{"id":"3e0e2afa-eb84-49df-9bd2-4657cffeadf5","arxiv_id":"2506.10923","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Vib2Move controls fingertip vibrations to modulate friction, letting a parallel gripper slide and rotate planar objects in hand under gravity with about 6 mm final position error.","lead":"A robotic gripper that vibrates its fingertips can slide and rotate flat objects in hand using only gravity, without touching the environment. The Vib2Move system reaches roughly 6 mm final positioning accuracy on real objects, offering a simple way to give two-finger grippers some dexterity.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The motion model's scale factor k in Eq. 6 is not derived and may have been fit per vibration action, making the Table I model evaluation potentially circular.","rationale":"The reader's weakest assumption (quasi-static stopping) is plausible, but the empirical success of the feedback planner already provides indirect evidence that stopping behavior is adequate for the tested objects. The scale-factor issue is more foundational: it affects whether the model contribution is a predictive model or a curve fit. The paper explicitly says 'we treat the scale of this limit surface intuitively as a scale factor for the twist' (Sec. III-A), so this is not hidden, but no calibration protocol is reported. Since no code or data are available, the ambiguity cannot be resolved from the text. This keeps the verdict conditional rather than accept. I agree with the reader's conditional assessment; my concern is a different load-bearing point, so agreement is partial.","tokens_in":12534,"tokens_out":7767,"duration_ms":86601,"concrete_test":"Split the 1564 recorded vibration actions into disjoint calibration and test sets. Fit k (or equivalently the A scale) on the calibration set, freeze it, and recompute RMSE on the test set. Also report the fitted k values per object and whether one global k reproduces the reported errors. If test RMSE is materially worse than Table I, the model evaluation is not a valid test of the predictive model.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing weakness is the undisclosed scale factor k introduced in Eq. 6. The paper claims the model depends only on contact geometry, but k = 1/(wf^T A wf) depends on the unknown normal-force/friction scale (μN) and on object mass through wf = [0, -mg, -mg xc]. If the object is actually sliding, the wrench lies on the limit surface, so wf^T A wf = 1 and k = 1; otherwise there is no physical justification for using k as a step-size multiplier. No procedure is given for choosing k per object or per vibration action, and the 1564-action evaluation in Table I appears to be a single-table RMSE without a train/test split. If k was fitted to minimize error on the same actions, the reported 1.849 mm / 2.189° prediction error does not demonstrate predictive power. This matters because the motion model is one of the three stated innovations and the basis for the planner's subgoal heuristics. It is distinct from the quasi-static concern: even granting quasi-static behavior, the model's magnitude is not derived.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"Vib2Move proposes to perform in-hand planar reconfiguration with a parallel-jaw gripper by vibrating the fingertips to reduce effective friction and letting gravity slide the object. The paper derives a quasi-static sliding motion model from an ellipsoidal limit surface, introduces a scale factor k in the pose update, and uses a three-stage closed-loop planner (centering, translation, orientation) with pose feedback. Experiments on six planar objects report a model prediction RMSE of 1.849 mm / 2.189 deg over 1564 vibration actions and a final reconfiguration RMSE of 5.91 mm / 1.21 deg over 60 trials.","tokens_in":12751,"tokens_out":6873,"duration_ms":70163,"significance":"The idea is attractive: if validated, a simple two-finger gripper with vibration motors can reconfigure objects in free space without external fixtures or force sensing. The hardware is simple, the use of limit-surface theory for predicting motion direction is reasonable, and the planner is transparent and reproducible. However, the central quantitative claims are not currently supported as stated: the abstract's sub-6 mm claim is contradicted by Table II, and the magnitude part of the motion model rests on an underived scale factor. These issues are fixable but need to be addressed before publication.","major_comments":[{"comment":"The abstract states that Vib2Move consistently yields final positioning errors below 6 mm, but Table II reports per-object RMSE values of 6.59 mm (Object 1), 6.36 mm (Object 5), and 7.04 mm (Object 6); the aggregate average of 5.91 mm is below 6 mm, so the claim is only true on average, not for each object or trial. Please rephrase the claim to match the data or report the full per-trial distribution, including worst-case error and standard deviation.","section":"Abstract / Table II"},{"comment":"The scale factor k = 1/(w_f^T A w_f) is introduced without derivation, yet it controls the magnitude of every pose update. Because A = Diag{(a_1 N)^-2, ...}, k depends on the unknown normal force N and friction coefficient mu; because w_f = [0, -mg, -mg x_c], it also depends on object mass. This contradicts the paper's claim that the model depends only on contact geometry. If the friction wrench lies on the limit surface during sliding, then w_f^T A w_f = 1 and k = 1, in which case the factor is redundant; if the wrench is inside the limit surface, no physical argument is given for using this ratio as a step-size multiplier. In addition, Eq. (6) adds a dimensionless normalized twist times delta_t and k to a pose vector whose translational components are in mm and angular component in radians, but no conversion between linear and angular scales is provided. The evaluation in Table I does not state how k was chosen, whether it was fit to the same 1564 actions, or whether any actions were held out; as reported, the 1.849 mm / 2.189 deg RMSE therefore does not establish predictive power. Please derive or fit k explicitly, report its values, and validate on held-out actions.","section":"Section III-A.2, Eq. (6) / Table I"},{"comment":"The paper states that inertial forces are negligible and the object comes to rest as soon as vibrations are turned off, and it claims this assumption is evaluated in the results section. The planner's stepwise execution depends on this assumption: stage transitions assume a binary stick/slide behavior controlled by the vibration on/off signal. Section VI reports only final-path RMSE and a qualitative path comparison; I found no direct experiment that measures stopping/starting transients, checks that the object is stationary with vibration off, or compares predicted versus actual motion duration. Please add such a test, or revise the claim to state that the assumption is an unvalidated modeling choice.","section":"Section III-A, quasi-static assumption"}],"minor_comments":[{"comment":"The sentence 'for a fixed contact patch size, A proportional to N' is incorrect as written; from the definition A = Diag{(a_1 N)^-2, ...}, A scales as N^{-2}, not N. The intended statement is presumably that the size of the limit surface is proportional to N. Please correct this, as it appears in the argument leading to Eq. (6).","section":"Section III-A.1"},{"comment":"The caption states that results are shown as a function of the timestep chosen, but the table contains a single set of values and no timestep value. Please specify the timestep or remove the phrase.","section":"Table I"},{"comment":"The model evaluation reports 1564 collected vibration actions but gives no per-object breakdown, standard deviation, or number of actions per object. Reporting these would make the evaluation in Table I interpretable.","section":"Section V-B"},{"comment":"The relative error is defined as RMSE normalized by the object's max length, but this normalization is not stated in the table caption or the surrounding text; per-trial standard deviations should also be reported alongside the means.","section":"Table II"},{"comment":"There are several typos, including 'the the object comes to rest' in Section III-A, 'foundation for to our planning' in Section III, and 'we can devide the motion' in the Fig. 9 caption. Please copyedit the manuscript.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the journal's scope. The main concern to monitor in revision is the status of k in Eq. (6): if k was tuned on the same data, the model evaluation is circular, and the paper currently does not rule this out. The abstract/Table II discrepancy also needs to be resolved. I see no novelty disclosure or citation concerns."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know before you read it. First, the core idea is new and worth knowing: a standard parallel gripper with cheap coin motors can slide a planar object in the vertical plane by toggling vibration, and the object's motion direction is predicted from a limit-surface model with no fitted parameters. Second, the paper's headline error claim is not supported by its own table, and the step-size multiplier in the motion model is ad hoc and possibly fitted, which weakens the claim that the model is predictive.\n\nWhat is genuinely good. The hardware is refreshingly simple and reproducible: two 10 mm coin motors, 3D-printed fingertips, paper tape, on-off control. The authors correctly identify that in the vertical plane the friction wrench is determined by force balance, so the direction of sliding follows from geometry alone (Eq. 4). The experiments are real-world: 1564 vibration actions for the model and 60 full reconfiguration trials across six objects with varied mass and surface. The closed-loop planner with three stages (center, translate, orient) is a sensible decomposition, and the authors are honest that open-loop would not work.\n\nWhere it gets soft. The abstract says errors ‘below 6 mm’, but Table II lists per-object RMSE from 4.36 to 7.04 mm. The average is 5.91 mm; only a reader squinting at averages can call that ‘below 6 mm’. That needs a rewrite or a better metric. The bigger issue is the scale factor k in Eq. 6. The paper says the model depends only on contact geometry, but k = 1/(wf^T A wf) is not derived; it depends on N and mg, which the authors do not know. There is no stated procedure for choosing k, and the 1564-action model evaluation is reported as a single RMSE with no train/test split. If k was tuned per object or per action on the same data, the 1.849 mm prediction error does not demonstrate out-of-sample prediction. The direction prediction is real, but the magnitude is not. The authors also promise a timestep study in Table I that is not there; the table just has one row. Finally, the quasi-static assumption (object stops immediately when vibration stops) is plausible but never directly tested.\n\nWho is this for? Anyone working on in-hand manipulation with simple grippers will want to know this idea exists. It deserves a serious referee, but the authors need to disclose how k is set, report variance, fix the abstract, and add the missing timestep analysis before it is acceptable. I would not cite it in its current form.","headline":"New mechanism, real robot results, but the error claim is overstated and the motion-model scale factor k is not derived.","tokens_in":13277,"tokens_out":2530,"would_cite":false,"duration_ms":26182,"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":"Vib2Move shows that fingertip micro-vibrations let a two-finger gripper slide planar objects in hand using gravity, reaching pose errors below 6 mm.","keywords":["in-hand manipulation","parallel gripper","fingertip vibration","friction modulation","limit surface","planar sliding","motion planning","quasi-static manipulation"],"falsifier":"Run a high-speed camera or miniature IMU on a grasped object through repeated vibration-on/off cycles: if the object continues moving after vibration stops, or if a heavier object fails to slide at the 5 N grasp force, the binary stick/slide quasi-static model is violated.","tokens_in":1473,"feed_emoji":"🤏","tokens_out":1532,"duration_ms":83126,"temperature":0.7,"pith_summary":"Vib2Move is a method for in-hand reconfiguration that gives an ordinary parallel-jaw gripper the ability to reposition a planar object in free space. The fingertips vibrate at high frequency, breaking contact repeatedly to lower the effective friction, which effectively emulates fast changes in grip force. Gravity then slides the object relative to the fingers, and the paper derives a limit-surface model that predicts the direction of that motion. A three-stage planner uses these predictions to center the object, translate it, and rotate it to a goal pose. Real-world trials across six objects report final position errors below about 6 mm and orientation errors near 1 degree.","feed_headline":"Vibrating fingertips slide objects in hand to sub-6 mm accuracy","feed_subtitle":"A parallel-jaw gripper repositions planar objects using micro-vibrations and gravity alone, no fixtures or force sensors.","key_machinery":"The load-bearing object is the ellipsoidal limit surface of the finger contact patch, the boundary of all frictional wrenches the patch can exert. Its axes are fixed by contact geometry and friction, so enforcing gravity balance fixes the wrench scale without knowing grip force or friction coefficient. The motion direction comes from the maximal-dissipation rule $\\dot{q} \\propto A w_f$, and the paper's modification scales each planning step by $k = 1/(w_f^T A w_f)$, proportional to the limit-surface size, to reproduce slow near-rotational motion instead of an unrealistic constant-speed step. Vibration acts as the binary switch: on means sliding, off means sticking, which makes the planner's discrete subgoals executable.","core_discovery":"The paper's central claim is that millisecond-scale fingertip vibrations can serve as a control input for in-hand manipulation: they lower the effective friction at the contact fast enough that a parallel gripper can slide a grasped planar object using gravity alone, without external fixtures or force sensing. Concretely, the paper derives a quasi-static sliding model in which the object is always in force-torque equilibrium, so the friction wrench is fixed by gravity and contact geometry. With the ellipsoidal limit surface $w^T A w = 1$ and the direction rule $\\dot{q} \\propto A w$, the update $q_{t+1} = q_t + \\frac{A w_f}{\\|A w_f\\|} \\delta t\\, k$, $k = 1/(w_f^T A w_f)$, predicts the object's slip direction. The planner alternates between rotating the gripper without vibration and holding it fixed while vibrating, with pose feedback closing the loop. Across 60 real-world paths on six objects the final error is 5.91 mm in position and 1.21 degrees in orientation, and single vibration actions are predicted with 1.849 mm and 2.189 degrees RMSE.","pith_inferences":["Because the reported measurement noise is above 1 mm and the errors are 4-7 mm, a meaningful share of the residual error may be tracking noise; higher-resolution pose estimation is a natural test of the model's true accuracy.","The same friction-modulation idea could extend to multi-fingered hands or to variable-amplitude vibration for finer control; the paper leaves those as future directions.","A direct residual-velocity measurement after vibration stops would determine how much object mass or speed the quasi-static assumption can tolerate before a dynamic model is needed."],"forward_implications":["An ordinary two-finger gripper, without external fixtures or force sensing, can perform free-space in-hand reconfiguration of planar objects.","The limit-surface model predicts single vibration actions well enough for closed-loop control: 1.849 mm and 2.189 degrees RMSE on 1,564 actions.","Across 60 real-world trials on six objects, the planner's final pose error is 5.91 mm in position and 1.21 degrees in orientation, roughly 3-5% of object size.","The two observed motion modes, near-rotational and translational, justify the three-stage centering, positioning, and orientation-adjustment strategy.","Objects with non-homogeneous friction, such as 3D-printed textured surfaces, still converge within the reported error, so the method tolerates friction anisotropy."],"supporting_citations":[{"why":"Establishes the limit surface for a finger sliding under combined torsion and shear, grounding the wrench-space friction model.","marker":"[34]"},{"why":"Introduces limit surfaces for planar sliding of rigid bodies with dry friction, the basis of the quasi-static sliding prediction.","marker":"[35]"},{"why":"Supplies the ellipsoidal limit-surface approximation and the equivalent-radius torque relation used in the motion update.","marker":"[36]"},{"why":"Shows quasi-static pushing and sticking-interaction planning that the gripper primitive structure adapts to the vertical plane.","marker":"[17]"},{"why":"Gives the uniform-pressure constant for a patch contact that sets the torque axis of the limit surface.","marker":"[37]"},{"why":"Presents a vibration-finger parallel gripper for thin objects on horizontal planes, the direct predecessor this work moves to vertical free space.","marker":"[2]"},{"why":"Extends the vibration-fingertip gripper to full-state in-hand manipulation, providing the baseline for arbitrary pose reconfiguration.","marker":"[3]"},{"why":"Demonstrates adaptive pivoting with a parallel gripper using visual and tactile feedback, an earlier gravity-driven in-hand method limited to orientation.","marker":"[13]"}],"fun_headline_variants":["Vibrating fingertips slide objects in hand to sub-6 mm","Micro-vibrations reposition objects via gravity alone","Vibration-controlled friction slides grasped objects to 6 mm","Fingertip vibrations slide objects in hand, no fixtures","Sub-6 mm in-hand sliding with zero external fixtures"],"cache_read_input_tokens":15488,"weakest_assumption_plain":"Everything rests on the quasi-static assumption that the object slides only while vibration is on and comes to rest the instant it is switched off, with inertial forces negligible; the paper does not directly measure that stopping behavior.","fun_headline_variants_meta":{"raw":{"variants":["Vibrating fingertips slide objects in hand to sub-6 mm","Micro-vibrations reposition objects via gravity alone","Vibration-controlled friction slides grasped objects to 6 mm","Fingertip vibrations slide objects in hand, no fixtures","Sub-6 mm in-hand sliding with zero external fixtures"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001365,"raw_usage":{"total_tokens":5532,"prompt_tokens":938,"completion_tokens":4594,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":554,"completion_tokens_details":{"reasoning_tokens":4512}},"tokens_in":554,"tokens_out":4594,"duration_ms":33012,"temperature":1.0,"reasoning_tokens":4512,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T04:14:29.055699+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a high-speed camera or miniature IMU on a grasped object through repeated vibration-on/off cycles: if the object continues moving after vibration stops, or if a heavier object fails to slide at the 5 N grasp force, the binary stick/slide quasi-static model is violated.","supporting_citations":[{"cited_title":"The sliding of robot fingers under combined torsion and shear loading","cited_arxiv_id":null,"evidence_quote":"Establishes the limit surface for a finger sliding under combined torsion and shear, grounding the wrench-space friction model."},{"cited_title":"Planar sliding of a rigid body with dry friction: limit surfaces and dynamics of motion","cited_arxiv_id":null,"evidence_quote":"Introduces limit surfaces for planar sliding of rigid bodies with dry friction, the basis of the quasi-static sliding prediction."},{"cited_title":"Practical force-motion models for sliding manipulation","cited_arxiv_id":null,"evidence_quote":"Supplies the ellipsoidal limit-surface approximation and the equivalent-radius torque relation used in the motion update."},{"cited_title":"Stable push- ing: Mechanics, controllability, and planning","cited_arxiv_id":null,"evidence_quote":"Shows quasi-static pushing and sticking-interaction planning that the gripper primitive structure adapts to the vertical plane."},{"cited_title":"Modeling of contact mechanics and friction limit surfaces for soft fingers in robotics, with experimental results","cited_arxiv_id":null,"evidence_quote":"Gives the uniform-pressure constant for a patch contact that sets the torque axis of the limit surface."},{"cited_title":"Robotic manipulation of thin objects within off-the-shelf parallel grippers with a vibration finger","cited_arxiv_id":null,"evidence_quote":"Presents a vibration-finger parallel gripper for thin objects on horizontal planes, the direct predecessor this work moves to vertical free space."},{"cited_title":"Adaptive control for pivoting with visual and tactile feedback","cited_arxiv_id":null,"evidence_quote":"Demonstrates adaptive pivoting with a parallel gripper using visual and tactile feedback, an earlier gravity-driven in-hand method limited to orientation."}],"review_version":1}