{"id":"2ee32bd0-af48-4c87-b4bb-ef22f7073a6a","arxiv_id":"2411.13289","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Passive knee flexion on a bipedal robot delays ankle push-off and increases the forward impulse of the trailing leg during the step-to-step transition.","lead":"A bipedal robot with spring-loaded ankles walked with two knee control modes: active knee motion and a freely swinging knee. The passive mode delayed ankle push-off and produced a larger forward push of the trailing leg during the step-to-step transition.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The direct robot result is solid, but the causal claim that delayed ankle plantarflexion drives the 87% impulse increase is underdetermined: AKFI and PKFI differ in four timing variables and knee torque simultaneously. A SAPF-isolated control is needed.","rationale":"The reader's weakest assumption pinpoints the same confound: AKFI and PKFI differ in multiple timing variables simultaneously, so the observed momentum difference cannot be uniquely attributed to SAPF timing. I do not see a fatal flaw in the direct empirical comparison. The robot experiment is reproducible, the effects are large relative to the reported standard deviations, and the authors honestly state that SAPF was indirectly manipulated and that transfer to human gait remains to be explored. The paired Wilcoxon test is questionable because gait cycles within a condition are autocorrelated and pairing across separate experiments is arbitrary, but the effect sizes are large enough that this would not overturn the direct comparison. The load-bearing issue is causal attribution. The title claim itself ('passive knee flexion increases forward impulse') is directly supported by the data. The abstract's 'leading to' claim, namely that the SAPF delay produces the impulse increase, is an additional inference that is not uniquely identified. The supplementary PKFI40 condition does not isolate SAPF because it changes the duration of the passive knee behavior, not SAPF alone. A test that independently shifts SAPF while holding other timing and control variables equal would settle the question. Since the reader's conditional verdict already reflects exactly this uncertainty, no verdict change is needed.","tokens_in":20694,"tokens_out":4423,"duration_ms":49673,"concrete_test":"Use the existing PKFI40 data as a first check: if PKFI40's SAPF delay and trailing-leg impulse lie on the same relation as AKFI versus PKFI, the timing mechanism gains support; if they do not, the effect is confounded by the other simultaneous changes. The decisive check is a new experiment or simulation in which SAPF onset is shifted by roughly 2-3%GC (for example, by a timed ankle torque or a spring-slack perturbation) while knee and hip command trajectories, knee impedance, and LLTD are held identical across conditions. The causal claim is confirmed only if the trailing-leg impulse increase tracks the SAPF shift when the other gait-event timings are fixed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The empirical contrast is real: the robot's trailing-leg horizontal impulse during the step-to-step transition is larger under PKFI than under AKFI, and the data and code are available. The load-bearing step is the paper's causal reading that this increase is produced by the 3%GC delay of SAPF, expressed in the abstract as 'leading to 87% larger increase'. That inference is underdetermined by the experiment as reported. The two conditions differ simultaneously in SKF onset (5%GC later), SHF onset (4%GC later), SAPF onset (3%GC later), LLTD timing (1%GC difference), and in the fact that the knee motor torque is zero for a large fraction of the gait cycle in PKFI. Any of these differences, or the change in knee joint impedance itself, could contribute to the impulse difference. The authors explicitly state that they 'indirectly manipulated' SAPF timing, so SAPF timing was not independently varied. Consequently, the abstract's causal chain—passive knee flexion, then delayed ankle plantarflexion, then larger trailing-leg impulse—is not uniquely supported. The direct empirical claim of the title survives, but the mechanistic claim, which is the paper's main contribution to the catapult-release discussion, does not yet have causal support.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper compares two knee-control modes on the bipedal robot EcoWalker-2: active knee flexion initiation (AKFI), where the knee motor follows a commanded trajectory throughout the gait cycle, and passive knee flexion initiation (PKFI), where the knee motor is commanded to zero torque from 35% of the gait cycle onward. The authors report that PKFI delays the start of knee flexion by 5% of the gait cycle (GC), the start of hip flexion by 4% GC, and the start of ankle plantar flexion (SAPF) by 3% GC, while also shifting leading-leg touch-down by 1% GC. During the step-to-step transition, PKFI increases the horizontal impulse of the trailing leg by 87% (0.14 kg·m/s), increases the magnitude of the center-of-mass momentum vector by 188%, and causes a larger decrease in remaining-body horizontal momentum, compared with AKFI. The authors interpret these findings as evidence that passive knee flexion can trigger the release of the ankle catapult and that the delayed SAPF timing is the causal driver of the impulse differences. Data and code are provided via a public repository.","tokens_in":20944,"tokens_out":4666,"duration_ms":47861,"significance":"If the reported effects are robust, the paper provides a valuable robotic demonstration of Perry's hypothesis that passive knee flexion can initiate the push-off sequence, and it quantifies momentum changes associated with altered gait-event timing. The strengths of the work include its use of a physical biped with passive spring-loaded ankles, the precision of the encoder-based measurements (reported SD < 0.4% GC), the inclusion of a PKFI40 robustness check, and the public release of data, code, and CAD files. However, the direct empirical contrast between AKFI and PKFI is strong, but the mechanistic interpretation that the SAPF delay specifically 'leads to' the larger impulse is underdetermined because multiple kinematic and actuation variables change simultaneously between conditions. The statistical analysis also uses a paired test on non-paired data. With corrected statistics and appropriately hedged causal claims, the study would be a solid contribution to the discussion of push-off function and knee-ankle coordination.","major_comments":[{"comment":"The Wilcoxon signed-rank test is applied to 120 gait cycles from each of two separate experimental runs. These cycles are not paired: there is no one-to-one correspondence between cycle i in the AKFI run and cycle i in the PKFI run, and the two runs were performed sequentially on the same robot rather than as a matched repeated-measures design. Using a paired test on unpaired samples can inflate significance. Please re-analyze with an unpaired two-sample test (e.g., Mann-Whitney U) or a mixed-effects model that treats gait cycle as a repeated measure within each run. Because the reported p-values are extremely small, the substantive conclusions will likely survive, but the inference procedure as written is incorrect.","section":"3.4"},{"comment":"The abstract states that passive knee flexion 'resulted in a 3% of the gait cycle later onset of ankle plantar flexion, leading to 87% larger increase in the trailing leg horizontal momentum.' However, PKFI and AKFI differ simultaneously in SKF onset (5% GC later), SHF onset (4% GC later), SAPF onset (3% GC later), LLTD timing (1% GC difference), and in the knee motor being torque-free from 35% GC onward (Fig. 3, Methods 3.3). The authors themselves describe SAPF as 'indirectly manipulated' (Introduction, Fig. 1b), so the experiment does not independently vary SAPF. The observed impulse difference could plausibly be caused by any of these coordinated changes, or by the change in knee joint impedance itself. An additional condition that varies SAPF timing without changing knee torque and hip-flexion timing, or a mediation analysis that controls for the other timing variables, is needed before the causal chain can be supported. At minimum, the abstract and conclusions should be reworded to describe the observed coordinated timing shift rather than asserting that the SAPF delay alone produces the impulse increase.","section":"Abstract and Discussion"}],"minor_comments":[{"comment":"In the SAPF definition, the text says 'We defined the moment of SAPF when the trailing leg's ankle plantar flexion angle reached its maximum.' Since SAPF is the start of plantarflexion after a period of dorsiflexion, the relevant event should be the maximum dorsiflexion angle. Please clarify the sign convention and event definition, as written the description is ambiguous.","section":"3.4"},{"comment":"The caption contains duplicated text: 'the the absolute' appears twice. Please correct.","section":"Table 1 caption"},{"comment":"The phrase 'A possible release mechanism can be derives from Perry's work' contains a grammatical error; it should read 'can be derived from Perry's work.'","section":"Introduction"},{"comment":"The sentence 'leading leg touch-down did not happen exactly at 50 %GC in either PKFI or AKFI experiment by neither legs' uses a double negative. Please reword to 'in either leg' or 'in neither leg.'","section":"Discussion"},{"comment":"The term 'SOLO actuator module' is used without explanation on first appearance. Please define it as a compact series-elastic actuator module or reference the appropriate prior work.","section":"Methods 3.1"}],"recommendation":"major_revision","confidential_remarks":"This manuscript reports a well-instrumented robot experiment with open data and code, and the direct AKFI-versus-PKFI contrast is clear. The main barrier is not the empirical measurement but the causal interpretation: the abstract's 'leading to' claim attributes the impulse increase specifically to the SAPF delay, yet the experimental manipulation changes several timing and actuation variables simultaneously. Combined with the invalid paired-test choice, this requires a revision rather than acceptance in its current form. I do not see a need to reject, because the direct empirical result and the open-data practices are valuable, and the causal claim can be fixed either by an additional control experiment or by substantially hedging the language."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth knowing: this is the first robot test of passive vs active knee flexion initiation, and the direct empirical contrast is solid—PKFI gives an 87% larger increase in trailing-leg horizontal impulse and a 188% larger increase in CoM momentum magnitude during the step-to-step transition, with public data and code. The EcoWalker-2 walking with zero knee torque is itself a nice mechanical demonstration.\n\nWhat's good: the experiment is clean, the gait event timing precision is genuinely better than human data, and the PKFI40 robustness check shows the result is not sensitive to the exact zero-torque start time. The authors are upfront that transfer to human gait remains open, and they do not oversell the leg-swing vs redirection debate resolution.\n\nThe soft spots are two. First, the statistics: they use a paired Wilcoxon test on 120 cycles from two separate runs. Those are not paired, so the p-values as reported are not valid. With effect sizes this large (differences an order of magnitude above the SD), an unpaired test would almost certainly still show p < 0.001, so I doubt the conclusion flips, but it needs to be redone properly.\n\nSecond, the causal reading. The abstract says the 3% later SAPF 'leads to' the 87% impulse increase. But the two conditions differ in knee flexion onset, hip flexion onset, SAPF onset, LLTD timing, and in whether the knee motor is active at all. SAPF was 'indirectly manipulated,' not independently varied. So the data show a correlation between the condition and the impulse change, but they don't uniquely identify SAPF timing as the causal driver. The title's direct claim survives; the mechanistic claim about the catapult release needs a control that isolates SAPF timing, or the language needs to be softened to 'associated with' rather than 'leading to.'\n\nMinor point: the 188% CoM momentum increase sounds dramatic but is 0.08 kg m/s on a 2.1 kg robot; still real, just context.\n\nWho should read it: legged robotics people, biomechanists working on push-off timing, and prosthetic/orthotic control. It deserves a serious referee. I'd send it to peer review with instructions to fix the stats and either add an isolating experiment or rewrite the causal claims as hypotheses.","headline":"Direct robot result is clean and worth knowing; the stats are mispaired and the causal chain from SAPF timing to impulse is underdetermined.","tokens_in":21469,"tokens_out":3403,"would_cite":true,"duration_ms":34747,"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":"Passive knee flexion, by delaying the start of ankle push-off until after the leading leg lands, increases the trailing leg's horizontal impulse by 87% and shifts ankle energy toward swinging the leg rather than accelerating the body.","keywords":["passive knee flexion","ankle push-off timing","step-to-step transition","bipedal robot","elastic ankle catapult","center of mass momentum","gait event timing","cost of transport"],"falsifier":"Run the robot in a third condition that keeps knee and hip flexion onsets identical to AKFI but commands the ankle plantar-flexion onset to occur 3% of the gait cycle later, matching PKFI's SAPF timing; if the trailing-leg horizontal impulse does not rise by roughly 87% relative to AKFI, the paper's claim that the SAPF delay drives the momentum change would be falsified.","tokens_in":20505,"feed_emoji":"🦵","tokens_out":8862,"duration_ms":86683,"temperature":0.7,"pith_summary":"This paper asks whether passively letting the knee flex, rather than actively driving it, can trigger the release of the leg's 'catapult'—the elastic ankle that stores energy in stance and releases it at push-off. Using the EcoWalker-2 bipedal robot, the authors compare active knee flexion initiation (AKFI) with passive knee flexion initiation (PKFI), where the knee motor is unpowered for most of stance. They find that passive initiation delays the start of ankle plantar flexion by about 3% of the gait cycle, from just before to just after the leading leg touches down, and this delay is associated with an 87% larger horizontal impulse of the trailing leg during the step-to-step transition. The authors argue that the timing of ankle release relative to touch-down determines whether ankle energy accelerates the remaining body or swings the trailing leg forward, which matters for understanding human push-off and for designing orthoses, prostheses, and efficient robots.","feed_headline":"Passive knee flexion boosts trailing-leg forward impulse by 87%","feed_subtitle":"Robotic walker shows delaying ankle push-off by 3% of the stride shifts energy into swing-leg acceleration","key_machinery":"The central object is the EcoWalker-2, a planar biped with hip and knee motors, passive spring-loaded ankles, and biarticular GAS and SOL spring-tendons that span the ankle and also couple to the knee. Its two control modes define the comparison: in AKFI the knee motor tracks a programmed rhythmic angle trajectory for the whole gait cycle; in PKFI the knee motor outputs zero torque from 35% of the gait cycle onward, so knee flexion arises only from the GAS spring and ground reaction forces. The mechanism that carries the argument is the timing gap between the start of ankle plantar flexion (SAPF) and leading-leg touch-down (LLTD): AKFI releases the ankle 2% of the gait cycle before LLTD, PKFI releases it 2% after LLTD, and the paper attributes the momentum differences to that shift. All momentum changes are evaluated over the step-to-step transition, defined as the interval from the minimum to the maximum vertical center-of-mass velocity.","core_discovery":"On the paper's own terms, the central discovery is that passive knee flexion initiation works as a gait-event timing mechanism: with the knee motor switched off during mid-to-late stance, the knee begins to flex 5% of the gait cycle later than with active control, and the ankle begins to plantarflex 3% later—after rather than before leading-leg touch-down. That shift moves the elastic recoil of the passive ankle springs from a pre-touch-down 'preemptive push-off' that mainly accelerates the rest of the body into a post-touch-down release that mainly propels the trailing leg into swing. Measured over the step-to-step transition, the trailing leg's horizontal momentum impulse is 0.14 kg m/s larger (87% larger), the center-of-mass momentum magnitude increase is 188% larger, and the remaining body's horizontal impulse decreases more in the passive condition. Both conditions walk at the same speed with sufficient toe clearance, and passive knee flexion slightly lowers the net positive cost of transport. The paper presents this as the first robotic test of the hypothesis that passive knee flexion is the 'catch' that releases the human lower-leg catapult.","pith_inferences":["Editorial inference: The causal attribution is narrower than the experimental manipulation—AKFI and PKFI differ in knee flexion onset, hip flexion onset, and touch-down timing as well as ankle push-off timing, so a follow-up varying only the commanded ankle release time would test whether the 87% impulse change is really driven by SAPF timing.","Editorial inference: Applied to prosthetics, a zero-torque or low-impedance knee window timed near push-off could let users tune how much ankle energy returns to the limb versus the torso, which may matter for users with limited hip control.","Editorial inference: Since human gait events vary by about 1.5% of the gait cycle across strides, the 3% timing shift in this robot suggests that natural stride-to-stride variability could produce appreciable differences in how push-off energy is distributed, even at the same average speed.","Editorial inference: The sagittal-plane constraint and upper-body stabilization may magnify the measured effects; a free 3D walker would likely show some of the push-off impulse absorbed by lateral momentum, so the 87% figure should be read as an upper-bound estimate for unconstrained walking."],"forward_implications":["If the timing claim holds, shifting ankle push-off a few percent of the gait cycle after leading-leg touch-down is enough to redirect a large share of ankle energy toward swinging the trailing leg rather than accelerating the body.","A robot can maintain the same walking speed and toe clearance with no knee torque during most of stance, so passive knee dynamics can replace active knee control in that phase without losing gait functionality.","The small cost-of-transport reduction (0.52 vs 0.56 in net positive cost of transport) suggests that letting the knee flex passively may also be energetically beneficial, not just kinematically feasible.","For assistive devices, the results imply that knee-flexion timing—not just ankle power—should be a control parameter in orthoses and prostheses, since it changes where push-off energy goes.","The event-ordering difference (SAPF before vs after LLTD) provides a robotic confirmation of the human-like preemptive push-off pattern and gives a concrete mechanical role to the knee in the leg catapult release."],"supporting_citations":[{"why":"Supplies the observation under test: the main part of knee flexion after leading-leg touch-down is passive, motivating the passive knee flexion condition.","marker":"[5]"},{"why":"Provides human trailing-leg and remaining-body momentum data during push-off and the impulsive ankle push-off view used to interpret the robot's momentum changes.","marker":"[4]"},{"why":"Documents that ankle power during push-off largely reflects elastic recoil of the Achilles tendon, the mechanical basis of the catapult.","marker":"[2]"},{"why":"Defines the step-to-step transition interval (minimum to maximum vertical CoM velocity) used as the measurement window for all momentum impulses.","marker":"[17]"},{"why":"Frames the swing-leg versus CoM-redirection debate over ankle push-off, which the paper's timing argument is meant to inform.","marker":"[16]"},{"why":"Introduces the EcoWalker robot and the GAS/SOL spring-tendon architecture that EcoWalker-2 inherits and modifies.","marker":"[12]"},{"why":"Supplies the human gait dataset whose event timing is compared with the robot's to show that active knee flexion resembles human ordering.","marker":"[42]"}],"fun_headline_variants":["Robotic walker: passive knee flexion boosts swing impulse 87%","Passive knee flexion delays push-off, boosts forward momentum 87%","Knee flexion timing: passive beats active for forward impulse","How passive knees improve walking impulse: robot study","87% more forward impulse with passive knee flexion in robot"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the later start of ankle plantar flexion—not the simultaneous differences in knee flexion onset, hip flexion onset, and leading-leg touch-down timing—is what produces the larger trailing-leg impulse.","fun_headline_variants_meta":{"raw":{"variants":["Robotic walker: passive knee flexion boosts swing impulse 87%","Passive knee flexion delays push-off, boosts forward momentum 87%","Knee flexion timing: passive beats active for forward impulse","How passive knees improve walking impulse: robot study","87% more forward impulse with passive knee flexion in robot"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000511,"raw_usage":{"total_tokens":2512,"prompt_tokens":997,"completion_tokens":1515,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":613,"completion_tokens_details":{"reasoning_tokens":1430}},"tokens_in":613,"tokens_out":1515,"duration_ms":13746,"temperature":1.0,"reasoning_tokens":1430,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T16:36:15.805837+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the robot in a third condition that keeps knee and hip flexion onsets identical to AKFI but commands the ankle plantar-flexion onset to occur 3% of the gait cycle later, matching PKFI's SAPF timing; if the trailing-leg horizontal impulse does not rise by roughly 87% relative to AKFI, the paper's claim that the SAPF delay drives the momentum change would be falsified.","supporting_citations":[{"cited_title":"Gait Analysis: Normal and Pathological Function (SLACK Incorporated, 1992)","cited_arxiv_id":null,"evidence_quote":"Supplies the observation under test: the main part of knee flexion after leading-leg touch-down is passive, motivating the passive knee flexion condition."},{"cited_title":"& Seyfarth, A","cited_arxiv_id":null,"evidence_quote":"Provides human trailing-leg and remaining-body momentum data during push-off and the impulsive ankle push-off view used to interpret the robot's momentum changes."},{"cited_title":"J., Prilutsky, B","cited_arxiv_id":null,"evidence_quote":"Documents that ankle power during push-off largely reflects elastic recoil of the Achilles tendon, the mechanical basis of the catapult."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the step-to-step transition interval (minimum to maximum vertical CoM velocity) used as the measurement window for all momentum impulses."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Frames the swing-leg versus CoM-redirection debate over ankle push-off, which the paper's timing argument is meant to inform."},{"cited_title":"J., Mundinger, E","cited_arxiv_id":null,"evidence_quote":"Supplies the human gait dataset whose event timing is compared with the robot's to show that active knee flexion resembles human ordering."}],"review_version":1}