{"id":"9658cd3a-2d02-47dc-9f5c-50dcf1a06fdc","arxiv_id":"2507.19928","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"An NMPC scheme that optimizes the reference orbit within a periodic orbit family, with a polynomial model of the family, reduces simulated fuel consumption for cislunar stationkeeping.","lead":"This paper presents a spacecraft controller that keeps a satellite within a chosen family of lunar libration-point orbits instead of forcing it onto one fixed path. The controller picks the best orbit in the family on the fly, which the authors say cuts fuel use versus standard orbit tracking.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Fuel-savings claim is largely guaranteed by the extra optimization variable χ; the fixed-orbit baseline comparison does not test the proposed method.","rationale":"The reader's weakest assumption was the accuracy and uniqueness of the MPR model in Eq. (2), which is indeed a serious gap: no fitting-error analysis, validation set, or uniqueness proof is provided, and the reference states in the NMPC depend entirely on that regression. However, the most load-bearing concern for the paper's central claim is the baseline comparison. Because the proposed formulation optimizes χ while the conventional baseline does not, the fuel savings in Figures 13 and 17 are partially guaranteed by construction. This does not make the method useless — family tracking may still be valuable for flexibility and for reducing sensitivity to initial conditions — but it means the quantitative headline is not yet supported. The paper deserves conditional acceptance with a required revision: either add a fair baseline, such as a best fixed-orbit NMPC with the same tuning and phase freedom, or explicitly reframe the contribution as 'optimal orbit selection within a family' rather than a universal fuel-saving controller. The MPR validation issue should also be addressed, but the baseline comparison is the more fundamental threat to the central claim.","tokens_in":9934,"tokens_out":4244,"duration_ms":63164,"concrete_test":"","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central quantitative claim — 'significant reduction in fuel consumption compared to conventional tracking methods' — follows almost directly from the problem formulation. In Eq. (3)–(4), χ_k is an optimization variable with χ_min ≤ χ_k ≤ χ_max, while the conventional baseline fixes the reference orbit (χ ≡ χ0 and, effectively, a prescribed ν schedule). Every feasible control sequence for the baseline is also feasible for the proposed problem (choose χ_k = χ0), so with identical initial states and disturbances the optimal proposed cost is always less than or equal to the baseline cost. Thus the ΔV reductions in Figures 13 and 17 are not empirical evidence of a better controller; they are the value of adding one degree of freedom to the optimizer. The size of the reported saving depends on how far the chosen fixed orbit lies from the actual disturbed trajectory, not on any demonstrated property of family tracking. This is especially consequential because the bias can be made arbitrarily large by selecting a baseline orbit that is poorly matched to the simulation scenario. A fair comparison must hold the task fixed: either constrain χ to a well-chosen fixed orbit while retaining all other formulation details, or report fuel consumption at equal achieved distance from the orbit family. Without such a test, the headline result cannot distinguish 'better tracking' from 'more freedom, looser target.'","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a Nonlinear Model Predictive Control (NMPC) scheme for keeping a spacecraft within a family of periodic orbits in the cislunar CR3BP, rather than tracking one predefined orbit. The authors use pseudo-arclength continuation to generate orbit families, parameterize each family member by two variables (χ, ν), and fit the resulting states with a multivariate polynomial regression (MPR) model. The NMPC then optimizes over the velocity impulses and the orbit parameter χ at each step, and the controller is integrated with an Extended Kalman Filter. Simulations for Lyapunov, halo, and near-rectilinear halo orbits near L1 and L2 are presented, and the paper claims a significant fuel reduction compared to conventional fixed-orbit tracking.","tokens_in":10247,"tokens_out":5498,"duration_ms":62966,"significance":"If rigorously supported, the idea of tracking an entire orbit family instead of a single reference orbit is a useful contribution to cislunar station-keeping, potentially reducing fuel by allowing the spacecraft to exploit the natural manifold structure. The paper makes a clear effort to build a practical pipeline: PAC for family generation, MPR for a compact reference model, an NMPC solver with CasADi, and an EKF-based GNC loop. However, the current evidence for the central quantitative claim is not convincing: the baseline comparison is structurally biased, the MPR model is unvalidated, the disturbance description is inconsistent, and the headline results are single deterministic runs without statistical support. The conceptual contribution is defensible, but the quantitative claims need substantial additional evidence.","major_comments":[{"comment":"The optimization over χ_k in Eqs. (3)-(4) makes the proposed problem a relaxation of any fixed-χ baseline: every feasible control sequence for the baseline (with χ_k ≡ χ0) is feasible for the proposed problem, so the proposed optimal cost is by construction no larger. The fuel reductions in Figures 13 and 17 therefore partly reflect the added degree of freedom rather than a demonstrated advantage of family tracking. To support the headline claim, please compare against the same NMPC with χ_k constrained to a fixed orbit that is selected by a principled rule (e.g., the family member closest to the initial state or a minimum-fuel fixed orbit), and report the achieved distance from the orbit family for both methods.","section":"V, Eqs. (3)-(4)"},{"comment":"The MPR model generates the reference states P(χ,ν) that appear in the NMPC cost, but no fitting error, validation set, or model-order selection is reported. The paper also asserts without proof that the (χ,ν) parameterization is unique and continuous over the region R. Since the controller tracks these regressed states, regression error directly affects the effective reference and can influence the measured fuel consumption. Please report per-sub-manifold fit errors (RMS and maximum position error) and, if possible, a closed-loop sensitivity analysis to the regression error.","section":"IV, Eq. (2)"},{"comment":"The disturbance model is internally inconsistent: the text introduces an 'unexpected biased disturbance' to simulate 'an unrealistic strong solar wind', but the next sentence states the disturbance is 'Gaussian with a zero mean and σ_q = 10^{-3}'. It is also not specified how the disturbance enters the propagation in Eqs. (5)-(6) (e.g., process noise on states, unmodeled acceleration, or measurement error). The robustness claims depend on this model, so please clarify and make the description consistent.","section":"VI"},{"comment":"The headline fuel-savings claim is based on single deterministic simulation runs without error bars or statistical comparison. The Monte Carlo study in Figure 14 is only for the proposed method, not for the baseline. Please report total ΔV statistics (mean, standard deviation, and histograms if feasible) for both the proposed method and the fixed-orbit baseline over the same set of initial states and disturbance realizations, so the claimed reduction can be assessed as a distribution rather than a single trajectory.","section":"VI, Figures 13 and 17"}],"minor_comments":[{"comment":"The internal section references are inconsistent: the results appear in Section 6 (Numerical Simulations) and Section 7 (NMPC-EKF), not 'Section 7 presents and discusses the results' as stated in the introduction.","section":"Introduction"},{"comment":"In the paragraph after Eq. (1), 'receptively' should be 'respectively', and 'in the CR3BP mode' should probably be 'in the CR3BP model'.","section":"II"},{"comment":"The caption reads 'the tow control strategies'; 'tow' should be 'two'.","section":"VI, Figure 16 caption"},{"comment":"In the paragraph discussing the variable-χ approach, the phrase 'it significantly increases computational complexity' is repeated; the duplicate should be removed.","section":"V"},{"comment":"The summation condition 'nχ+ncν+nsν ≤ N' should state explicitly that nχ, ncν, nsν are nonnegative integers, and the ranges of χ and ν should be given.","section":"IV, Eq. (2)"},{"comment":"The line 'X_k = X_0' is written as a constraint, but it is an initial condition; please rewrite it as 'given initial state X_0'.","section":"V, Eq. (4)"}],"recommendation":"major_revision","confidential_remarks":"The paper is a conference-style preprint with a promising idea but insufficient evidence for the central quantitative claim. The structural unfairness of the baseline comparison is the main obstacle; once the authors rerun the comparison with a properly constrained baseline and add regression validation and statistical reporting, the contribution may be suitable for publication. The manuscript also needs careful proofreading for typos and section-number consistency."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You asked for my take on the cislunar NMPC paper. The genuinely new thing is treating the reference trajectory as an optimization variable over an orbit family, with the family represented by a multivariate polynomial regression. That is a real extension of the usual track-one-orbit MPC schemes, and the paper does a decent job building the pipeline: PAC generates family members, the chi/nu parameterization is sensible, submanifold splitting is a reasonable way to keep local polynomial fits accurate, and the fixed-chi variant is a smart compromise between flexibility and computational cost. It is also tested on Lyapunov, halo, and NRHO families, includes a 500-run Monte Carlo for halos, and couples the controller to an EKF. That is more than many astrodynamics MPC papers do.\n\nThe main problem is the central quantitative claim. Because chi is an optimization variable in Eq. (3)-(4), the optimizer can always choose chi equal to the baseline orbit's parameter and reproduce the fixed-orbit control sequence. The proposed problem's optimal cost is no larger than the baseline cost by construction. The reported fuel reduction is therefore largely the value of relaxing the tracking target, not evidence that family tracking finds something better. The paper needs a fair comparison that holds the task fixed: either constrain chi to a well-chosen fixed orbit inside the same formulation, or report fuel consumption at equal achieved distance from the family. Without that, Figures 13 and 17 do not support \"significant reduction in fuel consumption.\"\n\nThe second soft spot is the regression model. It is load-bearing, because all reference states come from it, yet the paper reports no fitting error, no validation set, and no serious discussion of uniqueness and continuity beyond stating the assumption. A fit-error table should be standard here. Also, the simulation section calls the disturbance \"biased\" and then says it is Gaussian with zero mean; one of those is a typo, but it matters for reproducibility. And the baseline NMPC is underspecified: same horizons, weights, and disturbance realizations? Figures 13 and 17 appear to be single runs with no error bars.\n\nNone of this kills the idea. The family-tracking formulation is a legitimate contribution, and the fixed-chi-constant-along-the-horizon variant is practically motivated. The paper just overstates what the evidence shows. It is worth a serious referee: the concept deserves attention, and a good referee could push the comparison and the MPR validation into shape. I would send it to peer review with clear instructions to compare on equal terms and report regression accuracy. It may come back with a weaker fuel-savings story, but still useful as a formulation paper.","headline":"The core idea—optimizing the reference orbit within a fitted orbit family—is new and worth discussing, but the headline fuel-savings claim is partly built into the formulation and the baseline comparison needs rework.","tokens_in":10732,"tokens_out":3285,"would_cite":true,"duration_ms":43703,"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 finite-horizon controller that tracks an entire orbit family, not a fixed orbit, cuts fuel use in cislunar stationkeeping simulations.","keywords":["cislunar space","periodic orbit families","nonlinear model predictive control","Circular Restricted Three-Body Problem","libration points","orbit family tracking","multivariate polynomial regression","station-keeping"],"falsifier":"Hold out a second set of continuation-generated orbit members, refit Eq. (2) without them, and compute the maximum state-prediction error on the held-out set; if that error is comparable to the tracking-error weights in the cost function, the nominal reference is unreliable. A closed-loop test that starts the spacecraft on an orbit member outside the trained sub-manifold and checks whether it remains in the family would settle the claim directly.","tokens_in":9724,"feed_emoji":"🌙","tokens_out":11426,"duration_ms":117040,"temperature":0.7,"pith_summary":"The paper tries to establish that stationkeeping in cislunar space can use less fuel when the controller aims at an entire family of periodic orbits instead of a single predefined orbit. The proposed nonlinear model predictive controller keeps the orbit-family member an optimization variable, so at each horizon it can steer toward whichever nearby Lyapunov, halo, or near-rectilinear halo orbit is cheapest to reach. The family is encoded by a two-parameter polynomial regression fitted to continuation-generated orbit states, and the NMPC then treats that regression as the reference. Simulations with an extended Kalman filter and a biased disturbance show the spacecraft remaining inside the family while consuming less fuel than a conventional fixed-orbit tracker. This matters because future cislunar missions could spend the saved fuel on longer operations or larger margins.","feed_headline":"Tracking an orbit family cuts fuel in cislunar stationkeeping","feed_subtitle":"A finite-horizon controller chooses the cheapest nearby orbit and beats fixed-orbit tracking.","key_machinery":"The load-bearing object is the multivariate polynomial regression model of Eq. (2), which represents the state $\\hat X_i$ of any orbit-family member as a polynomial in $\\chi$ (which orbit in the family) and $\\nu$ (where along the orbit), with cosine and sine terms in $\\nu$ to handle periodicity. The model is fitted separately on sub-manifolds chosen so that the map is continuous and one-to-one over the region $R$. This compact two-parameter model is what lets the NMPC treat the reference as an implicit variable: the cost function in Eq. (3) is written against $P(\\chi_k,\\nu_{k+i})$, so the optimizer can move $\\chi_k$ within bounds to pick a cheaper nearby member of the same family. The paper also proposes a fixed-$\\chi$ variant, in which one family member is selected for the whole horizon, as the computationally efficient middle ground between full flexibility and rigid tracking.","core_discovery":"The central claim is that the reference trajectory for periodic-orbit tracking in the circular restricted three-body problem should be a decision variable, not a fixed input. The controller minimizes a finite-horizon cost that compares the propagated state with the regression model $P(\\chi_k,\\nu_{k+i})$ over the prediction and control horizons, subject to bounds on the family parameter $\\chi$, the along-orbit angle $\\nu$, and the velocity impulses. Because $\\chi$ can vary within the starting sub-manifold, the optimizer selects the orbit that is easiest to track while still keeping the spacecraft in the family. The paper reports that this formulation reduces fuel consumption compared with tracking a predefined reference orbit, across Lyapunov, halo, and near-rectilinear halo families near L1 and L2, including when an extended Kalman filter feeds estimated states to the controller.","pith_inferences":["Beyond the paper, the same reference-as-decision idea can be carried to quasi-periodic orbit families and tori, where a two-parameter regression becomes a three- or four-parameter model and the fuel savings would quantify how much flexibility the controller actually needs.","A natural extension is to hold out some continuation-generated orbit members from the regression training and measure the prediction error of Eq. (2) on them; that would show how much of the reported fuel saving depends on the fidelity of the fit.","If the savings persist under higher-fidelity dynamics such as ephemeris or four-body models, family tracking could become an operational stationkeeping strategy, with orbit-keeping tolerances replacing a single reference orbit."],"forward_implications":["A spacecraft can stay inside a chosen cislunar orbit family even when its initial state is poorly known and a biased disturbance acts on the dynamics.","The fixed-$\\chi$ family-tracking controller uses less control effort than tracking a single predefined orbit and roughly matches a fully variable reference at lower computational cost.","The same construction works for Lyapunov, halo, and near-rectilinear halo families near both L1 and L2, with prediction and control horizons tuned per family.","Coupling the NMPC with an extended Kalman filter keeps the estimated spacecraft within the orbit family, with the estimation error fluctuating around a stable mean."],"supporting_citations":[{"why":"Supplies the detailed PAC and differential-correction implementation used to generate the orbit-family members that train the regression model.","marker":"[27]"},{"why":"Provides the high-order differential correction scheme used to refine each periodic orbit before continuation.","marker":"[24]"},{"why":"Provides the nonlinear optimizer used to solve the non-convex NMPC problem at each control step.","marker":"[28]"},{"why":"Defines the Lyapunov orbit family, one of the three families the controller is demonstrated on.","marker":"[25]"},{"why":"Defines the halo and near-rectilinear halo families, the other two families in the numerical study.","marker":"[26]"}],"fun_headline_variants":["Cislunar fuel saver: let the controller pick the orbit","Finite-horizon control picks the cheapest orbit in cislunar space","Orbit-as-variable cuts fuel in cislunar tracking","Finite-horizon tracking: choose the orbit family to save fuel","Cislunar stationkeeping: let the controller roam the orbit family"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire fuel-saving result rests on the fitted polynomial model of each orbit family being accurate and one-to-one; if that fit is poor or the two parameters do not label every point cleanly, the controller is chasing wrong reference states and the reported savings do not carry over.","fun_headline_variants_meta":{"raw":{"variants":["Cislunar fuel saver: let the controller pick the orbit","Finite-horizon control picks the cheapest orbit in cislunar space","Orbit-as-variable cuts fuel in cislunar tracking","Finite-horizon tracking: choose the orbit family to save fuel","Cislunar stationkeeping: let the controller roam the orbit family"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00085,"raw_usage":{"total_tokens":3688,"prompt_tokens":930,"completion_tokens":2758,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":546,"completion_tokens_details":{"reasoning_tokens":2665}},"tokens_in":546,"tokens_out":2758,"duration_ms":22894,"temperature":1.0,"reasoning_tokens":2665,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T13:51:15.240814+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Hold out a second set of continuation-generated orbit members, refit Eq. (2) without them, and compute the maximum state-prediction error on the held-out set; if that error is comparable to the tracking-error weights in the cost function, the nominal reference is unreliable. A closed-loop test that starts the spacecraft on an orbit member outside the trained sub-manifold and checks whether it remains in the family would settle the claim directly.","supporting_citations":[{"cited_title":"Advances in Cislunar Periodic Solutions via Taylor Polynomial Maps","cited_arxiv_id":"2409.03692","evidence_quote":"Supplies the detailed PAC and differential-correction implementation used to generate the orbit-family members that train the regression model."},{"cited_title":"Dynamics near the three-body libration points via koopman operator theory,","cited_arxiv_id":null,"evidence_quote":"Provides the high-order differential correction scheme used to refine each periodic orbit before continuation."},{"cited_title":"Numerical exploration of the restricted problem, v,","cited_arxiv_id":null,"evidence_quote":"Defines the Lyapunov orbit family, one of the three families the controller is demonstrated on."},{"cited_title":"The ‘halo’family of 3-dimensional periodic orbits in the earth-moon restricted 3-body problem,","cited_arxiv_id":null,"evidence_quote":"Defines the halo and near-rectilinear halo families, the other two families in the numerical study."}],"review_version":1}