{"id":"f67bcb6b-a904-4f06-9b74-8a92fd64c053","arxiv_id":"2411.11788","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"A simulation demonstrates that a variable-length inverted pendulum model with thruster forces and a quadratic-programming controller can track a 40-degree slope reference motion, but hardware transfer is unproven.","lead":"The paper simulates a simplified inverted-pendulum model with thruster forces, controlled by a quadratic-programming optimizer, to climb a 40-degree slope. It is a step toward giving the legged-aerial robot Husky the ability to run up steep slopes using both legs and thrusters, similar to young birds.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'seamlessly transferred to full dynamics' claim in Sec. IV is contradicted by the paper's own conclusion, which defers kinematic/dynamic constraints and thrust-directioning to future work; no full-HROM or hardware validation supports the transfer step.","rationale":"I considered two additional internal risks: Eq. (13) has a dimensionally inconsistent input matrix relative to Eq. (12) (B's second row should be [-1/m, -(xcop,0-xcom,0)/(m y0)] if u=[lambda_x, lambda_y]), and Eq. (18) imposes only lambda_x < mu |lambda_y|, omitting lambda_x > -mu |lambda_y|. Both would be serious if the code mirrors the text and should be corrected in revision. However, the transferability claim is more load-bearing because the strongest claim explicitly includes it, and the authors' own conclusion admits the required control and hardware elements are future work. The VLIP simulation, qpSWIFT solve times, and the friction-ratio plot are useful internal evidence, but they do not speak to COP manipulation or thrust vectoring on Husky. The reader's weakest assumption already points to this gap; my read therefore does not move the verdict from CONDITIONAL.","tokens_in":8787,"tokens_out":11985,"duration_ms":124188,"concrete_test":"Run the full HROM dynamics (Eq. 8) on the 40-degree slope using the QP-MPC's optimal ground-reaction and thruster force profiles from Sec. IV, with the commanded fixed-COP realized through a leg force distribution and with actual thruster directions enforced. If the body diverges from the reference or any foot slips, the 'seamlessly transferred' claim is falsified. A minimal variant: repeat the same tracking experiment with a whole-body controller implementing the commanded COP and thrust vector before changing anything else.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim has two parts: the VLIP/QP simulation on a 40-degree slope, and the assertion in Sec. IV that the resulting forces 'can be seamlessly transferred to the full dynamics.' The simulation part is only demonstrated on the planar VLIP with a hand-designed, fixed-COP reference (Sec. V). The transfer part is not demonstrated at all, and the paper's own conclusion explicitly defers what transfer would require: 'strictly implementing the kinematic and dynamics constraints would be required,' including finding and enforcing body pitch and roll to vector thrust, and 'a method of thrust regeneration using baffles and/or posture manipulation is being looked into.' Commanding an arbitrary COP inside the support polygon is not the same as realizing it with four legs, and the computed two-dimensional thruster forces assume a thrust-vectoring capability that is neither modeled nor measured. Without a full-HROM or hardware test, the strongest claim's second half rests on an unvalidated leap; the verified contribution is only a reduced-order simulation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a planar modified Variable Length Inverted Pendulum (VLIP) model with thruster forces applied at the center of mass, and a QP-based MPC that computes optimal ground reaction forces and thruster forces to track a reference trajectory on a 40-degree slope. The authors simulate this reduced-order model in MATLAB, report tracking performance, friction-cone satisfaction, and qpSWIFT solve times, and claim that the resulting forces can be seamlessly transferred to the full Husky dynamics. The paper concludes by listing future work needed for hardware implementation, including kinematic and dynamic constraints, body pitch/roll thrust vectoring, and thrust regeneration via posture manipulation.","tokens_in":9033,"tokens_out":7519,"duration_ms":69754,"significance":"If the reduced-order control framework were validated against the full-order Husky model or hardware, it would be a useful step toward thruster-assisted steep-slope locomotion, building on established VLIP and QP-MPC techniques. The paper's strengths are its simple, real-time-compatible formulation, explicit friction-cone constraints in the optimization, and a slope-independent state model that depends only on the COP-COM offset. However, the current evidence is limited to a simulation of the reduced-order model, in which the thruster forces are computed from the same equations of motion used to propagate the state, so the results demonstrate internal consistency rather than independent predictive validity. The manuscript's own conclusion contradicts the strong transferability claim in Section IV, and the simulation parameters are not fully reported, which limits reproducibility.","major_comments":[{"comment":"The linearized dynamics in Eq. (13) are inconsistent with Eq. (12). Using x_cop - x_com = (x_cop,0 - x_com,0) - (x_com - x_com,0), Eq. (12) linearizes to A(2,1) = +λ_y0/(m y0) and B(2,2) = -(x_cop,0 - x_com,0)/(m y0), whereas the paper prints A(2,1) = -λ_y0/(m y0) and B(2,2) = -(x_cop,0 - x_com,0)/y0. The sign error changes the open-loop pole locations, and the missing factor 1/m rescales the control input. Unless a different normalization is intended but not stated, the QP in Section IV is built on a different model than the one derived in Section III.","section":"Section III, Eq. (13)"},{"comment":"The friction-cone constraint is one-sided. For positive normal forces λ_y,k ≥ λ_minn > 0, the correct no-slip condition is |λ_x,k| ≤ μ_s λ_y,k, i.e., -μ_s λ_y,k ≤ λ_x,k ≤ μ_s λ_y,k. The printed condition λ_x,k < μ_s |λ_y,k| imposes only an upper bound on λ_x and permits arbitrarily large negative tangential forces, which violates the no-slip assumption and undermines the claim that the QP output is constraint-admissible.","section":"Section IV, Eq. (18)"},{"comment":"The claim that the QP-derived thruster forces 'can be seamlessly transferred to the full dynamics' is not supported by the manuscript. The simulation in Section V is run on the planar VLIP model, not on the HROM described in Section II, and no mapping from the commanded ground reaction and thruster forces to joint torques, leg lengths, body orientation, or the thrust-vectoring mechanism is provided. The paper's own conclusion states that 'strictly implementing the kinematic and dynamics constraints would be required,' including body pitch and roll for thrust vectoring and a method of thrust regeneration. The transfer claim should be removed or substantiated with at least a full-HROM simulation.","section":"Section IV, last paragraph; Section VI"},{"comment":"The assumption that the COP is known, fixed, and controllable anywhere within the support polygon is load-bearing but never validated against the four-legged HROM. The MPC commands a scalar offset (x_cop - x_com) through the B matrix, and the resulting thruster forces are two-dimensional; realizing both on the actual robot requires posture manipulation and thrust vectoring that the paper explicitly defers to future work. The simulation therefore demonstrates internal feasibility of the reduced-order model, not feasibility on Husky.","section":"Sections III and V"},{"comment":"The simulation results are not reproducible as reported. The MPC weights Q and R, prediction horizon n_h, discretization time step, λ_minn, μ_s, the reference trajectory, and the COP update rule are not specified; only the slope angle and the solver's average solve time are given. Without these parameters, the tracking and constraint-satisfaction results in Figs. 4-7 cannot be independently checked.","section":"Section V"}],"minor_comments":[{"comment":"The paper structure says 'In Section 1... Section 2...' but the actual sections are labeled with Roman numerals II-V; renumber the outline accordingly.","section":"Section I"},{"comment":"The constraint line 'Umin and Umin' should read 'Umin and Umax'.","section":"Section IV"},{"comment":"The caption says the forces are 'from (9)', but the thruster forces are computed from Eq. (10); correct the equation reference.","section":"Fig. 6 caption"},{"comment":"The figure plots λ1 and λ2 but does not define these variables in the text or caption; define them.","section":"Fig. 7"},{"comment":"The spacing in 'W AIR' is inconsistent; use 'WAIR' consistently, and fix the incomplete sentence ending 'friction cone constraints and.' in Section I.","section":"Throughout"},{"comment":"The title contains a typo: 'referance' should be 'reference'.","section":"Fig. 4"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core of this paper is a reduced-order VLIP model with COM-collocated thruster forces, a fixed-ZMP assumption, and a QP-based MPC that tracks a 40-degree slope in simulation. That is a reasonable and modest extension of prior work, and the simulation itself looks internally consistent: the tracking plots show convergence, the solve times are respectable, and the equations of motion are mostly clear. The authors deserve credit for a tractable formulation and for being explicit about the planar, fixed-COP setup.\n\nThe main problem is the claim in Section IV that the thruster forces \"can be seamlessly transferred to the full dynamics.\" The stress-test note lands squarely on this. The paper's own conclusion defers exactly what transfer would require: enforcing kinematic and dynamic constraints, finding body pitch and roll to vector thrust, and regenerating thrust via baffles or posture manipulation. That is not a minor caveat; it changes what the paper has actually shown. What is demonstrated is a reduced-order feasibility study, not a controller ready for Husky. The fixed-COP assumption is also load-bearing and unvalidated against the full HROM or any hardware.\n\nThere are two additional technical issues worth flagging. The linearized A matrix in (13) appears inconsistent with the sign in (12); the state-dependent term from (xcop - xcom)·λy should produce a positive xcom coefficient when linearized, so the negative sign in A needs re-derivation or a clarifying assumption. The friction cone condition in (18) is one-sided, allowing arbitrarily large negative λx as long as λy is sufficiently positive; that should be a symmetric |λx| ≤ μ·λy constraint.\n\nOther soft spots are softer: parameter values are mostly undisclosed, the COP update rule is a hand-tuned heuristic, and there is no disturbance or robustness analysis. These are correctable rather than fatal.\n\nWho gets value from this? Researchers working on legged-aerial platforms or WAIR-style maneuvers, as a quick feasibility check and a starting point for full-order simulation. It is not a hardware validation and should not be cited as one. The paper deserves a serious referee — the idea is coherent and the simulation is reproducible in principle — but it needs substantial revision to align its claims with its evidence.\n\nRecommendation: send to peer review with the expectation that the transferability claim be either removed or heavily qualified, and that the sign and friction-cone issues be fixed.","headline":"Modest simulation-only VLIP/MPC extension for thruster-assisted steep-slope walking, undermined by an unsupported 'seamless transfer' claim that the paper's own conclusion contradicts.","tokens_in":9611,"tokens_out":2604,"would_cite":false,"duration_ms":27026,"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":"A VLIP model with center-of-mass thrusters plus QP-MPC tracks a 40-degree slope in simulation, and the paper argues the forces transfer to full dynamics.","keywords":["variable length inverted pendulum","wing-assisted inclined running","quadratic programming MPC","thruster-assisted locomotion","legged-aerial robot","friction cone constraints","steep slope walking","reduced-order model"],"falsifier":"Run the same QP-computed ground reaction and thruster force profiles on the full Husky reduced-order model (HROM) with no-slip feet and posture-based thrust vectoring, and check whether the body follows the reference trajectory; if the full dynamics diverge, or if the required COP lies outside the support polygon, the central transferability claim fails. Alternatively, on hardware, measure whether the actual thruster wrench from the ducted fans matches the commanded $F_x$ and $F_y$ within the friction cone while attempting a 40-degree slope walk.","tokens_in":1682,"feed_emoji":"🤖","tokens_out":2239,"duration_ms":65008,"temperature":0.7,"pith_summary":"The paper claims that a thruster-assisted legged robot can walk up a 40-degree slope if the control problem is reduced to a planar variable-length inverted pendulum with thruster forces applied at the center of mass and a known, fixed center of pressure. A quadratic-programming model predictive controller solves for ground reaction forces and thruster forces that track a reference trajectory while respecting friction-cone constraints, and the computed forces are said to be transferable to the full Husky dynamics. If correct, this gives a tractable reduced-order control route to wing-assisted-incline-running-style maneuvers on a real legged-aerial platform.","feed_headline":"Thruster-assisted walking hits a 40-degree slope in simulation","feed_subtitle":"A linear pendulum model plus quadratic programming picks ground and thruster forces fast enough for real-time use.","key_machinery":"The central object is a modified Variable Length Inverted Pendulum (VLIP): a point mass connected to a fixed center-of-pressure point on the slope, with an external thruster force at the mass and ground reactions at the center of pressure. The zero-moment-point equation at the center of mass reduces the planar dynamics to the linear time-invariant model in Eq. (13), whose state is $(x_{\\text{com}}, \\dot{x}_{\\text{com}})$ and whose inputs are ground reaction forces. The quadratic-programming MPC with friction-cone constraints is the mechanism that turns this model into a controller, and the paper argues that because the center of pressure can be placed anywhere inside the support polygon by distributing forces between the two contact legs, the thruster wrench from the VLIP can be handed to the full dynamics.","core_discovery":"The slope angle drops out of the reduced-order dynamics, leaving a linear time-invariant system whose behavior depends only on the relative position of the center of mass and the center of pressure. Solving the zero-moment-point equation about the center of mass under a constant height $¥ddot{y}=0$ yields $\\ddot{x}_{\\text{com}} = -(x_{\\text{cop}}-x_{\\text{com}})\\lambda_y/(m y_0) - \\lambda_x/m$, and the ground reaction forces $\\lambda_x,\\lambda_y$ become the inputs. The MPC then chooses these inputs over a horizon to minimize tracking error subject to bounds and the friction cone $\\lambda_{y,k} > \\lambda_{\\min,n}$, $|\\lambda_{x,k}| \\le \\mu_s \\lambda_{y,k}$. In simulation on a 40-degree slope the optimizer finds feasible forces, the body tracks the reference position and velocity, and the implied thruster forces point in directions consistent with the wing-assisted inclined running intuition that legs supply traction while thrusters offload normal force.","pith_inferences":["The fixed-COP assumption is the main gap: the paper does not simulate the full HROM or hardware, so the claim that the thruster wrench 'can be seamlessly transferred' is an extrapolation; I would test robustness by perturbing the COP within the support polygon and measuring tracking degradation.","The constant normal force result (normal ground reaction stays at $\\lambda_{\\min,n}$ because of the quadratic cost) suggests the optimizer actively offloads weight to the thrusters, which implies the achievable range of slope angles is bounded by propeller force limits.","A natural extension is to replace the fixed-foothold heuristic with online foothold or step-timing optimization, which would let the same VLIP-QP framework handle irregular terrain and discrete contact switches.","The decoupling of slope angle from the reduced-order dynamics means the controller could be reused across terrains without retuning, provided the COP can still be regulated."],"forward_implications":["The reduced-order model is linear and the QP solver takes about 65 microseconds per call, so the controller is fast enough for real-time implementation on the robot's onboard computer.","Ground reaction forces in the VLIP are independent of slope angle, so a single tuning of the MPC could apply to different inclines without re-deriving the model.","The framework gives an explicit split of roles: legs provide traction through friction-cone-respecting ground forces, while thrusters offload normal force and help drive acceleration up steep slopes.","The same QP formulation can be extended to 3D movements and whole-body control, as the paper identifies as future work.","The approach offers a path to dynamic two-point contact gaits on steep slopes, going beyond the static three-foot-contact gaits used by earlier slope-walking controllers."],"supporting_citations":[{"why":"It supplies the biological WAIR maneuver that motivates the thruster-assisted walking problem and the leg/traction wing/thrust split.","marker":"[1]"},{"why":"It establishes VLIP/LIP reduced-order modeling for a legged robot that can fly, slackline, and skateboard, which is the modeling paradigm this paper modifies.","marker":"[28]"},{"why":"It provides dynamic modeling of wing-assisted inclined running on a morphing multi-modal robot, supporting the transfer of the reduced-order results to the full Husky platform.","marker":"[7]"},{"why":"It presents prior thruster-assisted incline walking results on Husky that this paper builds upon and extends.","marker":"[11]"},{"why":"It supplies the real-time QP solver used in the MPC, which is what makes the framework tractable for online control.","marker":"[29]"},{"why":"It shows QP-based MPC applied to dynamic quadruped motions, supporting the real-time feasibility and methodology of the proposed controller.","marker":"[30]"}],"fun_headline_variants":["Steep slope walking via reduced-order pendulum and QP","QP MPC balances legs and thrusters on a 40-degree slope","Slope angle vanishes from model, enabling 40-degree climb","Reduced-order VLIP plus QP walks Husky up 40-degree slope","Linear model makes slope irrelevant; QP nails 40-degree walk"],"cache_read_input_tokens":11776,"weakest_assumption_plain":"The center of pressure is assumed known, fixed, and controllable anywhere inside the support polygon, and the commanded thruster forces are assumed physically realizable by pitching the robot body; if either fails on the real Husky, the simulated 40-degree tracking would not transfer to hardware.","fun_headline_variants_meta":{"raw":{"variants":["Steep slope walking via reduced-order pendulum and QP","QP MPC balances legs and thrusters on a 40-degree slope","Slope angle vanishes from model, enabling 40-degree climb","Reduced-order VLIP plus QP walks Husky up 40-degree slope","Linear model makes slope irrelevant; QP nails 40-degree walk"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000606,"raw_usage":{"total_tokens":2797,"prompt_tokens":888,"completion_tokens":1909,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":504,"completion_tokens_details":{"reasoning_tokens":1818}},"tokens_in":504,"tokens_out":1909,"duration_ms":12269,"temperature":1.0,"reasoning_tokens":1818,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T18:08:41.822716+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same QP-computed ground reaction and thruster force profiles on the full Husky reduced-order model (HROM) with no-slip feet and posture-based thrust vectoring, and check whether the body follows the reference trajectory; if the full dynamics diverge, or if the required COP lies outside the support polygon, the central transferability claim fails. Alternatively, on hardware, measure whether the actual thruster wrench from the ducted fans matches the commanded $F_x$ and $F_y$ within the friction cone while attempting a 40-degree slope walk.","supporting_citations":[{"cited_title":"Dynamic modeling of wing-assisted inclined running with a morphing multi-modal robot","cited_arxiv_id":"2311.09963","evidence_quote":"It provides dynamic modeling of wing-assisted inclined running on a morphing multi-modal robot, supporting the transfer of the reduced-order results to the full Husky platform."},{"cited_title":"Representation-Free Model Predictive Control for Dynamic Motions in Quadrupeds,","cited_arxiv_id":null,"evidence_quote":"It shows QP-based MPC applied to dynamic quadruped motions, supporting the real-time feasibility and methodology of the proposed controller."}],"review_version":1}