{"id":"bc05c905-b05a-432a-a360-35588ee8b5e8","arxiv_id":"2505.12649","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A low-cost, open-source quadruped robot with modular telescoping legs and backdrivable actuators is presented as a physical testbed for comparing animal morphologies.","lead":"The paper introduces TROT, a roughly $4,000 modular quadruped robot with telescoping, reconfigurable legs meant to mimic many animal body plans. It is offered as an open-source testbed for biomechanics and legged-robotics experiments that are usually done in simulation.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Quantitative COT demonstration rests on dimensionally inconsistent motor equations; Table 2's multi-kilowatt powers are implausible for a 10 kg robot, so the flagship biomechanical hypothesis test is unsupported.","rationale":"The reader's verdict is CONDITIONAL and I agree, but my stress-test pass identifies a different load-bearing concern than the reader's weakest_assumption. The GRF simplification in §2.6 is real and self-admitted, but the cost-of-transport experiment is the only quantitative demonstration of the paper's stated capability for biomechanical hypothesis testing. If Eqs. 3–5 are dimensionally wrong, the COT result provides no reliable evidence for the central claim. This is not an external-disagreement concern; it is an internal consistency problem in a key calculation. The modular hardware, open-source CAD, and kinematic comparison across limb ratios are credible contributions and can stand independently. A direct power measurement or corrected re-derivation would settle the issue cheaply. I therefore recommend keeping the conditional verdict: the platform may be valuable, but the paper needs a corrected or removed COT claim and a clear statement of which demonstrations actually support the abstract's promise.","tokens_in":11462,"tokens_out":8410,"duration_ms":96855,"concrete_test":"Recompute Table 2 using a consistent DC motor model: I = τ/K_t (K_t in Nm/A), V = ω·K_e + I·R, using the actuator constants from refs. [22] and [24], and the actual trial speed from motion capture. Then directly measure battery current and voltage during the same trotting trial on a power meter. If the measured mean electrical power and recomputed COT are within, say, 20% of Table 2, the concern fails. If the measured power is far lower (e.g., below 500 W) or differs from Table 2 by a large factor, the COT experiment and its biological conclusion are unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that TROT be usable for biomechanical hypothesis testing. The paper's only quantitative biological hypothesis test is the cost-of-transport experiment in §4.3, and it is not currently supported. Table 2 reports mean electrical power near 5 kW and COT values above 100 for a 10 kg robot moving at about 1 m/s; these are physically implausible (typical quadruped COT values are below 1). The calculation rests on Eqs. 3 and 5. Eq. 3 writes I = τ·Kτ, where Kτ is labeled a torque constant; dimensionally this yields N·m · N·m/A, not amperes, so the current is wrong unless Kτ is an undocumented reciprocal constant. Eq. 5 as printed, V = ω·K_B ∗ I·R, multiplies the back-EMF and resistive voltage terms instead of summing them; the right side has units of power, not voltage. The prose describes the intended sum, but the equation and the resulting table are inconsistent with a standard DC motor model. Additionally, §4.3 says the trial was at 1 m/s for 2 s while §3.3 says 0.3 m/s for 3 m; the COT denominator depends on V, so this ambiguity changes the result by a factor greater than three. Because the COT experiment is the paper's demonstration of biomechanical hypothesis testing, the quantitative support for the central claim is not yet established. A secondary, self-admitted limitation is the uncalibrated serial-link GRF simplification in §2.6; it affects the sensing demonstration but is less central than the broken energy calculation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents TROT, a low-cost (~$4,000), open-source, modular quadrupedal robot built from 3D-printed parts and off-the-shelf components. The authors describe the actuator, limb, torso, sensing, and control architecture, and they report three demonstrations: proprioceptive ground-reaction-force estimation validated against a force platform, kinematic foot-trajectory comparisons across three tibia-to-femur length ratios, and a cost-of-transport (COT) experiment intended to test a biological hypothesis about limb moment of inertia. The central claim is that TROT provides a reproducible physical testbed for biomechanical hypothesis testing across extant, extinct, and theoretical limb morphologies.","tokens_in":11716,"tokens_out":3796,"duration_ms":43667,"significance":"If the hardware, open-source CAD/software, and kinematic results hold up, TROT would fill a real gap between expensive commercial quadrupeds and one-off, species-specific robots. The explicit low cost, modularity, and external validation against a force platform are genuine strengths, as is the kinematic demonstration that simple limb-ratio changes produce large trajectory differences. The COT experiment is the paper's only quantitative biological hypothesis test, however, and it is currently compromised by dimensional errors in the electrical power model and by protocol inconsistencies. Because the platform claim itself is plausible and the errors appear correctable, the manuscript merits revision rather than rejection.","major_comments":[{"comment":"Equation (5), printed as V = ωK_B ∗ IR, is dimensionally inconsistent: the left side is a voltage, while the right side is the product of two voltages and therefore has units of voltage squared. The surrounding prose correctly describes the back-EMF term and the resistive drop as additive, so the intended equation is evidently V = ωK_B + IR. As printed, however, every power value in Table 2 is computed from a wrong voltage model, because Eq. (6) uses P = IV. This is a load-bearing error for the COT demonstration and must be corrected and the analysis re-run.","section":"§3.3, Eq. (5)"},{"comment":"Equation (3), I = τKτ, is dimensionally wrong for a conventional torque constant: if Kτ has units of N·m/A, the product τKτ has units of (N·m)^2/A, not amperes. The relation should be I = τ / Kτ unless Kτ is explicitly redefined as a reciprocal constant, which the text does not do. Since all current, power, and COT values in Table 2 propagate from this expression, the dimensional error invalidates the quantitative results as reported.","section":"§3.3, Eq. (3)"},{"comment":"The experimental protocol is inconsistent between sections: §3.3 states TROT ran three meters at 0.3 m/s, while §4.3 and Table 2 report data taken at 1 m/s for 2 seconds. The COT denominator in Eq. (6) contains the velocity V directly, so this discrepancy changes the reported COT values by a factor of more than three. The authors must state the exact protocol and ensure that the velocity used in Eq. (6) matches the actual trial.","section":"§3.3 versus §4.3 and Table 2"},{"comment":"The reported mean electrical powers (roughly 5.1, 9.4, and 9.6 kW for a 10 kg robot at about 1 m/s) and COT values (126–195) are physically implausible for a small quadruped; typical quadrupedal COT values are of order unity or below. Even after correcting Eqs. (3) and (5), if the recalculated powers remain in the kilowatt range, the authors should investigate potential unit, sign, or logging errors. As presented, the flagship biological hypothesis test is quantitatively unsupported.","section":"Table 2 and §4.3"},{"comment":"The paper acknowledges that modeling the 4-bar linkage as a serial chain introduces a linear scaling error that can be manually calibrated for each axis, but no calibrated values or dynamic validation during trotting are provided. Table 1 shows RMSE values up to 11.15 N in the vertical direction, which is non-negligible relative to the measured GRF magnitudes. This limitation weakens the onboard-sensing demonstration and should be addressed, at least by providing the calibration procedure and a dynamic validation trial.","section":"§2.6 and §4.1"}],"minor_comments":[{"comment":"The text says the three limb configurations each have a total length of 400 cm; for a 10 kg quadruped the intended value is presumably 400 mm, and the manuscript should be corrected to avoid confusion.","section":"§3.2"},{"comment":"The symbol V is used both for motor voltage in Eq. (5) and for robot velocity in Eq. (6), which invites confusion and may have contributed to the protocol inconsistency; a distinct symbol such as v for velocity would improve clarity.","section":"§3.3 and Eq. (6)"},{"comment":"The caption refers to the \"X-axis\" and \"y-axis\" with inconsistent capitalization; the axes should be defined consistently and the marker offset explained in the caption itself.","section":"Figure 4 caption"},{"comment":"The rendering of the Jacobian in Eq. (1) has spacing and subscript formatting issues that make it hard to read; a cleaner typeset version would help readers verify the derivation.","section":"Eq. (1)"}],"recommendation":"major_revision","confidential_remarks":"The self-citation to the prior actuator characterization [22,24] is legitimate reuse and not a concern. The kinematic and hardware contributions are likely sound and of interest to the robotics community. The COT errors are substantial but appear correctable: the authors should fix the electrical equations, re-run the analysis, and reconcile the speed protocol. If the originally reported kilowatt-level powers persist after correction, the paper's quantitative biological claim would remain unsupported and the recommendation would need to be revisited."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Urs et al. describe TROT, a ~$4k, open-source quadruped with telescoping two- or three-link limbs, reversible knee/elbow orientation, and quasi-direct-drive actuators. That combination is genuinely new among accessible platforms, and the hardware design, actuator choice, and adapted Cheetah MPC controller are credible for a biomechanics-oriented testbed. The kinematic experiment comparing three tibia:femur ratios is a reasonable demonstration, and the authors are honest about the serial-link simplification in the GRF estimator.\n\nThe soft spots are real and load-bearing for the quantitative claims. The cost-of-transport experiment in Section 4.3 is the paper's only biological hypothesis test, and it currently does not hold up. Eq. 3 as printed gives I = tau*K_tau, which is dimensionally wrong unless K_tau is secretly a reciprocal constant. Eq. 5 writes V = omega*K_B * I*R, multiplying back-EMF and resistive terms instead of summing them; the right side has units of power, not voltage. The prose describes the intended sum, but the equation as written is not a DC motor model. Table 2's mean electrical powers above 5 kW and COT values above 100 for a 10 kg robot at 1 m/s are implausible on their face; typical quadruped COT is well below 1. The speed inconsistency (0.3 m/s in Section 3.3 vs 1 m/s in Section 4.3/Table 2) further muddies the result. The stress-test note had this right.\n\nThe GRF validation also shows high errors (up to 11 N in one vertical channel), and the authors acknowledge the 4-bar linkage simplification produces linear scaling error that can be calibrated but provide no calibration constants or dynamic validation during actual locomotion. The sensing demonstration is thus partial, not conclusive. The abstract's promise of testing extant, extinct, and theoretical morphologies overreaches; the experiments only cover three extant-like limb ratios and one inertia manipulation, with no extinct or theoretical forms actually built or run.\n\nStill, the platform itself is a plausible contribution. The design choices - modular links, backdrivable QDD actuators, parameterized control - are well-motivated, and the code/CAD with a DOI is a genuine reproducibility feature. The central idea deserves referee time, but not in its current quantitative state. The paper needs the energy calculation corrected from first principles, the speed ambiguity resolved, the GRF calibration performed or clearly stated as a limitation with error bounds, and the claims trimmed to what was actually demonstrated.\n\nVerdict: conditional accept after major revision. The hardware is worth engaging with; the current energy and sensing numbers are not. I'd want to see the corrected equations and a rerun of the COT comparison before trusting the biology claim.","headline":"A genuinely useful modular quadruped platform, but the flagship cost-of-transport numbers rest on broken motor equations and need major revision before the biological claims can stand.","tokens_in":763,"tokens_out":2316,"would_cite":false,"duration_ms":32423,"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":"TROT is a $4,000 open-source quadruped whose modular, telescoping legs can be reconfigured to match a wide range of animal limb morphologies, giving biologists a physical testbed for locomotion hypotheses.","keywords":["modular quadruped robot","legged locomotion","morphology","quasi-direct drive actuator","ground reaction force","cost of transport","open-source hardware","additive manufacturing"],"falsifier":"Run TROT on a force-measuring treadmill at the same 0.3 m/s trot used in the cost-of-transport experiment and compare the forces estimated from motor torques with the treadmill's readings; if the dynamic errors are as large as the static errors reported in the paper (up to roughly 11 newtons vertically) or the measured cost of transport shifts by more than the claimed 7.5% effect, the sensing-based results are not supported.","tokens_in":11225,"feed_emoji":"🤖","tokens_out":7343,"duration_ms":70206,"temperature":0.7,"pith_summary":"The paper presents TROT, a 10 kg quadrupedal robot that can be built for under $4,000 from 3D printed parts and off-the-shelf components. Its limbs are modular: each leg has two or three rigid links, the proximal joint can be a knee or elbow, and telescoping sections change link lengths, so the robot can approximate the limb proportions of many quadrupedal mammal families. The authors argue that such a reconfigurable, proprioceptive platform lets researchers empirically test biomechanical hypotheses that currently require either expensive commercial robots or simulation. They demonstrate this with three experiments: actuator-based ground reaction force estimates, foot trajectories across tibia:femur ratios, and a cost-of-transport comparison of distally versus proximally added mass. The central claim is that the platform provides a reproducible, accessible physical testbed for comparing extant, extinct, and theoretical limb designs.","feed_headline":"A $4,000 modular robot reshapes its legs to test animal gaits","feed_subtitle":"Telescoping, torque-sensing legs let a cheap quadruped test how leg shape changes gait and energy cost.","key_machinery":"The load-bearing mechanism is the modular limb built around quasi-direct drive actuators—high-torque, backdrivable motors whose current draw measures joint torque—with power transmitted through a four-bar linkage that couples the tarsus angle to the femur angle, keeping actuators proximal to minimize limb inertia. Telescoping tibia and tarsus segments set link lengths, and swappable gear ratios (7.5:1 or 15:1) expand the reachable length-ratio space. Ground reaction force is computed as $\\mathrm{GRF} = (J^T)^{-1}\\tau$ from the leg Jacobian and actuator torques, treating the closed four-bar chain as a serial three-link chain; the paper states this simplification introduces a linear scaling error that can be manually calibrated per axis. This sensing scheme is what lets the robot measure whole-body dynamics without force plates or foot sensors.","core_discovery":"The central claim is that a low-cost quadruped with modular legs can serve as a physical model for testing how limb morphology shapes locomotion, and that the robot as built realizes that capability. TROT's legs can be arranged in three- or four-link configurations, with knee/elbow orientation, torso length, and link lengths all adjustable; with a 15:1 gear ratio the reachable limb-length ratios span felids, canids, rodents, lagomorphs, artiodactyls, and perissodactyls. The paper reports that actuator-derived ground reaction force estimates track a force platform with root-mean-square errors between roughly 1.3 and 11.2 N depending on axis and leg, that changing the tibia:femur ratio from 50:50 to 45:55 or 55:45 visibly changes foot trajectory and hip translation, and that adding 500 g to the tibiae instead of the femora raises cost of transport by 44.1%, supporting the hypothesis that distal limb mass is energetically costly.","pith_inferences":["If the ground reaction force estimates can be validated during dynamic trotting after per-axis calibration, the same torque-sensing scheme could be extended to field biomechanics, where force plates are unavailable.","The 44.1% cost-of-transport gap between femur- and tibia-loaded limbs is a strong candidate for a controlled test of elastic energy storage: adding virtual or physical series elasticity to the distal joints should reduce that gap if the penalty is chiefly inertial.","The modular limb design effectively samples a morphospace of limb proportions; a natural next step is to systematically sweep tibia:femur ratio at fixed speed and measure cost of transport, producing a map that could be compared with allometric data from living mammals.","The paper leaves the four-bar-to-serial scaling correction uncalibrated; publishing per-axis correction factors would let other users reproduce the sensing results on different builds."],"forward_implications":["Biology laboratories without large equipment budgets can build a quadrupedal testbed and vary morphology, gait, and controller in ways previously limited to commercial robots or simulation.","Hypotheses about extinct or theoretical limb configurations, such as limb proportions that no living animal exhibits, can be tested physically rather than only in simulation.","The cost-of-transport data support the claim that shifting mass distally along the leg increases energetic cost, giving a concrete empirical target for theories of limb design.","The foot-trajectory results imply that even a modest change in tibia:femur ratio meaningfully changes foot clearance and hip translation under identical control, so morphological changes alone can alter gait outcomes.","Because the actuators estimate ground reaction forces, locomotion can be studied on natural or uneven substrates without instrumented force plates, if the quoted calibration errors can be reduced."],"supporting_citations":[{"why":"Supplies the design and characterization of the quasi-direct drive actuators that provide torque sensing and high output torque.","marker":"[22]"},{"why":"Provides the open-source low-level control software that TROT adapts for parameterized locomotion and user-defined gaits.","marker":"[14]"},{"why":"Supplies limb-length ratio data from mammals that define the morphological space TROT can match.","marker":"[26]"},{"why":"Demonstrates the previous use of modeling a four-bar-linkage leg as a serial chain, justifying the simplified Jacobian and its linear scaling error.","marker":"[30]"},{"why":"Establishes the motor-commutation technique that converts motor current into torque values for ground-reaction-force estimation.","marker":"[27]"},{"why":"Provides motor selection metrics and the empirically determined back-EMF constant used in the cost-of-transport power calculation.","marker":"[24]"},{"why":"Motivates the cost-of-transport experiment by reporting no significant cost-of-transport difference between cheetahs, gazelles, and goats despite differing limb inertia.","marker":"[18]"},{"why":"Inspires the four-bar tarsus coupling from mammalian hindlimb kinematics.","marker":"[25]"}],"fun_headline_variants":["$4,000 modular robot swaps leg shapes to test animal gaits","TROT: a cheap quadruped that reshapes legs for gait studies","Low-cost robot with telescoping legs mimics diverse animal morphologies","Robot of Theseus tests how leg morphology changes locomotion cost"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The force and energy measurements assume that the motor torques in each leg simply balance the force from the ground, with the leg treated as a chain of three straight segments; the paper acknowledges that the real four-bar linkage makes this a scaling error that is left uncalibrated and untested while the robot is moving.","fun_headline_variants_meta":{"raw":{"variants":["$4,000 modular robot swaps leg shapes to test animal gaits","TROT: a cheap quadruped that reshapes legs for gait studies","Low-cost robot with telescoping legs mimics diverse animal morphologies","Robot of Theseus tests how leg morphology changes locomotion cost"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.0005,"raw_usage":{"total_tokens":2489,"prompt_tokens":1029,"completion_tokens":1460,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":645,"completion_tokens_details":{"reasoning_tokens":1386}},"tokens_in":645,"tokens_out":1460,"duration_ms":11960,"temperature":1.0,"reasoning_tokens":1386,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:29:56.328136+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run TROT on a force-measuring treadmill at the same 0.3 m/s trot used in the cost-of-transport experiment and compare the forces estimated from motor torques with the treadmill's readings; if the dynamic errors are as large as the static errors reported in the paper (up to roughly 11 newtons vertically) or the measured cost of transport shifts by more than the claimed 7.5% effect, the sensing-based results are not supported.","supporting_citations":[{"cited_title":"Design and characterization of 3d printed, open-source actuators for legged locomotion,","cited_arxiv_id":null,"evidence_quote":"Supplies the design and characterization of the quasi-direct drive actuators that provide torque sensing and high output torque."},{"cited_title":"Mini cheetah: A platform for pushing the limits of dynamic quadruped control,","cited_arxiv_id":null,"evidence_quote":"Provides the open-source low-level control software that TROT adapts for parameterized locomotion and user-defined gaits."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies limb-length ratio data from mammals that define the morphological space TROT can match."},{"cited_title":"Overconstrained coaxial design of robotic legs with omni-directional locomotion,","cited_arxiv_id":null,"evidence_quote":"Demonstrates the previous use of modeling a four-bar-linkage leg as a serial chain, justifying the simplified Jacobian and its linear scaling error."},{"cited_title":"Sensorless field oriented control of brushless permanent magnet synchronous motors,","cited_arxiv_id":null,"evidence_quote":"Establishes the motor-commutation technique that converts motor current into torque values for ground-reaction-force estimation."},{"cited_title":"Alternative Metrics to Select Motors for Quasi-Direct Drive Actuators","cited_arxiv_id":"2202.12365","evidence_quote":"Provides motor selection metrics and the empirically determined back-EMF constant used in the cost-of-transport power calculation."},{"cited_title":"Running in cheetahs, gazelles, and goats: energy cost and limb configuration,","cited_arxiv_id":null,"evidence_quote":"Motivates the cost-of-transport experiment by reporting no significant cost-of-transport difference between cheetahs, gazelles, and goats despite differing limb inertia."},{"cited_title":"Basic limb kinematics of small therian mammals,","cited_arxiv_id":null,"evidence_quote":"Inspires the four-bar tarsus coupling from mammalian hindlimb kinematics."}],"review_version":1}