{"id":"ff225e39-cf1e-4ca6-827d-2a914b2e071d","arxiv_id":"2506.12314","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A variable reduction ratio knee joint, whose transmission ratio drops during extension, improves explosive jumping in electrically driven humanoids.","lead":"This paper designs a humanoid knee joint whose gear ratio decreases as the leg extends, letting the motor deliver sustained high power during a jump. A one-legged test platform jumped 63 cm, and a 45 kg humanoid robot jumped 0.5 m high and 1.1 m forward.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The headline 28.1% advantage over fixed-ratio knees is computed from a simplified vertical CoM model whose Jacobian (Eq. 5) appears to omit a factor 1/2, and the paper provides no fixed-ratio hardware baseline.","rationale":"The paper reports a genuine hardware result: a 1-DOF platform with the EVRR-K knee jumped 0.63 m, and BHR8-J1 completed three jumps with recorded motor currents. The variable-ratio mechanism is a plausible way to keep the motor in a favorable torque-speed region. The issue is the central comparison. The 28.1% headline advantage is not an experimental value; it is the output of a simplified optimization. That optimization's Jacobian looks wrong by a factor of two, and the paper's own Section VI admits the ankle is not included. Because the full-robot jumps use active hip and ankle joints, they do not provide a direct test of the simplified model's prediction. I am not claiming the hardware is nonfunctional or the concept is invalid. But the strongest quantitative claim—sustained high power and 28.1% improvement over an optimal fixed-ratio joint—rests on a model that has a concrete algebraic suspect and no fixed-ratio control experiment. This is exactly the kind of load-bearing assumption that a conditional verdict should pin down. The reader's weakest_assumption already pointed at the simplified vertical model and the ankle; I agree in general and sharpen it to Eq. (5). There is also an internal inconsistency between 28.1% in the abstract and 20% in the conclusion, which makes the headline number hard to audit. The reader's verdict of CONDITIONAL remains appropriate; no change needed.","tokens_in":12944,"tokens_out":7147,"duration_ms":91204,"concrete_test":"Re-derive Eq. (5) from the two-link geometry and verify the missing 1/2. Then rerun the Section IV optimization with the corrected Jacobian, including the Table II initial angles, for both EVRR-K and FRR-K. If the jump-height margin does not remain near 28% (or if the optimal VRR curve changes materially), the claimed advantage over fixed-ratio joints is not established. As a weaker but independent check, run the same one-DOF platform with an equivalent-mass fixed-ratio knee under the same current command and compare the achieved height to 63 cm.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim is that EVRR-K outperforms the optimal fixed-ratio knee by 28.1% in jump height (abstract) or 20% (Section VI). This number is not measured against any fixed-ratio hardware; it comes from the optimization in Section IV, which uses the simplified single-joint vertical model of Section II-B. That model therefore carries the entire comparison. There is a concrete, checkable issue inside the model. For a symmetric two-link leg with CoM height y = [m1*a1 + m2*(l1+a2) + m3*(l1+l2)]*cos(q2/2)/M, the y-Jacobian is d y/d q2 = (1/2)*[same expression]*sin(q2/2)/M (up to sign). Equation (5) prints the same expression without the 1/2. Since lambda = 1/J in Eq. (4), all force/speed quantities in Eq. (6) and all optimized work/takeoff-energy values inherit this factor. Re-running the optimization with the correct Jacobian could shift the optimized VRR parameters and especially the FRR-K baseline, so the reported 28.1% margin may not survive. The model also forces the CoM to move vertically with hip and ankle passive; Section VI concedes the ankle is unaddressed. In the full-robot experiments, hip and ankle are active, so those experiments do not directly validate the model used to produce the claimed margin. A hardware or whole-body comparison against a comparable fixed-ratio knee is absent.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a variable-reduction-ratio knee joint (EVRR-K) for humanoid jumping, implemented by a linear-actuator-driven guide-rod mechanism whose reduction ratio decreases as the knee extends. The authors analyze motor output limits and a simplified vertical-CoM knee kinematic model, optimize the mechanism parameters (crank length, frame length, assembly offset) by maximizing simulated takeoff energy, and compare the result with an optimized fixed-ratio knee in simulation. Experiments on a 24.93 kg single-joint platform give a 63 cm vertical jump, and the 45 kg humanoid BHR8-J1 equipped with the mechanism performs a 0.5 m vertical jump, a 1.1 m forward jump, and a 0.5 m box jump. The reported 28.1% advantage over fixed-ratio knees is a simulation-based comparison rather than a hardware measurement.","tokens_in":13278,"tokens_out":7258,"duration_ms":79153,"significance":"The paper's concrete hardware achievements are significant: a compact motor-driven knee enables a 63 cm jump on a 24.93 kg single-joint platform and useful multi-task jumps on a full humanoid, with motor speeds kept below about 3000 rpm and the mechanism packaged in a practical form. The design paradigm is credible and worth reporting. However, the headline quantitative superiority over fixed-ratio knees is not measured; it is a model-based prediction, and the simplified model used for that prediction appears to contain a Jacobian error that may change the reported margin. The central claim is therefore defensible but not yet established at the strength claimed.","major_comments":[{"comment":"Equation (5) is missing a factor of 1/2. For the CoM height y_CoM = [m1 a1 + m2(l1+a2) + m3(l1+l2)] cos(q2/2)/(m1+m2+m3), differentiation gives d y_CoM/d q2 = -(1/2)[m1 a1 + m2(l1+a2) + m3(l1+l2)] sin(q2/2)/(m1+m2+m3). The magnitude, which is what Eq. (5) intends, should therefore contain the factor 1/2. Because lambda(q2)=1/J in Eq. (4) and Eq. (6) uses lambda to map joint torque to CoM force and joint rate to CoM velocity, all subsequent quantities in Section IV, including the optimized takeoff energy and the FRR-K comparison in Table II, are computed with lambda values that are a factor of two too large. The optimization should be rerun with the corrected Jacobian and the reported 28.1%/28.9% margin should be re-quantified.","section":"Section II-B, Eq. (5)"},{"comment":"The claimed improvement over fixed-ratio joints is not validated experimentally. Section V-A reports only the EVRR-K platform jump of 63 cm; no fixed-ratio knee was built or tested. The 'theoretical improvement of 28.1%' (abstract) and '20% improvement' (Section VI) both derive from the simulation in Table II, where the FRR-K baseline is optimized in the same simplified model. Since a model-based baseline is used to support the central quantitative claim, a hardware or whole-body-simulation comparison against a comparable fixed-ratio knee is needed before the improvement can be considered established.","section":"Section V-A and Table II"},{"comment":"The simplified model used for optimization constrains the CoM to vertical motion and treats the hip and ankle as passive, yet the full-robot validation in Section V-B uses active hip and ankle joints, and Section VI explicitly states that the ankle's role is unaddressed. The single-joint platform is consistent with the model, but the full-robot jumps cannot validate the optimized ratio curve or the simulated margin over FRR-K. A concrete resolution would be to evaluate the optimized EVRR-K parameters in a whole-body multi-joint model with active hip/ankle, or to compare hardware against a fixed-ratio knee on the same platform.","section":"Section II-B and Section V-B"}],"minor_comments":[{"comment":"The improvement percentages are inconsistent: 28.1% in the abstract, approximately 28.9% in Section IV-B, and 20% in Section VI. These should be reconciled and stated as simulated margins where appropriate.","section":"Abstract, Section IV-B, Section VI"},{"comment":"The second relation in Eq. (6) appears inconsistent with Eq. (4): if lambda(q2)=qdot2/ydot_CoM, then ydot_CoM = qdot2/lambda(q2), not qdot2*lambda(q2). Please correct the printed equation or the definition.","section":"Eq. (6)"},{"comment":"There are typographical errors: 'Index T erms' in the abstract, 'center of mas' in Section I, and inconsistent formatting of subscripts and Greek letters in several equations. The manuscript should be proofread.","section":"Section I and abstract"},{"comment":"The caption of Fig. 5 references 'Fig. 12a' and 'Fig. 12b' when it should reference panels (a) and (b) of Fig. 5; this can confuse readers.","section":"Fig. 5 caption"},{"comment":"In Eq. (18), H = W_takeoff/(m_tot g) - y_CoM,s; please clarify whether y_CoM,s is the CoM height in the fully extended pose, since this affects the reported jump height.","section":"Section IV-B, Eq. (18)"}],"recommendation":"major_revision","confidential_remarks":"The hardware results are real and would be of interest to RA-L readers, but the paper's quantitative claim of superiority over fixed-ratio knees needs correction. No concerns about attribution or novelty beyond the usual need to cite related variable-transmission mechanisms."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a real hardware contribution—a variable reduction ratio knee driven by a linear actuator—and the jump demos are genuine. The headline 28.1% advantage over fixed-ratio knees, though, is a simulation result, and the simulation has a concrete bug that needs checking.\n\nWhat's new: the idea of monotonically decreasing the knee reduction ratio as the leg extends to keep the motor in its high-power region is well motivated. The guide-rod mechanism is compact and gives a wide range of ratio curves. The single-joint platform achieving 63 cm on a 24.93 kg leg is a solid experimental demonstration, and the full 45 kg robot doing 0.5 m vertical, 1.1 m forward, and 0.5 m box jumps is real hardware performance. The authors also give an optimization procedure for choosing the mechanism parameters, which is a useful template.\n\nSoft spots, in order of softness. The claimed 28.1% improvement over an optimal fixed-ratio knee is not measured against any fixed-ratio hardware. It comes from the simplified vertical-CoM model of Section II-B, which is also the model used to optimize the parameters. That makes the improvement a model prediction. More concretely, Eq. (5) looks wrong: for the symmetric two-link leg, the CoM Jacobian should be (1/2)*[...]*sin(q2/2)/M, and the printed equation is missing the 1/2. Since lambda = 1/J, this doubles the transmission ratio and changes all force/speed mappings. Re-running the optimization with the correct Jacobian could shift the optimal parameters and the FRR-K baseline, so the reported margin may not survive. The conclusion also says 20% while the abstract says 28.1%, another sign the number isn't stable. And the model treats hip and ankle as passive, which Section VI concedes; the full-robot experiments use active hip and ankle, so they don't validate the model used for the improvement claim.\n\nMinor stuff: some references look off-topic ([4], [5], [7]), and there are typos like 'T erms' in the index.\n\nBottom line: the mechanism and the hardware results are worth knowing about. The quantitative comparison to fixed-ratio designs needs a corrected model and ideally a hardware baseline. I'd send it to peer review with a request to fix the Jacobian and report the sensitivity of the optimization to that error.","headline":"Real hardware, genuine jump demos, but the headline 28.1% advantage over fixed-ratio knees is a simulation result built on a model with a likely missing 1/2 factor in the Jacobian.","tokens_in":13797,"tokens_out":2572,"would_cite":true,"duration_ms":29977,"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 a knee whose reduction ratio falls as it extends keeps a fixed electric motor in its high-power band through takeoff, yielding a 63 cm single-joint jump and a 0.5 m humanoid jump.","keywords":["humanoid robots","explosive jumping","variable reduction ratio","knee joint design","linear actuator","ball screw","electric actuation","jump control"],"falsifier":"Run an A/B jump test on the same robot with the same motor, control law, and takeoff angle, swapping only the knee ratio curve: if the EVRR-K joint does not measurably out-jump the best fixed-ratio knee, or if the knee motor's speed exceeds the high-loss region before takeoff in the EVRR-K case, the central claim fails. A cheaper check: instrument the knee during the reported 0.5 m box jump and verify that motor speed stays below roughly 3000 rpm and joint power remains near 1.5 kW through the late takeoff phase.","tokens_in":12774,"feed_emoji":"🦿","tokens_out":8704,"duration_ms":91880,"temperature":0.7,"pith_summary":"Electric humanoid robots jump poorly because a fixed gear ratio between the knee motor and the body cannot deliver both high torque at the start of a jump and high speed at the end; the motor either stalls low or spins into a high-loss speed region. This paper proposes a knee whose reduction ratio starts high and falls as the joint extends, so the motor can sit in its high-power band through the whole takeoff. The idea is realized with a linear-actuator-driven guide-rod mechanism whose geometry makes the effective ratio a tunable function of knee angle, and the paper optimizes that geometry for maximum takeoff energy. On a one-leg test platform carrying a 20 kg load, the joint jumped 63 cm, which the paper reports as a theoretical improvement of 28.1% over the best fixed-ratio joint (the conclusion states 20%). Integrated into a 45 kg humanoid, the design produced a 0.5 m vertical jump, a 1.1 m forward jump, and a 0.5 m box jump.","feed_headline":"Knee that shifts ratio mid-jump lifts a humanoid 0.5 m","feed_subtitle":"The ratio falls as the leg extends, holding a fixed motor in its high-power band and beating fixed-ratio knees by ~28%.","key_machinery":"The central object is the explosive variable reduction ratio knee (EVRR-K): a coupling law in which the transmission ratio $k(q_2)$ decreases as knee angle $q_2$ extends, so a high ratio near the crouched position amplifies torque and a low ratio near full extension keeps motor speed down. The physical implementation is a linear actuator driving a crank-guide-rod: a ball screw pushes a link that rotates the knee through a crank, and the effective ratio is a function of the crank radius $r$, frame length $S_0$, and assembly offset $\\Delta\\theta$, given by $k = \\frac{2\\pi r (S_0+r)\\sin\\theta}{Q \\sqrt{2S_0 r - 2r^2\\cos\\theta + S_0^2 + 2r^2 - 2S_0 r \\cos\\theta}}$ with $Q$ the screw lead. The design is tuned by an optimization that maximizes takeoff mechanical energy $W_{\\text{takeoff}} = \\frac{1}{2}m_{\\text{tot}}\\dot{y}_{\\text{CoM}}(t_{\\text{to}})^2 + m_{\\text{tot}} g y_{\\text{CoM}}(t_{\\text{to}})$ over $(r, S_0, \\Delta\\theta)$ under a maximum-torque 'explosive' control law and structural constraints. This parameterized ratio-angle coupling is what lets one fixed electric motor act as both a torque amplifier at the start of a jump and a speed-friendly drive at the end.","core_discovery":"On its own terms, the paper's discovery is that the knee-to-CoM transmission ratio mismatch is not a control problem but a mechanical-design problem. For a fixed-ratio knee, the ratio between motor speed and CoM speed grows steeply as the knee extends, so a motor sized for the start of the jump is forced to very high speed, and therefore high loss, at the end. The EVRR-K couples the reduction ratio to the joint angle so that the product of motor torque and speed stays close to the motor's peak-power plateau: a high initial ratio builds torque quickly, and the declining ratio caps the motor-speed rise. The paper claims that optimizing the crank length, frame offset, and angular offset of the guide-rod mechanism, using a takeoff-energy objective under maximum-torque control, yields a monotonically decreasing ratio curve whose simulated jump height beats the optimal fixed-ratio knee by 28.1% (abstract; 20% in the conclusion), and that the mechanism delivers this in hardware: a 63 cm jump on a 24.93 kg single-joint platform and, on the 45 kg humanoid BHR8-J1, a 0.5 m vertical, 1.1 m forward, and 0.5 m box jump.","pith_inferences":["A natural extension the paper does not test: if the ankle were also given a variable-ratio treatment, takeoff energy should rise further; the paper identifies the ankle as its main unaddressed limitation, and a knee-limited model likely leaves that margin on the table.","The 28.1% (abstract) versus 20% (conclusion) improvement figure likely depends on which fixed-ratio baseline and initial angle is chosen; a reader comparing against other robots should focus on the absolute jump numbers rather than the single percentage.","The same variable-ratio principle might transfer to other explosive tasks, such as squatting lifts or stair-springing, where the load-speed profile changes over the motion, though the paper only tests jumps.","A direct A/B on the full robot, with the same control and motor and only the ratio curve swapped, would isolate the mechanism's contribution; the paper's evidence is a single configuration plus a simulated fixed-ratio comparison."],"forward_implications":["A single electric knee motor can cover the full explosive-jump torque-speed profile, so jump performance no longer requires a larger, heavier motor or hydraulic actuation.","The optimized ratio curve keeps the knee motor below about 3000 rpm during takeoff, in the high-power, low-loss band, whereas a fixed-ratio joint would need over 4000 rpm and enter the loss region.","Because the ratio curve is set by three geometric parameters, the same guide-rod mechanism can be re-tuned for different limb lengths, masses, and initial crouch angles.","On the full robot, the knee peaks at 286 Nm, 15.5 rad/s, and 1.5 kW during the box jump, with hip and ankle joints staying below 200 Nm and 1.2 kW, confirming the knee as the bottleneck the design targets.","The reported 0.5 m box jump was not limited by the mechanism: the paper notes neither the vertical nor the forward jump reached optimal performance, so the same joint should yield more once the control law exploits the variable ratio."],"supporting_citations":[{"why":"Supplies the jumping control and trajectory method used for the full-robot jump tests.","marker":"[10]"},{"why":"Underpins the explosive control strategy of driving the joint at its maximum available torque.","marker":"[12]"},{"why":"Motivates the trade-off between transmission efficiency and leg inertia used to justify the ratio design.","marker":"[14]"},{"why":"Provides a rotary-actuator comparison with a high-torque knee that still represents the fixed-ratio limit.","marker":"[16]"},{"why":"Gives a 0.5 m vertical-jump baseline from a compact electric humanoid.","marker":"[21]"},{"why":"Gives a high-torque cable-driven knee that still only jumps 0.3 m, supporting the mismatch diagnosis.","marker":"[23]"},{"why":"Provides the 0.5 m / 1.4 m / 0.4 m benchmark that the proposed design is compared against.","marker":"[24]"},{"why":"Supports the choice of a 10 mm lead dual-lead ball screw for transmission efficiency.","marker":"[32]"}],"fun_headline_variants":["Variable-ratio knee boosts humanoid jump by 28%","Knee with shifting gear ratio lifts robots 0.5 m","Angle-adaptive knee gear enhances humanoid jumps","Jumping knee adjusts ratio mid-jump to beat fixed gears","Single-joint knee with variable ratio jumps 63 cm"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The claimed improvement is computed with a model in which the knee is the only active joint and the center of mass moves straight up, and the paper acknowledges that the ankle's role is not addressed.","fun_headline_variants_meta":{"raw":{"variants":["Variable-ratio knee boosts humanoid jump by 28%","Knee with shifting gear ratio lifts robots 0.5 m","Angle-adaptive knee gear enhances humanoid jumps","Jumping knee adjusts ratio mid-jump to beat fixed gears","Single-joint knee with variable ratio jumps 63 cm"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000717,"raw_usage":{"total_tokens":3269,"prompt_tokens":1042,"completion_tokens":2227,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":658,"completion_tokens_details":{"reasoning_tokens":2144}},"tokens_in":658,"tokens_out":2227,"duration_ms":49469,"temperature":1.0,"reasoning_tokens":2144,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T00:53:16.381856+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run an A/B jump test on the same robot with the same motor, control law, and takeoff angle, swapping only the knee ratio curve: if the EVRR-K joint does not measurably out-jump the best fixed-ratio knee, or if the knee motor's speed exceeds the high-loss region before takeoff in the EVRR-K case, the central claim fails. A cheaper check: instrument the knee during the reported 0.5 m box jump and verify that motor speed stays below roughly 3000 rpm and joint power remains near 1.5 kW through the late takeoff phase.","supporting_citations":[{"cited_title":"Vertical jump of a humanoid robot with cop-guided angular momentum control and impact absorption,","cited_arxiv_id":null,"evidence_quote":"Supplies the jumping control and trajectory method used for the full-robot jump tests."},{"cited_title":"Motion planning for bipedal robot to perform jump maneuver,","cited_arxiv_id":null,"evidence_quote":"Underpins the explosive control strategy of driving the joint at its maximum available torque."},{"cited_title":"Proprioceptive actuator design in the mit cheetah: Impact mitigation and high-bandwidth physical interaction for dynamic legged robots,","cited_arxiv_id":null,"evidence_quote":"Motivates the trade-off between transmission efficiency and leg inertia used to justify the ratio design."},{"cited_title":"Unitree robotics - official website,","cited_arxiv_id":null,"evidence_quote":"Provides a rotary-actuator comparison with a high-torque knee that still represents the fixed-ratio limit."},{"cited_title":"Drive-train design in jaxon3-p and realization of jump motions: Impact mitigation and force control performance for dynamic motions,","cited_arxiv_id":null,"evidence_quote":"Gives a high-torque cable-driven knee that still only jumps 0.3 m, supporting the mismatch diagnosis."},{"cited_title":"Experimental evaluation of mechanical and electrical power consumption of feed drive systems driven by a ball-screw,","cited_arxiv_id":null,"evidence_quote":"Supports the choice of a 10 mm lead dual-lead ball screw for transmission efficiency."}],"review_version":1}