction-sensitive) reanalysis before the central claim is supported.","tokens_in":12214,"tokens_out":7739,"duration_ms":82040,"concrete_test":"Recompute the overshoot regression replacing each unsigned range δ_j with the signed net joint displacement θ_j(end) − θ_j(start) over the extracted rotation segment (or, equivalently, with the projection of the trajectory onto the rotation-consistent direction per grasping orientation), using the same min-max normalization and model. If the wrist (δx_w, δy_w, δz_w) and index DIP coefficients are no longer significant or change sign, the claim that these joints contribute to overshooting is not supported by the current range-based predictor. Additionally, fit the original range model with a linear mixed model including random intercepts for participant and fixed effects for grasping orientation; if the significant list changes materially, the pooled OLS p-values are unreliable.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on regressions (Section 3, Table 1 left) whose predictors are unsigned max-min ranges δ_j of each joint over the rotation segment (Section 2.6). A range conflates movement in the rotation direction with movement against it. The criterion is signed overshoot (δ_τ − 90°), so a positive coefficient for δ_j only means 'larger total excursion at this joint is associated with more overshoot', not that the joint moved in the direction that adds rotation. This matters because the two proposed mechanisms—wrist bias in an egocentric frame and fingertip rolling—are directional. The problem is aggravated by pooling the five grasping orientations in the overshoot regression: for a fixed counterclockwise rotation, the same joint can move in opposite angular directions depending on initial grasp; Table 1 right shows sign and membership of contributing joints change across orientations. A range predictor collapses these opposite-signed contributions, so the significant wrist and DIP coefficients in Table 1 left could be a mixture artifact rather than evidence that these joints drive overshoot. The absence of random effects and multiple-testing correction is a separate, valid concern, but even a perfect mixed model would not cure the unsigned-predictor problem.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes motion capture data from a haptic rotation task (90° counterclockwise, blindfolded, two-finger grasp) to identify which upper limb and hand joint angle ranges predict the signed overshoot (δτ − 90°). Using stepwise ordinary least squares regression with backward elimination, the authors report that the wrist joint, the sideways movement in proximal finger joints, and the distal finger joints contribute significantly to overshooting. They also run per-grasping-orientation regressions to examine which joint ranges contribute to the achieved rotation. The results are interpreted as evidence for two mechanisms: a hand-centered egocentric reference frame bias (wrist contribution) and fingertip rolling over the knob surface (finger joint contributions).","tokens_in":12376,"tokens_out":4280,"duration_ms":41689,"significance":"The paper addresses a clearly relevant question in haptics and human motor control: where in the kinematic chain the systematic overshoot in blindfolded rotary knob rotations originates. The experimental setup (motion capture with 18 participants, five grasping orientations, repeated trials) and the joint-level analysis framework are potentially valuable for future work in rehabilitation, robotics, and human-robot interaction. The paper is also transparent in reporting coefficients and p-values. However, the current analysis has serious statistical and conceptual limitations that prevent the central claim from being established. The proposed framework may be salvageable with a different predictor definition and a mixed-model approach, but as presented the evidence does not support the conclusions.","major_comments":[{"comment":"The predictors are unsigned max-min ranges δ_j of each joint angle over the rotation segment, while the criterion is the signed overshoot δτ − 90°. A positive regression coefficient for a range variable only indicates that larger total joint excursion is associated with more overshoot; it cannot distinguish movement in the direction that adds to the rotation from movement in the opposite (compensatory) direction. For example, a wrist rotation that opposes the knob movement and a wrist rotation that reinforces it can produce the same range value. The central claim that the wrist and finger joints 'contributed significantly to overshooting' (Abstract and Section 4) is therefore not supported by the regression as specified. The two proposed mechanisms are directional (egocentric reference frame bias and fingertip rolling), so the analysis needs a signed measure of joint displacement (e.g., net angle change over the rotation segment) or a decomposition into positive and negative contributions.","section":"Section 2.6 and Table 1 (left)"},{"comment":"The overshoot regression pools all trials from 18 participants, with each trial treated as an independent observation. Since each participant contributes multiple trials under each grasping orientation, within-participant correlations are likely substantial, and the reported p-values and standard errors are accordingly unreliable. Additionally, the stepwise backward elimination procedure performs many significance tests without any correction for multiple comparisons, so the final model (reported R² = 0.691) is likely overfitted and the list of 'significant' joints may be spurious. The lack of random effects for participants is a load-bearing problem for the central claim because the identification of contributing joints depends on the validity of these p-values.","section":"Section 3, Tables 1 and G (Tables 2–7)"},{"comment":"The per-orientation regression analyses of the rotation angle on joint ranges suffer from the same unsigned-predictor problem. A negative coefficient is interpreted as the joint 'compensating' for the rotation (Section 4), but a negative association between a range variable and the rotation angle does not imply that the joint moved in the opposite direction to the knob rotation. Moreover, these regressions also pool trials without participant random effects, and the constant term (ranging from 28.36 to 70.50) is treated as a catch-all for 'additional body movements' without a clear mechanistic interpretation. The reported p-values are conditional on the stepwise selection and are not corrected for multiplicity.","section":"Section 3, Table 1 (right) and Tables 3–7"}],"minor_comments":[{"comment":"The phrase 'the mean difference between the target and the encoded positions' is ambiguous; it should be clarified whether the signed error is computed as encoded minus target or vice versa.","section":"Section 1"},{"comment":"There is a typo: 'shown in shown in Table 1' should be 'shown in Table 1'.","section":"Section 3"},{"comment":"In the regression results for the free grasping condition, the predictor labeled 'δf/e i,MCP' appears twice with different coefficients (41.50 and 26.80). One of these is likely meant to be a different joint (e.g., PIP); please correct the labeling.","section":"Appendix G, Table 7"},{"comment":"The appendix is hosted on a password-protected website with the German word 'Passwort'; for an international readership and for review, the access information should be in English and ideally the supplementary material should be included with the submission.","section":"Introduction/Appendix access"},{"comment":"The description of the hand model states that the thumb has joints TMC, MCP, and IP, while the index finger has MCP, PIP, and DIP; this is clear, but the notation in Table 1 (e.g., δf/e_t,MCP) could be confusing because the subscript 't' is used for the thumb and 'i' for the index finger; a brief reminder in the caption of Table 1 would help.","section":"Section 2.5"}],"recommendation":"major_revision","confidential_remarks":"The central claim of the paper is not supported by the current analysis because the predictors are unsigned ranges. This is a fixable issue in principle, since the recorded joint angle trajectories could be re-analyzed using signed displacements or separate positive/negative components, and the pooled analysis could be replaced by a mixed model. However, the required changes are substantial and would affect the main results and their interpretation. I recommend major revision rather than rejection because the experimental framework and data collection are valuable and the proposed approach, once corrected, could make a meaningful contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Briefly: this paper does something genuinely new—it takes motion-capture data from a previously published haptic rotation experiment and asks which individual joints' excursions predict the well-known overshoot. That decomposition hasn't been done before, and it's a useful step for haptics and rehabilitation. The paper is also admirably transparent: it describes the preprocessing, includes detailed regression tables in the appendix, and flags the elbow joint's approximate tracking. The writing is direct and the reasoning is easy to follow.\n\nThe soft spot is the predictor construction. The regressions in Section 3 use unsigned max-min ranges δ_j as predictors for signed overshoot (δτ − 90°). A range cannot distinguish movement that adds rotation from movement that opposes it. So a positive coefficient on, say, the wrist range only tells you that larger total wrist excursion is associated with more overshoot—it does not tell you the wrist moved in the direction that produces overshoot. The two proposed mechanisms (egocentric wrist bias, fingertip rolling) are directional, so the evidence doesn't really test them. This is not a minor presentational issue; it is the difference between 'the wrist is involved' and 'the wrist drives overshoot.' The per-orientation regressions in Table 1 (right) make the problem concrete: the same joint appears with opposite signs across grasp angles, which is exactly what you'd expect if ranges are collapsing opposite-signed movements.\n\nThe other concerns are real but secondary. Trials from 18 participants are pooled as independent observations; no random intercepts, no correction for the stepwise selection. The R² values look fine, and the coefficients are large, but the p-values are optimistic. The paper reuses the authors' own earlier data, which is fine as long as the reader knows; it's stated clearly.\n\nMy take: the central claim about which joints contribute is plausible but not established. The paper deserves a serious referee, but I would send it back for major revision—add a mixed-effects model, and more importantly, replace or supplement the unsigned ranges with signed or phase-specific joint angles. As it stands, the mechanistic interpretation outruns the statistics.\n\nWho reads this: haptics researchers, motor control people, and anyone designing knob interfaces. It's a good case study for a reading group on regression pitfalls. I'd engage with it, but treat the conclusion as a conditional hypothesis.","headline":"Novel decomposition of haptic overshoot into joint excursions, but unsigned predictors and pooled trials make the mechanistic claims conditional.","tokens_in":12923,"tokens_out":1842,"would_cite":false,"duration_ms":17617,"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":"Blindfolded people systematically overshoot when rotating a knob to a target angle, and this paper traces that error to the wrist and to sideways and distal finger joint motions.","keywords":["haptic rotation","overshoot","joint angle analysis","motion capture","stepwise regression","egocentric reference frame","fingertip rolling","orientation bias"],"falsifier":"Refit the same model with a random intercept per participant (a linear mixed-effects model) and check whether the wrist and finger coefficients remain significant; if they do not, the claimed localization is an artifact of pooled trials. Alternatively, measure fingertip contact-point displacement during rotation directly: if overshoot occurs without any rolling of the fingertips over the knob surface, the paper's rolling mechanism is falsified.","tokens_in":11982,"feed_emoji":"🎛️","tokens_out":9797,"duration_ms":78114,"temperature":0.7,"pith_summary":"This paper asks why people overshoot when they rotate a rotary knob blindfolded to a target angle, and it locates the cause in specific joints rather than in a whole-arm systematic bias. Motion capture of 18 participants performing a 90-degree counterclockwise rotation with thumb and index finger yields joint-angle ranges for 15 hand and upper-limb degrees of freedom. Stepwise regression with backward elimination shows that the wrist joint, sideways movement at the proximal finger joints, and the distal finger joints significantly predict the overshoot, while the elbow and several thumb movements have negative coefficients, meaning they compensate. The authors conclude that two mechanisms are at play: a hand-centered egocentric reference frame that undercounts the wrist's own rotation, and un-compensated rolling of the fingertips over the knob surface that adds rotation without the person's awareness.","feed_headline":"Wrist and finger joints drive the blindfolded knob overshoot","feed_subtitle":"A regression on joint-angle ranges shows the overshoot is local, not whole-arm, and names two probable mechanisms.","key_machinery":"The machinery is a motion-capture-to-regression pipeline. Reflective markers on the hand and upper limb are converted, via a 16-joint, 26-DoF kinematic hand model and a 7-DoF shoulder-elbow-wrist model, into joint-angle trajectories, synchronized with a rotary-knob apparatus that records absolute orientation with 0.022-degree resolution. Each trial is compressed to one predictor per joint — the range (maximum minus minimum angle) over the rotation segment — and to the criterion, the signed error from the 90-degree target. Stepwise ordinary least squares regression with backward elimination, using min-max normalization per participant per joint, then selects the joints whose ranges significantly predict the error, which is the step that localizes the overshoot.","core_discovery":"The central discovery is a statistical localization of the haptic-rotation overshoot. Regressing the signed error (actual rotation minus 90 degrees, normalized per participant and per joint to [0,1]) on each joint's range of motion yields a final model with $R^2 = 0.691$, in which the three wrist axes (coefficients 30.61, 21.45, 20.13), the thumb's TMC flexion/extension (24.05), the index MCP abduction/adduction (8.55), and the index DIP flexion/extension (34.67) all contribute positively and significantly to overshoot. Elbow flexion/extension (-13.62), thumb MCP flexion/extension (-25.96), and thumb TMC abduction/adduction (-17.46) contribute negatively, i.e., they limit the overshoot. The paper interprets the wrist effect as evidence for a hand-centered egocentric reference frame and the finger effects as evidence that the fingertips roll over the knob surface, changing the contact point, without the participant compensating.","pith_inferences":["If the egocentric reference-frame account is right, then providing participants with visual or haptic feedback about hand orientation during the movement should reduce the wrist's contribution to overshoot, a testable prediction the paper does not make.","The per-orientation regressions imply that the set of significant predictors is not universal: a joint can contribute positively in one grasp and negatively in another, so the overshoot mechanism likely shifts with task geometry rather than being a fixed motor error.","Compressing each joint trajectory to a single range discards timing and phase; using velocity or temporal-alignment features might reveal that the order of joint recruitment matters for the size of the overshoot.","A direct non-invasive check is feasible: track the fingertips with small markers or high-speed video during rotation to confirm the rolling motion and quantify its angular contribution."],"forward_implications":["The overshoot in blindfolded haptic rotation is a local, joint-specific phenomenon: the wrist and particular finger joints carry the error, while the elbow and some thumb motions actively compensate.","The wrist's significant positive contribution supports a hand-centered egocentric reference frame, meaning the brain systematically under-weights the hand's own rotation.","Fingertip rolling over the knob surface, driven by sideways proximal-joint motion and distal flexion, adds rotary movement that the blindfolded participant does not compensate.","Grasping orientation changes the joint usage pattern: the shoulder contributes in all five orientations, the wrist drops out at the 90-degree grasp, and the index finger shifts from MCP abduction/adduction to PIP and DIP flexion as the grasp angle increases.","Future experiments can test the two mechanisms directly by restricting hand movement or preventing rolling, and the recorded data can be re-analyzed to explain why using more fingers reduces overshoot."],"supporting_citations":[{"why":"Supplies the experimental data set — 18 participants, five grasping orientations, 90-degree counterclockwise two-finger rotations — that this paper re-analyzes.","marker":"[4]"},{"why":"Introduced the Twister apparatus for recording knob orientation and established the overshoot effect in haptic rotation.","marker":"[3]"},{"why":"Provides comparison overshoot magnitudes for rotary versus motionless knobs in different grasping setups.","marker":"[1]"},{"why":"Identified movement strategies during knob turning that motivate the joint-angle analysis of this paper.","marker":"[2]"},{"why":"Describes the automatic motion tracking software used to turn marker trajectories into hand joint angles.","marker":"[7]"},{"why":"Provides the kinematic hand model with 16 joints and 26 degrees of freedom that the tracking software builds upon.","marker":"[9]"},{"why":"Supplies the upper-limb model with 7 degrees of freedom that guided the computation of shoulder, elbow, and wrist angles.","marker":"[14]"},{"why":"Evidence for a biased egocentric reference frame in haptic spatial judgments, used to interpret the wrist effect.","marker":"[10]"},{"why":"Background on egocentric versus allocentric reference frames in haptic space, supporting the hand-centered frame interpretation.","marker":"[13]"}],"fun_headline_variants":["Wrist and finger joints behind blindfolded knob overshoot","Overshoot in haptic rotation traced to wrist and fingers","Fingertip rolling and wrist bias cause knob overshoot","Haptic rotation overshoot: not whole-arm, but wrist and fingers"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The result depends on treating every trial from every participant as an independent observation in a single regression; if within-participant correlations are large, the reported p-values and the list of significant joints could be spurious.","fun_headline_variants_meta":{"raw":{"variants":["Wrist and finger joints behind blindfolded knob overshoot","Overshoot in haptic rotation traced to wrist and fingers","Fingertip rolling and wrist bias cause knob overshoot","Haptic rotation overshoot: not whole-arm, but wrist and fingers"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000191,"raw_usage":{"total_tokens":1357,"prompt_tokens":971,"completion_tokens":386,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":587,"completion_tokens_details":{"reasoning_tokens":313}},"tokens_in":587,"tokens_out":386,"duration_ms":23108,"temperature":1.0,"reasoning_tokens":313,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T18:50:24.925748+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Refit the same model with a random intercept per participant (a linear mixed-effects model) and check whether the wrist and finger coefficients remain significant; if they do not, the claimed localization is an artifact of pooled trials. Alternatively, measure fingertip contact-point displacement during rotation directly: if overshoot occurs without any rolling of the fingertips over the knob surface, the paper's rolling mechanism is falsified.","supporting_citations":[{"cited_title":"In: 2019 IEEE World Haptics Conference (WHC)","cited_arxiv_id":null,"evidence_quote":"Supplies the experimental data set — 18 participants, five grasping orientations, 90-degree counterclockwise two-finger rotations — that this paper re-analyzes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduced the Twister apparatus for recording knob orientation and established the overshoot effect in haptic rotation."},{"cited_title":"motionless knobs","cited_arxiv_id":null,"evidence_quote":"Provides comparison overshoot magnitudes for rotary versus motionless knobs in different grasping setups."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Identified movement strategies during knob turning that motivate the joint-angle analysis of this paper."},{"cited_title":"In: 2015 IEEE-RAS 15th International Conference on Humanoid Robots (Humanoids)","cited_arxiv_id":null,"evidence_quote":"Describes the automatic motion tracking software used to turn marker trajectories into hand joint angles."},{"cited_title":"In: 2014 IEEE International Conference on Robotics and Automation (ICRA)","cited_arxiv_id":null,"evidence_quote":"Provides the kinematic hand model with 16 joints and 26 degrees of freedom that the tracking software builds upon."},{"cited_title":"Sensor Review 39(4), 504–511 (2019) 14 K","cited_arxiv_id":null,"evidence_quote":"Supplies the upper-limb model with 7 degrees of freedom that guided the computation of shoulder, elbow, and wrist angles."},{"cited_title":"Psychol Behav Sci3, 212–221 (2014)","cited_arxiv_id":null,"evidence_quote":"Evidence for a biased egocentric reference frame in haptic spatial judgments, used to interpret the wrist effect."},{"cited_title":"Experimental Brain Research188, 199– 213 (2008)","cited_arxiv_id":null,"evidence_quote":"Background on egocentric versus allocentric reference frames in haptic space, supporting the hand-centered frame interpretation."}],"review_version":1}