{"id":"478c7b30-20a2-42e7-a3bf-e2a8330afcbb","arxiv_id":"2411.16055","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A curved passive end-effector can pick large objects by tilting them against a support and tucking underneath, demonstrated on multiple objects with simple hardware.","lead":"This paper shows a robot can pick up objects too large for a normal gripper by tipping them against a wall and sliding a curved 'palm' underneath. The method uses simple hardware and contact mechanics rather than a complex hand, and the authors demonstrate it on several boxes and containers.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Planning in Sec. IV-B assumes a monotonic rolling relation and a straight-line (θ,δ) path inside force-closure, but Sec. V-D admits the actual open-loop path differs because palm geometry is ignored; without measuring the executed path, the mechanics cannot support the reliability claim.","rationale":"The reader's weakest assumption—that the Sec. IV-B planning relies on monotonic rolling and a force-closure straight-line path—is the same concern that I find most load-bearing. The authors are transparent about the gap in Sec. V-D, and the experimental successes are real evidence that the overall concept works, which is why I do not move the verdict to rejection. The concern does not overturn the central claim; it identifies a missing validation step. The paper would be materially stronger if the executed trajectories were measured and compared against the force-closure region, and if the monotonic rolling assumption were checked directly. Since the reader already issued a CONDITIONAL verdict, my stress-test pass leaves that verdict unchanged while sharpening the specific requirement: the model-to-experiment link must be demonstrated, not assumed. No ad hominem is intended; this is a technical gap in the argument, not a claim about authorial conduct.","tokens_in":9045,"tokens_out":6028,"duration_ms":61174,"concrete_test":"Track object pose and the object-palm contact point during tilting using motion capture or vision for both the 2-DOF palm and the UR3 experiments, computing θ, δ, and palm orientation over time. Overlay the measured trajectory on the force-closure map from Fig. 3(b) for each object using measured friction coefficients. If measured paths leave the shaded region yet picking succeeds, force-closure is not necessary and the planning criterion is unsupported; if paths stay inside, the monotonic rolling assumption is empirically confirmed. A stronger variant is to compare the same object with 90° and 61° supports, checking whether the fixture change shifts the measured path into the force-closure region as the paper's post-hoc explanation claims. This test would settle whether the Sec. IV-B model is predictive or merely illustrative.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing assumption is the planning model in Sec. IV-B: \"It is expected that during tilting, the object rolls on the curved palm such that while θ is increasing, δ is decreasing monotonically,\" and that a straight-line path between initial and target configurations stays inside the sampled force-closure region (Fig. 3(b)). This relation is not derived from the palm profile; the palm's curvature determines how the contact point C moves along the object edge as the palm reorients, so δ(θ) is a nontrivial function of palm geometry that the paper never models. Sec. V-D explicitly concedes that the actual trajectory \"differs from the nominally feasible path\" because palm geometry is ignored. The experiments are teleoperated or executed open-loop, and no (θ,δ) measurements are reported, so there is no evidence that any executed trajectory stayed inside the force-closure region. The failure mode in Fig. 7(a) is exactly the predicted consequence of leaving that region: the object loses equilibrium and falls. Thus the theoretical planning does not explain the successful trials; the reliability claim rests on an empirical demonstration with manual control and a post-hoc fixture change (61° support) that is not derived from the model. This is load-bearing because the paper's central claim is that dexterous interaction can be realized with a passive curved palm, and the planning analysis is the claimed mechanistic support for that realization. The claim may well be true, but as presented the model-to-experiment link is unverified.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a nonprehensile picking method in which a curved, passive end-effector (the 'palm') tilts an object against two flat supports and then tucks underneath it to obtain a gravity grasp. The authors formulate the task as a quasistatic planar problem with three frictional contacts, classify two feasible tilting modes and one infeasible mode, and compute force-closure regions in a (theta, delta) configuration space to plan a tilting path. They demonstrate the method on a custom two-DOF palm and on a UR3 arm with several objects, reporting mostly 5/5 success rates, with one 0/5 condition for a steel box at a 90-degree support angle that was fixed by changing the support angle to 61 degrees. The paper concludes that simple hardware and control can realize the dexterous interactions needed for this form of picking.","tokens_in":9430,"tokens_out":3889,"duration_ms":38445,"significance":"If the result holds, the paper provides a useful addition to nonprehensile and extrinsic-dexterity manipulation: a passive curved palm, with no additional degrees of freedom or force/torque sensing, can pick objects too large for prehensile grasping by exploiting environment contacts. The mechanics analysis in Sec. IV gives a clear qualitative classification of tilting modes, and the force-closure plots in Fig. 3(b) offer a falsifiable prediction about which configurations should be stable. The hardware demonstration, including a conventional robot-arm implementation, is a meaningful proof of concept. However, the paper does not release code or raw experimental data, and the connection between the theoretical planning model and the open-loop experiments is not established, which limits the strength of the claimed reliability result.","major_comments":[{"comment":"The planning strategy in Sec. IV-B assumes that during tilting the object rolls on the curved palm so that delta decreases monotonically as theta increases, and that a straight-line path in (theta, delta) between initial and target configurations lies inside the force-closure region. Neither property is derived from the palm profile or from the contact model. Sec. V-D then explicitly concedes that the actual open-loop path differs from the nominally feasible path because the geometry of the palm is ignored. Since no executed (theta, delta) trajectories are reported, the experiments do not show that the object remained inside the force-closure region during tilting. This gap is load-bearing: without it, the theoretical analysis in Sec. IV-B does not explain the successful trials. The paper should either measure or estimate the executed paths, or explicitly reposition the planning analysis as a heuristic and support the empirical claim independently.","section":"Sec. IV-B and Sec. V-D"},{"comment":"The experimental evidence for reliable picking is thin: only 5 trials per condition are reported, the steel box at 90 degrees failed 0/5, and the subsequent fixture change to 61 degrees is explained only qualitatively via Sec. IV-A. The text says that reorienting support #2 redirects the contact normal downward, which predicts the direction of the improvement, but it does not provide a quantitative derivation of the 61-degree threshold, a parameter sweep, or trials across a broader range of support angles and object shapes. Given the small sample sizes, the success-rate differences between conditions (e.g., 3/5 versus 5/5) are not statistically meaningful, and the teleoperated execution makes it difficult to separate the method's robustness from operator skill. Please report confidence intervals, additional trials, or a clear statement that the empirical claim is at the level of a demonstration rather than a reliability characterization.","section":"Table I and Sec. V-C"},{"comment":"After presenting the force-closure planning method, the paper states that 'feasible initial and target configurations are chosen empirically' in the experiments. This creates a disconnect between the proposed planning blueprint and the actual experimental protocol. If the planning method was not used to select configurations, then the experiments do not validate the planning method; they only validate a manually chosen implementation of the general idea. The authors should clarify whether the force-closure computation was used at all in choosing the experimental trajectories, and if not, what role the planning section is intended to play in the claimed contribution.","section":"Sec. IV-B, experimental usage"}],"minor_comments":[{"comment":"The word 'ellipsical' should be 'elliptical' in the description of the palm's curved surface.","section":"Sec. V-A"},{"comment":"The phrase 'gently slammed' in the description of the initial contact is contradictory; consider replacing it with 'gently pressed' or 'brought into contact'.","section":"Sec. V-B"},{"comment":"The sentence 'Tilting supposedly begins (ends) when theta = 0 (delta = 0)' is vague; please define the initial and target configurations precisely, including the relation of these values to the object and palm geometry.","section":"Sec. IV-B"},{"comment":"The caption and text do not state the object dimensions or the exact friction coefficients used to generate the force-closure regions; adding these details would make the plot reproducible.","section":"Fig. 3(b)"},{"comment":"The claim that high-friction rubber on the palm 'seems sufficient' to attain high success rates is not tested by any ablation; comparing a lower-friction palm or a quantitative friction sensitivity study would strengthen this assertion.","section":"Sec. V-D"},{"comment":"The paper does not provide the LP force-closure software, experimental data, or trial logs; releasing these would substantially improve the reproducibility of the claimed success rates.","section":"Reproducibility"}],"recommendation":"major_revision","confidential_remarks":"The manuscript fits the journal's scope and addresses a real problem in nonprehensile manipulation. The central empirical demonstration is informative, but the gap between the planning model and the open-loop execution is significant and should be addressed in a revision. I do not think rejection is warranted, because the qualitative mechanics and the demonstration are still valuable; however, the authors should either measure the executed paths in (theta, delta) or explicitly reframe the planning section as a heuristic, and they should strengthen the experimental evidence with more trials and a quantitative treatment of the fixture-angle change. Please also consider requesting that the authors release their data and code."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things you should know. First, the paper demonstrates something real: a curved passive palm plus a wall support can pick objects too large for a parallel-jaw gripper, using simple hardware and no force-torque sensor. Second, the mechanics/planning section is not actually the explanation of the successful trials; the model assumes a monotonic rolling relationship and a straight-line path that the experiments never measure, and the paper itself concedes the executed path differs from the nominal one.\n\nWhat is new: the two-mode classification of initial tilting (two-contact wedged tilting vs three-contact sliding tilting) is a clean way to think about the problem, and the force-closure regions computed from rigid-body contact models give a useful qualitative picture. The target-configuration kinematic restraint analysis using the Reuleaux method is also nice and explains why palm orientation matters at the end. The experimental work is honest: varied objects, success rates mostly 5/5, and the authors report the 0/5 steel-box failure and the fix by changing the support angle to 61 degrees. The fixture change is qualitatively predicted in Sec. IV-A (redirecting the friction cone at support #2), so it is not arbitrary, but the specific angle and the success rates are empirical.\n\nThe soft spots are real. The planning in Sec. IV-B assumes the object rolls on the palm such that delta decreases monotonically as theta increases, and that the straight-line path lies inside the sampled force-closure region. But the palm geometry is never modeled, so delta(theta) is unconstrained, and no (theta, delta) measurements are reported. Sec. V-D openly says the actual path differs from the nominally feasible one. That means the force-closure analysis does not certify the executed trajectories; it is scaffolding, not mechanism. The demonstrated reliability comes from the experiments themselves, teleoperated or open-loop, with small trial counts, no baseline, and no code or data. The 3/5 for one steel box is a bit at odds with the word 'reliable' in the abstract, though the authors do not hide it. These are addressable concerns, not a fundamental blow.\n\nWho is this for: nonprehensile manipulation and low-cost logistics robotics. A serious referee should engage; the idea deserves publication with revisions that either close the model-experiment gap (measure theta-delta while tilting, or model the palm profile) or weaken the mechanistic claims.","headline":"A credible proof-of-concept for picking large objects with a passive curved palm and a wall, but the planning model is not tied to the executed paths, so the mechanics section is scaffolding rather than the explanation of the successful trials.","tokens_in":9900,"tokens_out":2979,"would_cite":true,"duration_ms":22513,"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":"A curved passive palm, helped by two flat supports, can reliably pick objects too large for prehensile grasping by tilting them and tucking underneath.","keywords":["nonprehensile manipulation","in-hand manipulation","tilting","force-closure","passive end-effector","quasistatic manipulation","object picking","curved palm"],"falsifier":"Run the tilt with a palm of known curvature while tracking the object's pose and contact point; if the measured (θ, δ) trajectory leaves the precomputed force-closure region before δ reaches zero, the object will fall exactly as in Fig. 7(a), disproving the sufficiency of the straight-line planning assumption.","tokens_in":8845,"feed_emoji":"🖐️","tokens_out":3988,"duration_ms":33424,"temperature":0.7,"pith_summary":"This paper tries to show that an object too large for any prehensile grasp can still be picked up reliably by a robot that uses only a curved, passive palm and two flat environmental surfaces. The key maneuver is to tilt the object against a wall-like support, creating a gap underneath, and then tuck the palm into that gap to support the object against gravity. If true, this would let simple hardware—just two motor axes or a conventional arm with a rigid curved attachment—perform a manipulation task usually thought to require a dexterous hand. The authors support the claim with a quasistatic planar analysis of tilting, a force-closure-based planning method, and picking experiments on objects of various sizes and shapes with two different robot setups.","feed_headline":"A tilted palm can pick objects too big to grasp","feed_subtitle":"Two flat supports plus a curved passive end-effector replace a gripper's fingers.","key_machinery":"The central object is the quasistatic planar contact model with three frictional point contacts, analyzed through force-closure via composite wrench cones and moment labeling. The configuration space (θ, δ)—the object's tilt angle and the location of the object-palm contact along the object's edge—is precomputed with a linear-programming force-closure test, and tilting is planned as a straight-line path between an initial and a target configuration lying inside the force-closure region. At the target configuration, first-order (Reuleaux) mobility analysis determines whether the palm's orientation kinematically prevents the object from ungrasping by clockwise rotation.","core_discovery":"The paper claims that the dexterous interactions needed to pick an object too large for prehensile grasping—tilting it up against a wall-like support and then tucking a curved passive palm underneath—can be realized with a zero-DOF rigid end-effector and simple control. It formulates picking as a quasistatic planar process with three frictional point contacts (object with bottom support, side support, and palm), analyzes two feasible tilting modes (two-contact wedged rotation and three-contact sliding), and plans a tilting path as a straight line in the (θ, δ) configuration space that stays inside the force-closure region. Experiments with a custom two-DOF palm and a conventional six-DOF arm demonstrate reliable picking of several objects, with reorienting the side support to a more acute angle (61°) resolving a failure mode where the object falls during tilting.","pith_inferences":["If the rolling assumption is the bottleneck, a closed-loop controller using tactile or proprioceptive feedback to track (θ, δ) could convert the open-loop blueprint into a robust plan without changing the hardware.","The same tilt-and-tuck template may generalize to other environment layouts, such as a wall corner and a ceiling, or to compliant objects where local contact curvature stabilizes tilting, as hinted by the cardboard box experiments.","The authors' comparison of support angles suggests a design principle for fixtures: making the blocking surface more acute turns an unreliable three-contact slide into a repeatable wedge-and-tilt maneuver.","A testable extension is to precompute the force-closure region for a library of palm profiles and choose a palm shape that maximizes the (θ, δ) area where tilting succeeds, since palm geometry is currently ignored."],"forward_implications":["Objects too large for parallel-jaw or multi-finger grips can be picked with a single curved rigid palm, using environmental surfaces as the other 'fingers.'","The technique works without force/torque sensors: a motion-force hybrid control (position on one motor, constant torque on the other) suffices for reliable tilting with fewer commands.","Reorienting the side support to a more acute angle redirects the side contact normal downward and suppresses the failure mode where the object falls before the palm tucks underneath.","High-friction palm material enlarges the set of configurations in which the object is force-closure during tilting, compensating for open-loop execution.","The planar quasistatic analysis extends to extruded or revolved inverted truncated cone objects, giving a class of 3D shapes for which the method is applicable."],"supporting_citations":[{"why":"Supplies the moment-labeling and Reuleaux methods used to analyze tilting feasibility and ungrasping restraint.","marker":"[23]"},{"why":"Provides the linear-programming force-closure test used to compute the (θ, δ) feasible region.","marker":"[24]"},{"why":"Establishes prior work on zero-mobility passive end-effectors for object picking, which this paper extends to a curved palm.","marker":"[4]"},{"why":"Reports the same failure mode of the object falling during tilting and presents hybrid force-velocity control, giving context for the failure analysis.","marker":"[21]"},{"why":"Defines dexterous ungrasping, used to analyze whether the object can escape from the target configuration.","marker":"[25]"},{"why":"Presents a tilt-and-pivot picking technique for thin objects, the closest prior tilting-based picking approach.","marker":"[22]"}],"fun_headline_variants":["Palm tilt trick picks oversized objects","Curved palm turns tilt into grasp","Tilt and tuck: picking without fingers","Oversized objects? Tilt and tuck","Robotic palm picks big by tilting"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The plan assumes that as the object tilts, the contact point on the palm moves monotonically so that δ decreases while θ increases, and that a straight-line path in the (θ, δ) space stays inside the force-closure region; the authors admit that the actual open-loop path deviates because the palm's curvature is not modeled.","fun_headline_variants_meta":{"raw":{"variants":["Palm tilt trick picks oversized objects","Curved palm turns tilt into grasp","Tilt and tuck: picking without fingers","Oversized objects? Tilt and tuck","Robotic palm picks big by tilting"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000418,"raw_usage":{"total_tokens":2129,"prompt_tokens":897,"completion_tokens":1232,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":513,"completion_tokens_details":{"reasoning_tokens":1165}},"tokens_in":513,"tokens_out":1232,"duration_ms":9215,"temperature":1.0,"reasoning_tokens":1165,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T13:35:54.144631+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the tilt with a palm of known curvature while tracking the object's pose and contact point; if the measured (θ, δ) trajectory leaves the precomputed force-closure region before δ reaches zero, the object will fall exactly as in Fig. 7(a), disproving the sufficiency of the straight-line planning assumption.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the moment-labeling and Reuleaux methods used to analyze tilting feasibility and ungrasping restraint."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the linear-programming force-closure test used to compute the (θ, δ) feasible region."},{"cited_title":"Object picking through in-hand manipulation using passive end-effectors with zero mobility,","cited_arxiv_id":null,"evidence_quote":"Establishes prior work on zero-mobility passive end-effectors for object picking, which this paper extends to a curved palm."},{"cited_title":"Robust execution of contact-rich motion plans by hybrid force-velocity control,","cited_arxiv_id":null,"evidence_quote":"Reports the same failure mode of the object falling during tilting and presents hybrid force-velocity control, giving context for the failure analysis."},{"cited_title":"Planning for dexterous ungrasping: Secure ungrasping through dexterous manipulation,","cited_arxiv_id":null,"evidence_quote":"Defines dexterous ungrasping, used to analyze whether the object can escape from the target configuration."},{"cited_title":"Picking thin objects by tilt-and-pivot manipulation and its application to bin picking,","cited_arxiv_id":null,"evidence_quote":"Presents a tilt-and-pivot picking technique for thin objects, the closest prior tilting-based picking approach."}],"review_version":1}