{"id":"b94c4443-4504-4cf0-b0d1-f596bc567fa8","arxiv_id":"2608.10387","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A hip-actuated monoped can stabilize pitch and energize its hop with the same torque, and its steady-state gait has closed-form fixed points and eigenvalues from hybrid averaging, validated on the Penn Jerboa robot.","lead":"This paper shows that a hopping robot can use the torque from keeping its body upright to also add energy to its hop, giving simple formulas for the resulting speed and height. The authors demonstrate steady hopping on the Penn Jerboa robot at speeds up to 1.77 m/s, with an analysis predicting that such hip-energized hopping must use asymmetric leg placement.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1's stability conclusion is conditional on Conjecture 1, which asserts the continuous-time exponential-stability and BIBO property the cascade argument needs; no verification of this conjecture is supplied.","rationale":"The reader's weakest-assumption analysis identifies Conjecture 1, and my reading of Section 2.3.10 and the proof of Theorem 1 confirms that this conjecture is the place where the argument moves from discrete return-map stability to continuous-time cascade stability. The paper is unusually explicit about the gap: it labels the needed property as a conjecture, explains that hybrid exponential stability is not automatic from return-map hyperbolicity, and uses fixed-time flight and stance assumptions that make the averaging setup transparent. The simulation and hardware results give empirical support for the overall controller, but they do not validate the conjecture itself, because observed stable hopping could result from a larger basin even if the local continuous-time BIBO statement fails. The proposed Floquet-plus-impulse check directly targets the two parts of the conjecture: transverse contraction of the limit cycle and boundedness of the response to small continuous-time disturbances. If the check passes, the conditional theorem becomes a proven one; if it fails, the proof needs repair. Since the reader already returned CONDITIONAL, my stress-test does not change the verdict.","tokens_in":49475,"tokens_out":8548,"duration_ms":115923,"concrete_test":"Compute the monodromy (Floquet) multipliers of the unperturbed 2-DoF SLIP limit cycle at the fixed point q_s* in (30b) using the stance dynamics (1a)-(1b), the stepping controller (22), and the reset map (19), including saltation matrices at the touchdown and liftoff guards. If any nontrivial multiplier has modulus >=1, Conjecture 1 is false. If all multipliers are inside the unit circle, then test the BIBO portion by adding a small impulsive disturbance to \\dot r at mid-stance and checking that the deviation from the limit cycle decays exponentially over the following strides; failure to decay would show that the continuous-time property in Conjecture 1 does not follow from the return-map analysis alone.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Theorem 1 invokes Conjecture 1 as an essential bridge. Proposition 2 proves stability of the unperturbed translational return map S_phi* in discrete liftoff coordinates, but Theorem 1 must also absorb the exponentially decaying pitch-state disturbance that acts continuously during stance. That requires the unperturbed 2-DoF limit cycle to be locally exponentially stable in continuous time and to have the bounded-input/bounded-output property stated in Conjecture 1. The authors explicitly note in Section 2.3.10 that exponential stability of the return map does not automatically imply exponential stability of the hybrid limit cycle, which is why the conjecture is needed. The conjecture is neither proved nor numerically checked; the paper's simulations and hardware runs demonstrate that the full closed-loop system can hop stably, but they do not isolate the property asserted in Conjecture 1. If Conjecture 1 fails, the formal asymptotic-stability guarantee in Theorem 1 collapses even though the controller might still work in practice. The standard total-stability argument for the cascade would need either this continuous-time ISS property or an explicit discrete-time ISS proof, and neither is supplied independently of the conjecture.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a hip-only control strategy for pitch-unlocked planar monopedal hopping, in which the pitch-stabilizing hip torque also injects energy into the translational SLIP dynamics. The stance dynamics are decomposed into a 1-DoF pitch subsystem cascading into a 2-DoF SLIP subsystem. Closed-form fixed points and eigenvalue estimates are derived via hybrid averaging, with a new coordinate transformation treating SLIP as a unitary 2-DoF system. Theorem 2 establishes that hip-energized period-one gaits require asymmetric stepping. The analytical predictions are compared against simulations of a 5-link biped and a model of the Penn Jerboa, and against physical experiments on the Penn Jerboa, reporting stable hopping at 1.02--1.77 m/s. The formal stability result (Theorem 1) is explicitly conditional on an unproved conjecture about the continuous-time exponential stability and BIBO stability of the unperturbed SLIP limit cycle. The free parameters gamma and chi are fitted to the same simulation and hardware data used for validation, so the reported accuracy is calibrated rather than predicted.","tokens_in":49741,"tokens_out":3716,"duration_ms":43617,"significance":"If the central claims hold, the paper makes a valuable contribution to legged locomotion: it demonstrates that a single hip actuator can simultaneously stabilize pitch and replenish translational energy, gives closed-form expressions for the resulting fixed points and eigenvalues, and documents fast, sustained hopping on a physical robot. The hybrid-averaging extension to a full 2-DoF SLIP with attitude is a methodological advance, and Theorem 2's necessity result for asymmetric stepping is a clean conceptual insight. The paper is unusually transparent in listing its assumptions and in stating Conjecture 1 as an unproved hypothesis. However, the headline formal stability guarantee is conditional on that conjecture, and the quantitative validation is in-sample because gamma and chi are fitted to the data being compared. These limitations materially temper the strength of the stated conclusions.","major_comments":[{"comment":"Placeholder","section":"Section 2.3.10, Theorem 1"},{"comment":"Placeholder","section":"Section 3.4, Tables 9 and 10"}],"minor_comments":[{"comment":"Placeholder","section":"Abstract and Introduction"},{"comment":"Placeholder","section":"Abstract, Section 1.2"},{"comment":"Placeholder","section":"Section 2.3.6, footnote 17"},{"comment":"Placeholder","section":"Section 6.1, Table 8"}],"recommendation":"major_revision","confidential_remarks":"The paper is a strong empirical and analytical contribution, but the formal theorem is conditional on an unverified conjecture and the validation is in-sample. These issues are fixable within the scope of the manuscript by adding verification of Conjecture 1 and by re-labeling or re-analyzing the validation protocol. The novelty claim relative to prior VPP-based work is appropriately careful, though the comparison to tail-energized hopping in Section 6.7 should be qualified by the different experimental setups."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The real news here is a hip-only monoped that hops 5–9 leg lengths/s on hardware, with closed-form fixed-point formulas and a new theorem (Theorem 2) proving that hip-energized period-1 gaits require asymmetric stepping. The hybrid-averaging treatment of the full 2-DoF SLIP with attitude is a genuine extension of De's framework, and the paper is honest about what is assumption, what is conjecture, and what is post-hoc fitting. The hardware videos and the fourfold energy range across two morphologies give the qualitative story real weight.\n\nThe soft spots are exactly where the reader put them. Theorem 1's asymptotic-stability guarantee is conditional on Conjecture 1, which asserts the continuous-time exponential stability and BIBO property that the cascade argument needs. That conjecture is neither proved nor numerically checked, and the paper says so. The stability guarantee is therefore real only up to that unverified bridge. I agree with the stress-test note: this is load-bearing, not a cosmetic gap. Separately, gamma and chi are fitted to the same simulation and hardware fixed points that Tables 9 and 10 then compare against, so the 6–16% error numbers are internally calibrated, not externally validated. The authors flag this in Section 3.4, but it means the quantitative predictive claim is weaker than a first glance at the tables suggests. The hardware also uses supplemental discrete controllers (energy loop, feedforward torque) that sit outside the formal analysis; again, this is disclosed, and the paper argues they do not move the fixed points, but the formal certificate does not cover them.\n\nNone of this kills the paper. Theorem 2 is a structural result that stands on its own, the fixed-point formulas give designers a genuinely useful control-to-gait map, and the empirical demonstration is solid for a platform with this little actuation. The caveats are proportionate because the paper itself is unusually candid about them.\n\nThis paper deserves a serious referee. The right revision would either prove Conjecture 1 in a restricted setting or replace the conditional claim with the actual (still worthwhile) statement: the return map is exponentially stable at the averaged fixed point, and simulations/hardware show stable hopping in practice. I would also ask for a version of Tables 9 and 10 that separates calibration data from any genuinely predictive check.\n\nBring it to reading group; cite it for the hybrid-averaging extension and Theorem 2.","headline":"Genuine advance in hip-energized hopping with a first hybrid-averaging analysis of 2-DoF SLIP with attitude, but the flagship stability theorem leans on an unproved conjecture and the model accuracy is post-fit.","tokens_in":50207,"tokens_out":1450,"would_cite":true,"duration_ms":20169,"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":"This paper claims that a single hip torque can simultaneously stabilize body pitch and pump energy into the leg spring, sustaining steady monopedal hopping at user-selected speeds and heights, and it backs the claim with closed-form…","keywords":["monopedal hopping","hip-energized control","SLIP","pitch stabilization","hybrid averaging","stepping control","underactuated legged locomotion","Penn Jerboa"],"falsifier":"Simulate or instrument the full 3-DoF closed-loop system and measure the local return-map Jacobian at the predicted fixed point $q^*$ (30) for several small $\\epsilon$ values; if any eigenvalue leaves the unit circle, or the numerically exact fixed point is not $O(\\epsilon)$-close to the predicted one, Theorem 1's conclusion fails. To test Conjecture 1 directly, perturb the unperturbed SLIP limit cycle with small continuous disturbances and check that the linearized response decays; a marginal or growing linearized cycle would falsify the formal stability claim. Empirically, force a symmetric stepping law (touchdown angle exactly the negative of liftoff angle) on the hardware hopper, since Theorem 2 predicts sustained period-one hip-energized hopping then becomes impossible.","tokens_in":49266,"feed_emoji":"🦘","tokens_out":7424,"duration_ms":77662,"temperature":0.7,"pith_summary":"This paper claims that a single hip torque can do two jobs at once in a planar monoped: keep the body from pitching over, and pump enough energy into the springy leg to sustain steady hopping. By displacing the center of mass and tuning a feedforward torque bias, the reaction torque that counters pitching becomes a controllable energy source for the translational spring-loaded inverted pendulum (SLIP) motion. A stepping rule that lands at an intentionally asymmetric touchdown angle redistributes energy between the radial and angular directions, replacing what damping removes. The paper derives closed-form expressions for the gait's fixed points and eigenvalues, proves local stability of the 3-degree-of-freedom return map (modulo one explicit conjecture), and reports hardware hopping from 1.02 to 1.77 m/s, or 5.1 to 8.85 leg lengths per second.","feed_headline":"One hip motor sustains hopping to 8.85 leg lengths/s","feed_subtitle":"Pitch stabilization doubles as the energy pump, and a closed-form model predicts the hopping fixed points within 16 percent.","key_machinery":"The load-bearing machinery is a cascade decomposition plus a change of coordinates. The paper first isolates the 1-DoF pitch dynamics, using Assumption 5 to replace the leg-force moment by a constant $\\chi$-scaled disturbance, so the pitch loop with its discrete torque-bias integrator becomes an LTI hybrid system with an explicit stable fixed point (13). That fixed point feeds a constant hip torque $\\bar{\\tau}^*$ into the otherwise unactuated 2-DoF SLIP, whose stance dynamics are then re-expressed in coordinates (14) with master phase $\\psi_r$, energy ratio $\\psi_e$, leg angle $\\theta$, and energy $a_e$; hybrid averaging, a technique for approximating Poincar\\'e return maps when fast phases and slow energy states coexist, yields the averaged field (17), fixed point (18), and eigenvalue estimates (27)-(29). The stepping law (22), a filtered Raibert-style law augmented by a term $-\\pi/\\omega_e(a)$ that encodes the required asymmetry, keeps $\\psi_e$ near its target and stabilizes the leg angle. Theorem 2 completes the picture by proving that any hip-energized period-one gait must land asymmetrically, which is why the extra term is a necessity rather than a hack.","core_discovery":"On the authors' own terms, the central discovery is that pitch stabilization and translational energy injection are not competing uses of a single hip actuator but one combined mechanism: the PD-plus-feedforward torque that holds the body near a desired pitch exerts a reaction force on the leg that adds energy to the SLIP subsystem at a rate proportional to the steady torque bias $\\bar{\\tau}^*$. The paper formalizes this by decomposing the 3-DoF pitch-unlocked SLIP into an isolated linear pitch subsystem cascading into a 2-DoF translational SLIP subsystem, then applies hybrid averaging in new coordinates, the energy ratio $\\psi_e$ and twice the mass-specific square-root energy $a_e$ of (14), to obtain the averaged fixed point $\\hat{x}^*$ (18) and eigenvalue approximations (27)-(29). Theorem 1 assembles these into explicit fixed points $q^*=[q_a^*, q_s^*]$ (30) and asymptotic stability for small $\\epsilon$ under the assumptions in Tables 3 to 5, assuming Conjecture 1 on exponential stability and BIBO stability of the underlying limit cycle. The result predicts that stable, user-selectable fore-aft speed and apex height can be tuned through the COM offset $d_x$ and target energy ratio $\\psi_e^d$, with simulated and physical data matching the predictions at roughly 6 to 16 percent mean error.","pith_inferences":["Our inference: if any controller injects a nonzero net hip torque per stride, Theorem 2's asymmetry argument should apply beyond this specific law, so virtual-pivot-point-style torque laws that stabilize pitch may also require asymmetric foot placement to avoid unbounded angular energy.","A testable extension: replace the fitted constants $\\gamma$ and $\\chi$ with state-dependent estimates; the paper notes that its fixed-point accuracy depends on these fitted values, so an online estimator could broaden the operating regime where the closed-form map predicts hardware behavior.","Our inference: the same cascade decomposition may transfer to spatial robots by treating yaw and roll as isolated attitude subsystems perturbing a planar SLIP, potentially yielding closed-form fixed points for more than sagittal-plane motion, although the paper does not claim this."],"forward_implications":["A monoped can hop stably using only the hip motor during stance; no shank actuator is needed to restore energy lost to damping.","Steady-state fore-aft speed and apex height become user-selectable through two scalar setpoints, the COM offset $d_x$ and target energy ratio $\\psi_e^d$.","Hip-energized hopping that returns to the same state each stride must use an asymmetric stepping strategy; symmetric neutral-point stepping cannot work when the hip injects nonzero net energy.","The closed-form eigenvalues give explicit gain-scheduling guidance: $k_e$ governs energy-ratio convergence, $\\alpha$ governs leg-angle damping, and the energy eigenvalue shows why high-energy setpoints lose stability.","On the Penn Jerboa the strategy sustains speeds from 1.02 to 1.77 m/s, about 5.1 to 8.85 leg lengths per second, with the analytical model predicting the observed fixed points to roughly 6 to 16 percent mean error."],"supporting_citations":[{"why":"Supplies the hybrid averaging theorem that yields epsilon-close fixed points and stability guarantees for the return map; it is the central analytical method of the paper.","marker":"De et al. (2018)"},{"why":"Provides the multi-phase averaging checklist that the paper extends to systems not satisfying Definition 1(iii).","marker":"De (2017)"},{"why":"Classic SLIP return map and scissor-stepping framework; the stepping law (22) is a Raibert-style law with an added asymmetric term.","marker":"Raibert (1986)"},{"why":"Prior hip-torque-energized SLIP numerical study whose nearly constant open-loop torque inspires the stance torque law, and serves as a comparison baseline.","marker":"Ankarali and Saranli (2010)"},{"why":"Identified the virtual pivot point phenomenon, the only prior account of hip torque stabilizing pitch while energizing hopping; the paper compares to it and shows its controller yields a VPP.","marker":"Maus et al. (2010)"},{"why":"Introduces the Penn Jerboa platform and tail-energized hopping results, providing the hardware comparison baseline and the robot model parameters.","marker":"Shamsah et al. (2018)"},{"why":"Introduces the SLIP template used as the translational model for the unperturbed 2-DoF subsystem.","marker":"Saranli et al. (1998)"},{"why":"Provides the small-angle approximations and spring-mass running analysis used in Assumptions 7 and 9 to decouple the pitch and SLIP subsystems.","marker":"Geyer et al. (2005)"}],"fun_headline_variants":["Pitch torque pumps energy to hop at 8.85 leg lengths/s","Hip motor doubly used: stabilize and energize the gait","Closed-form model predicts hopping speed, height within 16%","COM offset tunes hopping speed and apex height"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The formal stability conclusion of Theorem 1 depends on an unproved conjecture, Conjecture 1: that the averaged unperturbed SLIP limit cycle is locally exponentially stable and bounded-input bounded-output stable under small continuous disturbances, and that this property persists for small $\\epsilon$ in the unaveraged dynamics; if that conjecture is false, the proof's stability guarantee for the cascade collapses, even though the controller might still work in practice.","fun_headline_variants_meta":{"raw":{"variants":["Pitch torque pumps energy to hop at 8.85 leg lengths/s","Hip motor doubly used: stabilize and energize the gait","Closed-form model predicts hopping speed, height within 16%","COM offset tunes hopping speed and apex height"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000666,"raw_usage":{"total_tokens":3076,"prompt_tokens":1019,"completion_tokens":2057,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":635,"completion_tokens_details":{"reasoning_tokens":1987}},"tokens_in":635,"tokens_out":2057,"duration_ms":15733,"temperature":1.0,"reasoning_tokens":1987,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T00:44:06.700457+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate or instrument the full 3-DoF closed-loop system and measure the local return-map Jacobian at the predicted fixed point $q^*$ (30) for several small $\\epsilon$ values; if any eigenvalue leaves the unit circle, or the numerically exact fixed point is not $O(\\epsilon)$-close to the predicted one, Theorem 1's conclusion fails. To test Conjecture 1 directly, perturb the unperturbed SLIP limit cycle with small continuous disturbances and check that the linearized response decays; a marginal or growing linearized cycle would falsify the formal stability claim. Empirically, force a symmetric stepping law (touchdown angle exactly the negative of liftoff angle) on the hardware hopper, since Theorem 2 predicts sustained period-one hip-energized hopping then becomes impossible.","supporting_citations":[{"cited_title":"and Schwind, W","cited_arxiv_id":null,"evidence_quote":"Introduces the SLIP template used as the translational model for the unperturbed 2-DoF subsystem."}],"review_version":1}