{"id":"a0a8124b-ddd7-428d-8d59-d4106bdfb602","arxiv_id":"2507.12751","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"For the Unitree A1 bounding gait, simulation shows duty factor near 0.22 to 0.30 and longer stride durations minimize cost of transport, with hardware tests at 0.5 m/s only partially matching.","lead":"This paper sweeps duty factor, phase shift, and stride duration for a simulated Unitree A1 quadruped doing a bounding gait, measuring cost of transport in Gazebo and on a treadmill. It reports which gait settings save the most energy and shows that the best settings depend on speed.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Hardware data in Fig. 5(B) contradict the central claim that phase shift has subtle effects: COT varies ~4x and decreases up to 0.64, while simulation predicts a minimum near 0.50.","rationale":"The most load-bearing concern is the direct contradiction between the paper's central claim about phase shift and its own hardware data. The reader's weakest assumption about the work-based COT proxy is a plausible root cause, but it is not the most direct problem: even without knowing why the hardware differs, the reported hardware results already falsify the statement that phase shift has 'more subtle effects.' The paper should either revise the central claim, add repeated trials with statistical error bars to confirm the hardware trend, or explicitly explain the discrepancy. This is a serious issue that should be addressed before the optimal parameters are treated as validated for the physical robot. However, the simulation study, code release, and the consistent duty-factor trend provide partial support, so the conditional verdict remains appropriate. The reason for the condition should be sharpened from 'proxy metric may be invalid' to 'the validation data directly contradict part of the central claim.' This does not change the reader's overall CONDITIONAL verdict, hence UNCHANGED.","tokens_in":10163,"tokens_out":5646,"duration_ms":62298,"concrete_test":"Run a dedicated hardware experiment on the A1 at 0.5 m/s testing phase shifts phi = 0.44, 0.50, 0.56, 0.60, 0.64, 0.68, with at least 5 repeated trials per setting, and compute COT from Eq. (4) for each trial. If the mean COT at phi=0.64 is significantly lower than at phi=0.50 (paired t-test) and the COT does not rise at larger phi, the simulation's minimum near 0.50 and the claim of subtle phase-shift effects are contradicted.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's strongest claim (Sec. III) asserts that 'variations in phase shift have more subtle effects on energy efficiency' and that the hardware tests show 'similar trends' to simulation. The hardware data in Fig. 5(B) do not support this. In the simulation (Fig. 3(D)), COT falls from about 2.7 at phi=0.32 to about 1.6 near phi=0.44-0.52, then slightly rises. On hardware, COT continues to decrease up to the largest tested phi=0.64, with 'much larger variations (about four times)' in COT. This is not a subtle effect; it is a dominant one, and the optimal phase shift is not 0.50. Likewise, Fig. 5(C) shows no clear optimal stride duration within the tested range, unlike the simulation's clear minimum near 0.22-0.26 s. The paper's conclusion that longer stride durations and higher duty factors lower COT, while phase shift has minor influence, is therefore not a faithful summary of the presented evidence. The omitted motor dynamics and friction mentioned in Sec. III.C are a plausible cause, but regardless of the cause, the validation data contradict part of the central claim. This is not just a proxy issue: it is an internal inconsistency between the reported results and the paper's stated findings.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper investigates how three bounding gait parameters (duty factor, phase shift, and stride duration) affect the cost of transport of a Unitree A1 quadrupedal robot. The authors build on their prior SLIP-based controller and trajectory-optimized default gait, implement it in Gazebo, and sweep each parameter at 0.5, 1.5, and 2.5 m/s, reporting COT computed from absolute mechanical joint power. They then perform hardware validation on a treadmill at 0.5 m/s. The paper claims that higher duty factors (in the two-flight-phase bounding regime) and longer stride durations lower COT, that phase-shift variations have subtler effects, that the optimal parameter combination depends on speed, and that the hardware results show similar trends to simulation.","tokens_in":10495,"tokens_out":6353,"duration_ms":73753,"significance":"If the claims held, the paper would provide a useful, systematic mapping of gait-parameter sensitivity for bounding on a commercially available quadruped, with the additional value of a direct hardware comparison. The simulation sweep is systematic and internally consistent, and the public code/video links are a reproducibility strength. The hardware duty-factor trend (optimal near 0.30) is consistent with the simulation's low-cost range of 0.22-0.30, which is a genuine point in favor of the approach. However, the paper's central validation claim is weakened by the phase-shift and stride-duration discrepancies between simulation and hardware, and by the absence of error bars and statistical analysis in the hardware section.","major_comments":[{"comment":"The hardware phase-shift data contradict the statement in the Fig. 3 caption and Section III.A that phase-shift variations have 'more subtle effects' and that hardware shows 'similar trends' to simulation. In simulation at 0.5 m/s (Fig. 3(D)) COT decreases to a minimum near phi=0.44-0.52 and then rises slightly, whereas on hardware (Fig. 5(B)) COT decreases through the largest tested phi=0.64 and varies by about a factor of four. The paper itself reports the hardware optimum at phi=0.64 versus 0.50 in simulation. Because the interpretation of phase shift is load-bearing for the central claim, this discrepancy must be addressed, either by adding motor losses/friction to the simulation, by reporting electrical-power measurements, or by explicitly reframing the phase-shift conclusion as simulation-only.","section":"§III.C, Fig. 5(B)"},{"comment":"The hardware stride-duration data do not show the simulation's optimal region. Simulation at 0.5 m/s finds a COT minimum near T=0.22-0.26 s and reports instability beyond 0.34 s, while Fig. 5(C) shows the hardware's most efficient tested point at T=0.36 s with no clear optimum inside the range. The text in Section III.C notes that 0.36 s was 'less stable,' but the conclusion still states that longer stride durations lower COT at all speeds. This overgeneralizes a hardware trend measured at one speed and apparently confounded by stability, so the claim must be restricted and the discrepancy explained.","section":"§III.C, Fig. 5(C)"},{"comment":"The hardware validation reports no number of repeated trials, no error bars, and no statistical tests for the COT data in Fig. 5. Given the variability visible in the simulation box plots and the small differences between some parameter settings, the reader cannot judge whether the hardware optima (e.g., phi=0.64 versus 0.50) are significant. At minimum, the authors should report per-condition trial counts, means and standard deviations, and ideally a paired comparison of the claimed optimal settings against neighboring settings.","section":"§III.C"},{"comment":"The experimental validation is performed only at 0.5 m/s, while the paper's conclusions state that the optimal combination varies with the robot's average speed and that longer stride durations lower COT at all speeds. The three-speed dependence is therefore supported only by simulation; the hardware data cannot validate the speed-dependence claim. The authors should either add hardware tests at 1.5 and 2.5 m/s or explicitly limit the experimental conclusions to the low-speed case.","section":"§III.A, §III.C"},{"comment":"The work-based COT in Eq. (4) uses absolute mechanical joint power, while Section III.C concedes that the simulation omits motor dynamics, mechanical damping, and friction. The hardware mismatches appear in exactly the parameters (phase shift and stride duration) where these losses could re-order the ranking. A concrete fix is to compare Eq. (4) with electrical energy measured from the motor drivers on a subset of the swept settings; if the ranking changes, the optimal-parameter claims must be revised.","section":"§II.D, §III.C"}],"minor_comments":[{"comment":"The word 'demonstrate' in the abstract is too strong given the partial validation; 'suggest' or 'indicate' would be more accurate.","section":"Abstract"},{"comment":"The fitted coefficients a1-a4 in Eq. (2) are not reported numerically and no fitting data or procedure is given; this limits reproducibility of the controller. Adding the values or a reference to the fitting details would help.","section":"§II.C, Eq. (2)"},{"comment":"The QP weights W1, W2, W3 and scalars alpha and beta are not given numerically, and the friction coefficient mu is stated only as 0.6. These parameters should be listed in the paper rather than only in the code release.","section":"§II.C, Eq. (3)"},{"comment":"The text says data sets were omitted when the robot failed to reach a steady state, but it does not report how many configurations were omitted or at which parameter values; this information is needed to assess the completeness of the parameter sweep.","section":"§III.C"},{"comment":"Fig. 4 shows absolute power for three single strides, but the text does not say whether these are representative individual strides or averages; clarifying this would improve the interpretation.","section":"§III.B, Fig. 4"},{"comment":"Reference [28] is listed as 'under review' and should be updated to its published status if available, and the access dates for the Unitree and ROS web references should be consistent.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the paper contains a solid and reproducible simulation study, but the hardware validation is weaker than the text claims, and the mismatches are in the paper's main parameters (phase shift and stride duration). I recommend major revision rather than rejection because the discrepancies are explicitly reported and can be addressed by reframing the claims, adding statistical detail, and/or adding a loss model. The authors should also clarify the novelty boundary relative to their prior work [26], [27], since the controller and default gait parameters are taken from those papers."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis paper is a straightforward empirical sweep of three gait parameters—duty factor, phase shift, and stride duration—for bounding on the Unitree A1. The simulations cover three speeds, the hardware test covers one speed, and the COT is computed from mechanical joint work. The main new content is the systematic parameter sweep and the sim-to-hardware comparison for a commercially relevant platform. They ship code and video, which is good practice.\n\nThe simulation results are clear and the paper is candid about dropping unstable configurations. The trend that higher duty factors and longer stride durations lower COT is consistent across speeds and is a useful engineering guideline. The hardware duty-factor data match the simulation well, with an optimum around 0.30.\n\nThe soft spot is the phase shift claim. The paper's stated conclusion says phase shift has 'subtle' effects on efficiency, but the hardware data in Fig. 5(B) show the opposite: COT varies by about a factor of four and continues decreasing up to the largest phase shift tested, 0.64, whereas the simulation minimum was near 0.50. Describing those as 'similar trends' is not supported by the figure. The stride duration comparison also shows the hardware optimum shifted beyond the simulated range. The acknowledgement of missing motor dynamics and friction in Section III.C is a plausible explanation, but the summary should have flagged the discrepancy directly instead of presenting the validation as confirming the simulation.\n\nMinor issues: the hardware section has no error bars, trial counts, or statistical tests. The work-based COT in Eq. (4) ignores electrical and mechanical losses, which the authors acknowledge. The citation pattern is fine; using the authors' prior controller work is appropriate, though the paper should separate simulation-derived recommendations from hardware-validated ones.\n\nBottom line: this is an honest, useful engineering study with real data, but the phase-shift conclusion needs to be revised and the validation section needs more rigor. It deserves a serious referee and a major revision, not a desk reject.","headline":"Useful empirical sweep of bounding gait parameters, but the paper's own hardware data contradict its phase-shift conclusion and the validation needs more rigor.","tokens_in":10997,"tokens_out":2564,"would_cite":true,"duration_ms":28107,"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":"The paper argues that in quadrupedal bounding, duty factor and stride duration — not speed alone — set the energy cost, with optimal values that shift with speed.","keywords":["quadrupedal bounding","cost of transport","duty factor","phase shift","stride duration","energy-efficient locomotion","gait optimization","A1 quadruped robot"],"falsifier":"Measure battery current on the A1 at 0.5 m/s while holding each gait parameter at the values in the paper's sweep for about 30 steady-state strides; if energy per meter is lower for duty factors below 0.22 or for shorter strides than for the recommended settings, the simulation-based ranking is wrong.","tokens_in":9980,"feed_emoji":"🦿","tokens_out":8977,"duration_ms":95795,"temperature":0.7,"pith_summary":"The paper hypothesizes that duty factor, phase shift, and stride duration are the main gait-level levers for energy cost in quadrupedal bounding, and tests it on a simulated and physical quadruped. It finds that raising duty factor into the 0.22–0.30 range and lengthening stride duration lowers cost of transport at every speed tested, while phase shift has a smaller effect. If the finding holds on hardware, a bounding robot can extend battery life by adapting these timing parameters to its speed instead of running a fixed gait. The result suggests speed-dependent gait scheduling rather than one universal optimal setting.","feed_headline":"Longer strides and higher duty factors cut quadruped energy cost","feed_subtitle":"Bounding tests on the A1 show best efficiency depends on speed, so gaits should adapt.","key_machinery":"The load-bearing metric is the work-based cost of transport in Eq. (4): the sum of absolute mechanical power at all twelve joints over a stride, divided by weight times forward distance. The control machinery is a floating-base model of the A1 with point-contact feet, a gait generator that reduces bounding to three free parameters — duty factor $\\gamma$, phase shift $\\varphi$, and stride duration $T$ — and a controller in which each stance leg pair is treated as a spring-loaded inverted pendulum for hip height, ground reaction forces are solved by a quadratic program under a single-rigid-body approximation and friction-cone constraints, and swing legs follow a placement strategy. The controller deliberately lets the torso pitch rather than forcing a fixed posture, which is what makes long, efficient strides feasible.","core_discovery":"The discovery is that the energy economy of a two-flight-phase bounding gait is largely set by two timing parameters: duty factor and stride duration. Across simulated speeds of 0.5, 1.5, and 2.5 m/s, cost of transport falls as duty factor rises into the 0.22–0.30 range and as stride duration lengthens toward the stability boundary, while mid-range phase shifts (0.44–0.52) keep cost low and consistent. Short strides are costly because they multiply touch-down events and power fluctuations; overly long strides destabilize the torso and cause slipping. Treadmill tests on the A1 at 0.5 m/s confirm the duty-factor trend and the benefit of longer strides, but put the best phase shift near 0.64 rather than 0.50, so the paper concludes the optimal parameters depend on speed and should be scheduled adaptively.","pith_inferences":["The authors leave implicit that the same sweep could be run online: a controller that nudges duty factor and stride duration while measuring cost of transport could find the speed-dependent optimum without a precomputed table.","Because the controller already treats legs as springs and lets the torso rotate freely, the optimal stride duration may track a mechanical resonance; plotting optima against Froude number could collapse all three speeds onto one curve.","The simulation-to-hardware gap in phase shift and stride duration suggests that adding motor electrical losses and joint damping to the model would shift predicted optima toward the hardware values."],"forward_implications":["A bounding controller can lower energy use by holding duty factor near 0.22–0.30 and using the longest stride duration that keeps the torso stable, with no hardware changes.","Optimal gait parameters depend on forward speed, so an adaptive scheduler that retunes $\\gamma$, $\\varphi$, and $T$ as speed changes should beat any single fixed gait.","Since mid-range phase shifts give low and stable COT in simulation, phase shift can be set coarsely without a major energy penalty, simplifying control.","On physical hardware, best efficiency sits at higher phase shift and longer stride than simulation predicts, so deployment should include hardware-in-the-loop tuning even if simulation guides the search."],"supporting_citations":[{"why":"Defines the work-based COT in Eq. (4) and supplies the default duty factor, phase shift, and stride duration used as the simulation baseline.","marker":"[26]"},{"why":"Provides the bounding controller that treats each stance leg pair as a spring-loaded inverted pendulum and lets the torso pitch freely.","marker":"[27]"},{"why":"Gives the passive SLIP model whose hip-height dynamics the controller approximates in Eq. (2).","marker":"[28]"},{"why":"Supplies the swing-leg foot-placement strategy used to generate swing-phase joint trajectories.","marker":"[17]"},{"why":"Documents the quadratic-programming ground-reaction-force controller on the A1 that this work builds on.","marker":"[29]"},{"why":"Provides the physics simulation environment in which all gait-parameter sweeps were run.","marker":"[31]"}],"fun_headline_variants":["Duty factor and stride length dominate quadruped energy cost","Optimal bounding gait depends on speed, tests on A1 show","Longer strides, higher duty factor cut robot bounding energy","Adaptive gait scheduling beats fixed parameters for efficiency","Why bounding robots should tune duty factor and stride length"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The results assume the mechanical joint work measured in simulation ranks gaits the same way the robot's real battery consumption does, even though motor losses, damping, and friction are left out.","fun_headline_variants_meta":{"raw":{"variants":["Duty factor and stride length dominate quadruped energy cost","Optimal bounding gait depends on speed, tests on A1 show","Longer strides, higher duty factor cut robot bounding energy","Adaptive gait scheduling beats fixed parameters for efficiency","Why bounding robots should tune duty factor and stride length"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000457,"raw_usage":{"total_tokens":2272,"prompt_tokens":901,"completion_tokens":1371,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":517,"completion_tokens_details":{"reasoning_tokens":1291}},"tokens_in":517,"tokens_out":1371,"duration_ms":11651,"temperature":1.0,"reasoning_tokens":1291,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T16:39:01.566790+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure battery current on the A1 at 0.5 m/s while holding each gait parameter at the values in the paper's sweep for about 30 steady-state strides; if energy per meter is lower for duty factors below 0.22 or for shorter strides than for the recommended settings, the simulation-based ranking is wrong.","supporting_citations":[{"cited_title":"Energy-optimal asymmetrical gait selection for quadrupedal robots,","cited_arxiv_id":null,"evidence_quote":"Defines the work-based COT in Eq. (4) and supplies the default duty factor, phase shift, and stride duration used as the simulation baseline."},{"cited_title":"All common bipedal gaits emerge from a single passive model,","cited_arxiv_id":null,"evidence_quote":"Gives the passive SLIP model whose hip-height dynamics the controller approximates in Eq. (2)."},{"cited_title":"Legged robots that balance,","cited_arxiv_id":null,"evidence_quote":"Supplies the swing-leg foot-placement strategy used to generate swing-phase joint trajectories."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Documents the quadratic-programming ground-reaction-force controller on the A1 that this work builds on."},{"cited_title":"Design and use paradigms for gazebo, an open-source multi-robot simulator,","cited_arxiv_id":null,"evidence_quote":"Provides the physics simulation environment in which all gait-parameter sweeps were run."}],"review_version":1}