{"id":"9ed74bc8-d052-4611-ba10-6e823e9fd2f7","arxiv_id":"2607.16739","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A scan-to-fabrication pipeline that adds surface-only mutual-capacitance touch sensing to existing 3D objects with optimized drive/sense line layouts.","lead":"This paper presents a pipeline that turns existing 3D objects into touch-sensitive surfaces by scanning them, generating and cutting electrode patterns, and attaching them without modifying the interior. A generalist readership might care because it extends touch interaction to irreplaceable, everyday, or mass-produced objects that cannot be redesigned or reprinted.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Trace placement accuracy is never measured; Section 9.5 evaluates digital projection over the object, not the adhered copper traces, and the second layer is attached via landmarks only, so the ~2 mm localization error in Table 3 cannot be separated from layout-to-physical misalignment.","rationale":"The reader's weakest assumption—that physical copper traces end up where the design says they do—is exactly the load-bearing concern. The paper evaluates projection overlay accuracy (Table 4) but never measures the final placement of the adhered traces, and the second layer is attached using only small landmarks rather than projection (Section 7.3). Since localization reconstructs touch positions from the designed intersections, any unmeasured placement error directly contaminates the reported localization accuracy (Table 3). This is not a disagreement with the consensus or a circularity; it is an internal gap between the claimed quantitative result and the evidence provided. The concern is concrete and testable by measuring physical placement after attachment. The reader's CONDITIONAL verdict remains appropriate: the core pipeline demonstration is plausible and fabricated prototypes exist, but the quantitative localization claim needs this missing measurement. No other concern—SNR variance, untested multi-touch, absent artifacts—is as directly tied to the central quantitative claim as the placement-accuracy link. Hence the verdict should not change; the condition should be made explicit.","tokens_in":15891,"tokens_out":4795,"duration_ms":47227,"concrete_test":"On one fabricated object (e.g., the Stanford bunny), after attaching all traces, capture calibrated RGB images and segment the copper/Kapton traces; compute per-trace RMS distance between physical trace centerlines (or pad centers) and the projected design layout using the same registration as Section 7.3. Then repeat the 15-location localization protocol with ground truth defined by physically measured points on the object surface (e.g., a coordinate-measuring probe or rigid marker grid) rather than by designed intersections. If placement RMS is below ~1 mm and localization error remains ~2 mm, the concern is resolved. If placement RMS is 3-4 mm or localization tracks designed intersections instead of physical locations, Table 3 overstates physical accuracy and the pipeline needs a placement-calibration or measurement step.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing link in the central claim is that each capacitance measurement maps to the designed 3D intersection of drive curve a_i and sense curve b_j (Sections 6 and 8). This holds only if physical traces lie where the design places them. Section 7.3 states that projection guides the first layer, but for the second layer projection was found 'less helpful' and attachment relies on small printed landmarks. Section 9.5 and Table 4 measure projection accuracy (mean 3.0-4.4 mm, stdev up to 2.45 mm) of the reprojected virtual curves against annotated camera captures, not the post-adhesion positions of the copper traces. The localization evaluation in Table 3 reports errors of 0.65-1.60 mm, apparently against the designed layout. If physical traces are displaced by the measured projection-scale errors, the true physical localization error is at least as large as the placement error, and the '~2 mm' claim is an artifact of comparing sensed positions to intended rather than actual geometry. This is an internal-consistency gap: the reported localization precision cannot exceed the unmeasured physical placement precision.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a computational fabrication pipeline that retrofits existing 3D objects with surface-conforming mutual-capacitance touch sensing. The pipeline scans a real object to obtain a mesh, generates two layers of surface-intrinsic curves via discrete trivial connections, selects subsets of drive/sense curves and electrode-pad locations through two ILP stages, unfolds the selected curves into 2D stencils, cuts them from copper laminate, attaches them with projection guidance, and connects them to a mutual-capacitance digitizer for real-time touch localization. The approach is demonstrated on four fabricated objects (bunny, mouse, plate, spoon) with quantitative SNR, localization, and projection-accuracy measurements, plus optimization studies on ten additional objects.","tokens_in":16192,"tokens_out":3200,"duration_ms":34609,"significance":"If the pipeline works as claimed, it is a useful contribution to computational fabrication and touch sensing: it avoids internal modification, supports arbitrary genus-0 geometries, and provides a complete surface-only workflow from scan to interactive object. The paper's strengths include a detailed, reproducible description of the geometric and optimization pipeline; an ablation of the secondary objective in Appendix B; and quantitative evaluations on multiple physical objects. The central claims, however, depend on unmeasured physical placement accuracy and on SNR statements that are not supported by the reported aggregate statistics. These issues are fixable and do not invalidate the overall approach, but they must be addressed before the claims can be accepted.","major_comments":[{"comment":"The localization accuracy claim (~2 mm or less, Table 3) is not supported because the physical positions of the adhered copper traces are never measured. The localization pipeline (Section 8) maps each capacitance measurement to the designed intersection of drive/sense curves; this mapping is only valid if the traces end up where the design places them. Section 7.3 states that projection guides the first layer but that the second layer relies on printed landmarks, and Section 9.5/Table 4 reports projection accuracy of mean 3.0–4.4 mm with standard deviations up to 2.45 mm. These placement-scale errors are comparable to or larger than the reported localization errors (0.65–1.60 mm), so the true physical localization error is at least as large as the placement error. Please measure post-adhesion trace positions (e.g., via the same camera/projection setup) or explicitly report localization","section":"§7.3, §9.5, Tables 3 and 4"},{"comment":"The statement 'All our SNR measurements surpassed both thresholds by a wide margin' is contradicted by the aggregate data. Table 2 reports mean SNR values of 52.51–95.55 with standard deviations of 32.68–81.42. Under any reasonable distribution, a substantial fraction of the 15 locations × 10 repetitions must fall below the recommended threshold of 15, and likely below the minimum threshold of 7. For example, the spoon has mean 52.51 and stdev 51.04, implying a large lower tail. Reporting only mean/stdev is insufficient; please provide per-location minima, quartiles, or the full distribution so the threshold claim can be verified.","section":"§9.3, Table 2"},{"comment":"The abstract and contributions claim 'multi-touch interaction' and Section 8 states that interpolation enables multi-touch, pinches, and swipes. However, the localization evaluation in Section 9.4 uses only single touches (15 locations, 10 repetitions each) and reports no multi-touch results, accuracy, or even a demonstration of simultaneous contact. Since multi-touch is a central capability claim, please either add a multi-touch evaluation (e.g., two-finger localization error or a qualitative demonstration with distinct contacts) or temper the claim to single-touch localization with multi-touch as a system capability.","section":"§8, §9.4"}],"minor_comments":[{"comment":"The threshold values for the non-sensing base region (normal-deviation angle and geodesic distance) and the number of seed points / rejection thresholds are not specified. As these are free parameters that affect the layout, please report the exact values used for the four fabricated objects.","section":"§5.1"},{"comment":"The description of surface sampling for the coverage objective says 'uniformly sample 300 points' with 'distance-threshold rejection,' but the rejection criterion is not defined. Please clarify how the threshold is chosen.","section":"Appendix A"},{"comment":"The plate's Voronoi CV is 0.594, which is substantially higher than the other objects, yet the text calls the layouts 'highly uniform.' The discussion attributes this to boundary effects; please provide a quantitative justification or soften the claim accordingly.","section":"Table 1 / §9.2"},{"comment":"The statement 'no observed false positives or negatives' is based on 150 touches per object; please state the total number of events and the test procedure, as this claim is strong relative to the sample size.","section":"§9.4"}],"recommendation":"major_revision","confidential_remarks":"The core pipeline appears sound and the ablation is a positive sign. The main blocker is the unmeasured physical placement accuracy, which undermines the quantitative localization claim as written. The SNR claim also needs correction. If the authors measure trace placement or reframe the claims, the paper could become acceptable. I would not reject because the issues are empirical gaps, not fundamental flaws."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know: this is a real contribution, not a repackaging. The combination of trivial-connection tracing with a coverage ILP yields layouts that previous work (Multitouch-Kit, Palma et al.) can't produce without developable assumptions or internal routing. Four fabricated objects, quantitative SNR and localization numbers, and an honest discussion of limitations. The strongest parts are the geometric formulation and the optimization; the weakest is the evaluation of physical placement.\n\nSpecifically, the stress-test note lands. Section 9.5 measures projection accuracy of the virtual curves (3–4.4 mm mean), not the post-adhesion positions of the copper traces. The second layer is attached by printed landmarks rather than projection (Section 7.3). Table 3's localization errors (0.65–1.60 mm) are computed against the designed intersections. If the physical traces deviate by the measured projection-scale errors, the true physical localization error is bounded below by that placement error. The authors never measure it. That gap doesn't break the pipeline—the objects clearly sense touches—but it does mean the headline 'Rmm localization' is a digital-model claim, not a physical one.\n\nTwo smaller issues. First, the SNR claim: Table 2's means are high but stdev is large (52 mean / 51 stdev for the spoon), so 'all measurements surpassed thresholds by a wide margin' is not justified from aggregate stats. Show per-location distributions. Second, multi-touch is claimed throughout but the localization eval is single-touch only. Interpolation to centroids is mentioned, not validated.\n\nThe ablation in Appendix B is good evidence that the secondary objective matters; the citation pattern looks fair. No code/data, which limits reproducibility for a fabrication pipeline; that should be a revision item, not a reject.\n\nWho this is for: anyone working on computational fabrication, capacitive sensing, or retrofitting interactive surfaces. It deserves a serious referee and likely a conditional accept given the gaps. I'd send it to review with a request for placement-accuracy measurement, per-condition SNR reporting, and a multi-touch interaction test.","headline":"Genuine advance in surface-only mutual-capacitance retrofit, but the headline localization claim is against the designed layout, not the physically placed traces; solid systems contribution that needs an evaluation revision.","tokens_in":16722,"tokens_out":2216,"would_cite":true,"duration_ms":24756,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A computational pipeline can retrofit existing 3D objects with surface-only mutual-capacitance touch sensing, achieving real-time multi-touch with localization errors around 2 mm or less.","keywords":["mutual-capacitance sensing","3D object retrofit","surface-conforming electrodes","intrinsic curve tracing","integer linear programming","projection-guided assembly","multi-touch interaction","computational fabrication"],"falsifier":"Measure the actual 3D positions of the adhered traces on one fabricated object (for example, by high-resolution scanning or annotating camera images) and compare them to the designed intrinsic layout. If the placement error approaches the electrode pad spacing set by the minimum-distance constraint, the localization mapping that assigns each capacitance measurement to its designed intersection will shift touches to incorrect surface locations.","tokens_in":15765,"feed_emoji":"🖐️","tokens_out":3210,"duration_ms":35233,"temperature":0.7,"pith_summary":"The paper tries to establish that an existing physical 3D object can be turned into a multi-touch surface without modifying its interior: scan the object into a mesh, generate two families of surface-intrinsic electrode curves, optimize their intersections for uniform coverage under hardware and fabrication constraints, cut the unfolded curves from copper foil, attach them with projection guidance, and resolve touches in real time. If true, this shifts touch sensing from a feature that must be manufactured into the object to a retrofit that preserves the object's structure and appearance except for the attached traces. The authors report signal-to-noise ratios far above recommended thresholds and mean localization errors between 0.65 mm and 1.60 mm across four fabricated demonstrators.","feed_headline":"Copper curves turn everyday 3D objects into multi-touch surfaces","feed_subtitle":"Scan, optimize, cut, and project: four objects gained real-time touch with localization under 2 mm.","key_machinery":"The central mechanism is the mutual-capacitance grid defined by two layers of surface-intrinsic curves: drive (Tx) and sense (Rx) lines that intersect at most once per pair, each intersection serving as a unique sensing location. Curve generation uses discrete trivial connections—a way to transport tangent directions across mesh edges without a global parameterization—so that curve families conform to arbitrary genus-0 surfaces. Two integer linear programming stages then select a hardware-limited subset of curves and place electrode pads with a minimum separation, and a strip-based unfolding (Mitani–Suzuki) converts each 3D curve into a planar outline for vinyl cutting.","core_discovery":"The central claim is that spatial multi-touch sensing can be computationally designed as a retrofit for a given 3D surface. The method treats each conductor as an intrinsic curve on the surface, with touch locations at intersections between drive and sense curves, and casts layout generation as two integer linear programs: one selects curve subsets that keep intersections evenly distributed and within the scanning controller's channel limits, and the other picks electrode pad positions separated by a minimum distance to avoid interference. Fabrication unfolds each selected curve into a 2D stencil, cuts it from a three-layer Kapton–copper–Kapton laminate with a vinyl cutter, and guides manual","pith_inferences":["The reported localization accuracy implicitly assumes the manually attached copper traces land where the projected design puts them; the paper measures projection overlay accuracy but not final trace placement, so an independent measurement of adhered-trace deviation would bound the claimed ~2 mm error.","If conductive paints or hydrographic transfers replaced copper foil, the surface appearance would be less occluded, but their sheet resistance and adhesion would likely change signal integrity; this is a testable extension the paper mentions as future work.","The optimization currently ignores electrical impedance and parasitic capacitance of long or tightly curved traces; integrating electrical models could reveal trade-offs between uniform coverage and signal strength that are not captured by geometric uniformity alone.","For non-genus-0 shapes or narrow protrusions (like the bunny's ears), additional singularities or reseeding would be needed; the paper notes coverage can be sparse there, so evaluating those cases would clarify the method's practical generality."],"forward_implications":["Any genus-0 object with a clear base can in principle be instrumented without 3D printing, internal wiring, or redesign.","The layout optimization cleanly separates geometric coverage goals from hardware limits, so changing the scanning controller or desired resolution only changes the channel-count constraints.","Standard touch interactions—tap, swipe, pinch, multi-touch—can be mapped onto curved surfaces with millimeter-level localization, as demonstrated on the four prototypes.","The pipeline makes interactive input available on unique or irreplaceable objects whose geometry and interior must remain intact, extending computational fabrication from creating interactive objects to augmenting existing ones."],"fun_headline_variants":["Scan an object, cut copper curves, get multi-touch","Adding touch to any 3D object: scan, design, cut, attach","Copper curves retrofit 3D objects for real-time touch","Make any 3D shape touch-sensitive with copper stencils","Multi-touch on existing 3D objects via computational sensor design"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing assumption is that the manually attached copper traces end up exactly where the designed layout says they should be, since every capacitance reading is mapped to a precomputed curve intersection; if the physical traces deviate, the touch locations are wrong.","fun_headline_variants_meta":{"raw":{"variants":["Scan an object, cut copper curves, get multi-touch","Adding touch to any 3D object: scan, design, cut, attach","Copper curves retrofit 3D objects for real-time touch","Make any 3D shape touch-sensitive with copper stencils","Multi-touch on existing 3D objects via computational sensor design"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00029,"raw_usage":{"total_tokens":1503,"prompt_tokens":685,"completion_tokens":818,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":429,"completion_tokens_details":{"reasoning_tokens":726}},"tokens_in":429,"tokens_out":818,"duration_ms":7415,"temperature":1.0,"reasoning_tokens":726,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T20:04:28.430565+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the actual 3D positions of the adhered traces on one fabricated object (for example, by high-resolution scanning or annotating camera images) and compare them to the designed intrinsic layout. If the placement error approaches the electrode pad spacing set by the minimum-distance constraint, the localization mapping that assigns each capacitance measurement to its designed intersection will shift touches to incorrect surface locations.","supporting_citations":[],"review_version":1}