{"id":"01b642cf-e924-44a2-b5fa-b17ec9cad929","arxiv_id":"2412.02613","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Operators can discriminate remote object stiffness using only force and displacement haptic feedback from a teleoperated robot hand, with displacement feedback offering a small, not consistently significant, advantage for similar-stiffness objects.","lead":"This paper tests whether people can judge how soft a remote object is while controlling a robot hand through an exoskeleton glove that pushes back on their fingers. Ten participants squeezed five silicone objects of different stiffnesses with no visual feedback, and discriminated them better than chance.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Uncontrolled auditory and other incidental cues from the robot hand, exoskeleton motors, and object contact could carry the stiffness discrimination, so the 'haptic feedback alone' claim is not yet isolated.","rationale":"I agree with the reader's conditional verdict and with the identified weakest assumption. The statistical problems noted by the reader (residual df = 1 in the ANOVA, Mann-Whitney U applied to paired data, uncorrected multiple comparisons, unexplained outlier exclusion) mainly undermine the secondary comparison between Method I and Method II; the primary above-chance claim could still hold despite those issues. The uncontrolled auditory channel is more load-bearing because it directly threatens the strongest claim that stiffness is perceived through haptic feedback alone. A replication with auditory masking is the cleanest way to settle this: if performance remains above chance under masking, the central claim survives; if not, the reported discrimination may be carried by incidental cues. Since the reader already recommended a conditional verdict, no verdict adjustment is needed.","tokens_in":12453,"tokens_out":3536,"duration_ms":41482,"concrete_test":"Repeat the Section IV protocol with an auditory-masking control condition: participants wear active noise-cancelling headphones playing continuous white noise, or the follower robot is placed in an acoustically isolated enclosure, for both feedback methods. Compare the ABX and S success rates against the original values using a paired one-sided test on participant means. If the masked condition still shows mean success rates significantly above 50% (e.g., around 74% for ABX and 65% for S with p < 0.05), the haptic-alone claim is supported; if the means drop toward chance, the original result is attributable at least in part to incidental auditory cues.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that operators perceive remote stiffness 'relying on haptic feedback alone and without any visual feedback.' The only sensory masking reported is visual: Section IV-C says 'a panel was used to hide the scene,' with no mention of auditory masking or isolation of the follower robot. The leader exoskeleton motors, the follower Allegro motors, and the contact events during squeezing all generate audible signals that are correlated with the measured contact force and therefore with object stiffness. Because the above-chance success rates in Section VI are the entire evidence for the central claim, the possibility that participants used these non-haptic cues is load-bearing. The concern is not about author conduct; it is that the experimental contrast does not yet separate haptic feedback from incidental correlated sensory cues.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a bilateral telemanipulation setup (HGlove leader exoskeleton, Allegro hand with tactile fingertips) and a 10-participant study comparing two haptic stiffness feedback methods: Method I, which renders force proportional to measured contact force, and Method II, which additionally scales the force by a leader/follower displacement ratio. Participants performed two-interval discrimination tasks (ABX similarity and softer-object identification) with vision hidden, and the reported group-mean success rates are about 75% for ABX and about 65% for Task S, above the 50% chance level. The authors conclude that operators can perceive remote object stiffness using haptic feedback alone and that the displacement component may help in the most difficult discriminations.","tokens_in":12594,"tokens_out":7600,"duration_ms":79431,"significance":"If the central claim is established, this is a useful contribution: to my knowledge it is the first participant study showing stiffness discrimination in real-world telemanipulation with a dexterous hand, tactile fingertips, and kinesthetic exoskeleton feedback, with no visual feedback. The study also compares two practical feedback formulations and addresses kinematic mismatch between leader and follower, which is a genuine problem in teleoperation. Strengths include the use of naive participants, balanced ordering of feedback methods, a repeated-trial design, and clearly reported mean performances per method and task. However, the experimental isolation of haptic feedback is incomplete, and several load-bearing statistical analyses are flawed, so the strength of the conclusions currently exceeds what the evidence supports.","major_comments":[{"comment":"The central claim that operators perceive stiffness 'relying on haptic feedback alone' is not yet isolated, because the only sensory cue removed is vision: Section IV-C states that 'a panel was used to hide the scene,' with no mention of auditory masking, physical separation of the follower robot, or a no-haptic control condition. The follower Allegro motors, the HGlove actuators, and the contact events during squeezing all generate audible and vibratory cues that are correlated with the measured contact force and therefore with object stiffness. A participant could plausibly perform above chance on Task ABX and Task S using these incidental cues alone. I request a control condition without haptic feedback, auditory masking (e.g., white-noise headphones), and/or explicit evidence that non-haptic cues were absent, before the 'haptic feedback alone' claim can be accepted.","section":"IV-C, VI"},{"comment":"The ANOVA in Table III is not a valid analysis of the experiment: with two groups, two days, and two tasks, the design has only eight aggregate cells, and the reported residual degrees of freedom is 1. This means the F-statistics and p-values are computed from a single residual degree of freedom and are uninterpretable, and the text also mislabels a three-factor model as a 'two-way ANOVA.' The conclusions drawn from this table (no significant group/day/task effects) are therefore unsupported. Please replace this with a mixed-effects model or repeated-measures ANOVA on participant-level trial data, with participant as a random effect.","section":"V-D, Table III"},{"comment":"The binomial calculations contain errors. For n=24, p=0.5, the exact probability of at least 16 correct is about 7.6%, not 10.6%, and the probability of at least 17 correct is about 3.2%, not 4.3%. The corrected values still support the qualitative conclusion that 17/24 is significant at the 5% level, but the reported confidence values (89.4% and 95%) should be corrected. More importantly, the manuscript moves from a participant-level binomial criterion to group mean success rates in Section VI; please clarify whether the 'all participants performed above 50%' claim is per participant or per group.","section":"V-A"},{"comment":"The treatment of the outlier is inconsistent and could affect the reported statistics. The text first states that 'all participants performed above the 50% chance level' and then states that a 'noticeable outlier, excluded from the statistical analysis' achieved 95% in Task ABX. If the outlier is one of the ten participants, then 'all participants' is ambiguous, and excluding a high-performing participant without a pre-specified criterion can bias the mean and variance estimates. Please report results with and without this participant, justify the exclusion rule (e.g., a defined outlier test), and state whether the outlier was excluded from Table II and Figs. 3-6.","section":"VI"},{"comment":"Method II is introduced as incorporating 'the squeezing displacement between the leader and follower devices,' but Eq. (9) is a force scaled by the product ΔZ·β = (ΔZ_L/ΔZ_F)(ΔZ_max,F/ΔZ_max,L). No derivation is given to show that this product is a stiffness estimate or that it constitutes 'displacement feedback' to the operator. The text in Section III-B2 refers to a real-time stiffness K_F,j in Eq. (6), but K_F,j is not used in Eq. (9). Please clarify the physical model, the units, and why this particular normalization is appropriate; as written, the second contribution is difficult to evaluate.","section":"III-B2, Eq. (9)"},{"comment":"The pair-by-pair Mann-Whitney comparisons involve roughly 14 tests (7 pairs × 2 tasks) with no correction for multiple comparisons. The single reported p=0.048 for pair (1-US, 4-LH) in Task S is therefore well within the range expected by chance, and the later near-significant p=0.061/0.067 values should not be interpreted as trends without an adjustment or a pre-specified analysis plan. The claim that displacement feedback 'may enhance discrimination' is correspondingly weaker than the current text suggests.","section":"V-E"}],"minor_comments":[{"comment":"There are typographical errors: 'classfication' in Section II and 'ANOV A' in Section V-D, and the text in Section V-D refers to 'Task X' where it should refer to 'Task S.'","section":"II, V-D"},{"comment":"The force thresholds F_min and F_max are introduced with numerical values but without units; please state the units (e.g., mN or N) and specify what happens when the measured force falls outside the valid range in Eq. (2).","section":"III-A"},{"comment":"Table I lists the sequence for Task ABX only; please clarify whether Task S used the same stimulus pairs and presentation order, or provide the corresponding sequence.","section":"IV-E"},{"comment":"The spider plots would be easier to interpret if the axis scale and success-rate range were explicitly labeled, and if the statistically significant pair were marked.","section":"Figs. 5-6"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a relevant and timely topic for the robotics/haptics community. The main scientific concern is the lack of control for non-haptic cues; this is fixable but requires additional experimental work, and the statistical reporting needs a thorough revision. I do not see grounds for rejection, but the manuscript is not ready in its current form."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The real contribution here is a working demonstration that stiffness discrimination is possible through a hand-exoskeleton teleoperation rig using tactile-derived force feedback, plus a new displacement-ratio feedback term (Eq. 9) that sensibly compensates for kinematic mismatch between leader and follower. The hardware setup is solid, the ABX/distance-level protocol is thoughtful, and the 10-person participant study is real work, not a simulation. The mean success rates of ~75% (ABX) and ~65% (S) are well above chance, so the core perceptual claim is probably true. No circular fitting: the thresholds and scaling constants are hardware-limited, not tuned to participant responses. Credit where due.\n\nNow the soft spots. The statistical reporting is genuinely bad. The ANOVA in Table III has residual df=1, which makes the F-tests meaningless for a repeated-measures design. The Mann-Whitney U tests are applied to paired data (each participant contributed both methods), which violates the test's independence assumption. Multiple comparisons across pairs and distance groups go uncorrected, so the single significant p=0.048 is not evidence. The unexplained exclusion of a high-performing outlier should at least be shown with and without.\n\nThe bigger conceptual issue is the 'haptic feedback alone' claim. Vision is masked with a panel, but there is no mention of masking auditory or other incidental cues. The exoskeleton motors, the Allegro hand, and the contact events all make sounds that correlate with force and therefore stiffness. Participants could have used those cues. That doesn't sink the paper—it still shows the system works in a realistic setting—but it does mean the strong claim that discrimination is achieved through haptic feedback alone is not yet isolated. A control condition with the haptic feedback turned off (or with white noise masking) would settle it.\n\nThe displacement feedback advantage for similar-stiffness objects is appropriately hedged ('suggest'), so I don't consider that a flaw; it's just not established.\n\nWho should read this: people designing teleoperation or telesurgery systems who need evidence on exoskeleton-based haptic feedback and a new compensation trick. The paper deserves a serious referee because it is a real system evaluation with a useful design idea and a likely-true central result. The referee should insist on a proper statistical reanalysis (paired tests, correction for multiple comparisons) and at least one control condition for non-haptic cues before publication. Send it to review, but expect major revision.","headline":"A genuinely useful teleoperation study with a new feedback law and a plausible result, but the 'haptic alone' claim needs a control for audible cues and the statistics need a proper redo.","tokens_in":13146,"tokens_out":2097,"would_cite":true,"duration_ms":23681,"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 claims that a person wearing a force-feedback exoskeleton glove can perceive the stiffness of remote objects by squeezing them with a teleoperated robotic hand, using only haptic feedback derived from tactile fingertip sensors…","keywords":["haptic feedback","stiffness perception","telemanipulation","hand exoskeleton","tactile sensing","ABX discrimination","kinesthetic feedback","teleoperation"],"falsifier":"Run the two tasks with the haptic rendering disabled or replaced by a sham force that is uncorrelated with measured stiffness, keeping the same hidden-scene setup and all incidental sounds identical; if success rates in that control drop to chance, the haptic channel is necessary, and if they stay above chance, the current experiment has not isolated haptic feedback.","tokens_in":12245,"feed_emoji":"🧤","tokens_out":11719,"duration_ms":108827,"temperature":0.7,"pith_summary":"Telemanipulation—using a robot to handle objects at a distance—is useful in remote or hazardous settings, but how soft an object is can be hard to judge through a camera. This paper tests whether force feedback generated from tactile sensing on a robot hand is enough for an operator to tell how soft an unseen object is. Ten naive participants wore an exoskeleton glove that drove a three-fingered robotic hand; contact forces measured by the robot's tactile fingertips were rendered back as squeezing forces on the operator's fingers while the scene was hidden. Participants distinguished five silicone objects of different stiffness with success rates well above chance in both an ABX similarity task and a softer-than comparison task. If the result holds, remote stiffness assessment can be achieved through touch alone, without visual feedback, which matters for palpation, telesurgery, and handling delicate objects in inaccessible environments.","feed_headline":"A hand exoskeleton lets operators feel remote object stiffness","feed_subtitle":"Ten users squeezed soft silicone samples through a robot hand and told them apart by touch alone, with up to 75% accuracy.","key_machinery":"The load-bearing mechanism is the sensor–actuator haptic mapping between the follower's tactile fingertips and the leader exoskeleton: the maximum normal force $F_{F,j}$ measured on each robot fingertip is scaled by $\\alpha = F_{L,\\max}/F_{F,\\max}$ and rendered as $F^1_{L,j}$ on the operator's finger. Method II augments this force term with the normalized displacement ratio $\\Delta Z = \\Delta Z_{L,j}/\\Delta Z_{F,j}$ and a kinematic-range normalization $\\beta$, producing $F^2_{L,j} = F^1_{L,j} \\cdot \\Delta Z \\cdot \\beta$. Since the paper defines object stiffness as $K_{F,j} = F_{F,j}/\\Delta Z_{F,j}$, Method II is effectively rendering a stiffness-weighted feedback signal, while Method I renders force alone; comparing the two is what isolates the contribution of the displacement information.","core_discovery":"The paper sets out to show that a bilateral telemanipulation system with an exoskeleton glove on the operator side and a dexterous robot hand with tactile fingertips on the remote side can convey object stiffness to a naive operator without any visual feedback. Ten participants squeezed five same-sized silicone samples of different Shore hardnesses (Ecoflex 00-10 through Dragon Skin 30) using the robot hand while the scene was hidden; they completed an ABX similarity task and a softer-than comparison task under two rendering methods. The measured contact forces from the robot fingertips were mapped to kinesthetic forces on the operator's fingers, with Method II additionally incorporating finger-displacement differences between the leader and follower. Average success rates were 74–75% in Task ABX and 64–68% in Task S, both above the 50% chance level. The paper concludes that haptic feedback from tactile sensing alone supports remote stiffness perception, with the displacement component helping mainly for objects of similar stiffness.","pith_inferences":["Not stated in the paper: the same follower-side signals could generate a continuous stiffness estimate in real time, turning ordinal \"which is softer?\" judgments into a numeric stiffness readout that the operator could feel directly.","Because the experiments did not mask auditory cues, a replication with haptic feedback disabled would clarify whether the observed discrimination is truly attributable to the haptic channel alone.","A testable extension would measure the smallest stiffness difference a user can reliably detect by using finer stiffness increments around each sample, checking whether displacement feedback lowers that threshold.","The displacement-normalization idea is not limited to this glove-and-hand pair; any leader–follower combination with different motion ranges could apply the same normalization and be tested against force-only feedback."],"forward_implications":["If the central claim is correct, force-proportional haptic feedback through a hand exoskeleton is sufficient for an operator to tell remote objects apart by stiffness without visual feedback, within the tested Shore hardness range.","Including displacement feedback does not consistently beat force-only feedback, so the displacement component should be treated as a task-specific aid rather than a universal improvement.","The above-chance discrimination suggests practical use in remote palpation and soft-object sorting, where visual access is limited or absent.","Because performance improved between the first and second experimental day, longer training could push discrimination accuracy higher.","Since displacement feedback showed its clearest benefit at the closest stiffness distance, it is most valuable when the objects being compared are nearly identical in compliance."],"supporting_citations":[{"why":"It supplies the leader-side exoskeleton glove that renders kinesthetic forces on the operator's fingers.","marker":"[14]"},{"why":"It validates the bilateral tactile telemanipulation setup and provides the preliminary force-based rendering that Method I uses.","marker":"[15]"},{"why":"It describes the force-sensitive fingertips on the robot hand, which are the source of the contact-force measurements for both feedback methods.","marker":"[35]"},{"why":"It provides the ABX discrimination testing paradigm used in Task ABX.","marker":"[41]"},{"why":"It supplies the guidelines on trial numbers that justify the 24-trial session design and the statistical confidence calculations.","marker":"[42]"},{"why":"It is the closest prior teleoperation result on force-feedback stiffness discrimination, against which the exoskeleton-hand setup is positioned.","marker":"[33]"},{"why":"It shows that a prior exoskeletal-glove teleoperation system provided limited force feedback and focused on grasp detection rather than stiffness differentiation, marking the gap this study fills.","marker":"[25]"}],"fun_headline_variants":["Feel remote object stiffness through a robotic hand","Operators perceive remote stiffness by touch alone","Haptic exoskeleton reveals remote object stiffness","Tactile feedback lets users feel remote stiffness","Exoskeleton teleoperation conveys remote stiffness"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The experiments block visual feedback but do not mask auditory or other incidental cues from the robot hand, exoskeleton motors, or object contact, so the above-chance discrimination could in principle come from non-haptic cues rather than from the rendered haptic feedback alone.","fun_headline_variants_meta":{"raw":{"variants":["Feel remote object stiffness through a robotic hand","Operators perceive remote stiffness by touch alone","Haptic exoskeleton reveals remote object stiffness","Tactile feedback lets users feel remote stiffness","Exoskeleton teleoperation conveys remote stiffness"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000381,"raw_usage":{"total_tokens":2014,"prompt_tokens":930,"completion_tokens":1084,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":546,"completion_tokens_details":{"reasoning_tokens":1016}},"tokens_in":546,"tokens_out":1084,"duration_ms":9262,"temperature":1.0,"reasoning_tokens":1016,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T23:16:34.651813+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the two tasks with the haptic rendering disabled or replaced by a sham force that is uncorrelated with measured stiffness, keeping the same hidden-scene setup and all incidental sounds identical; if success rates in that control drop to chance, the haptic channel is necessary, and if they stay above chance, the current experiment has not isolated haptic feedback.","supporting_citations":[{"cited_title":"Hglove: A wearable force- feedback device for the hand,","cited_arxiv_id":null,"evidence_quote":"It supplies the leader-side exoskeleton glove that renders kinesthetic forces on the operator's fingers."},{"cited_title":"Feeling good: Validation of bilateral tactile telemanipulation for a dexterous robot,","cited_arxiv_id":null,"evidence_quote":"It validates the bilateral tactile telemanipulation setup and provides the preliminary force-based rendering that Method I uses."},{"cited_title":"Leveraging symmetry detection to speed up haptic object exploration in robots,","cited_arxiv_id":null,"evidence_quote":"It describes the force-sensitive fingertips on the robot hand, which are the source of the contact-force measurements for both feedback methods."},{"cited_title":"Standardizing auditory tests,","cited_arxiv_id":null,"evidence_quote":"It provides the ABX discrimination testing paradigm used in Task ABX."},{"cited_title":"Abx discrimination task,","cited_arxiv_id":null,"evidence_quote":"It supplies the guidelines on trial numbers that justify the 24-trial session design and the statistical confidence calculations."},{"cited_title":"Evaluation of force feedback for palpation and application of active constraints on a teleoperated system,","cited_arxiv_id":null,"evidence_quote":"It is the closest prior teleoperation result on force-feedback stiffness discrimination, against which the exoskeleton-hand setup is positioned."},{"cited_title":"Intuitive and interactive robotic avatar system for tele-existence: Team snu in the ana avatar xprize finals,","cited_arxiv_id":null,"evidence_quote":"It shows that a prior exoskeletal-glove teleoperation system provided limited force feedback and focused on grasp detection rather than stiffness differentiation, marking the gap this study fills."}],"review_version":1}