{"id":"bec66e0d-c1f6-4447-9685-902db3336b12","arxiv_id":"2501.14173","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"First indirect-method solutions to constrained 6DOF powered descent guidance, with closed-form gimbal-limited steering and an empirically validated Lagrange multiplier relation.","lead":"This paper computes fuel-optimal and time-optimal rocket landing trajectories in full six-degree-of-freedom dynamics using an indirect optimal control method, enforcing attitude, gimbal, and glideslope constraints through penalty functions and numerical continuation. It provides high-accuracy reference solutions that can be used to validate direct and convex guidance algorithms.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Final quaternion boundary condition in Table 2 is not unit-norm; the solved problem is not a valid 6DOF attitude maneuver, and 'no impact' is asserted without sensitivity test.","rationale":"The paper's central claim is the first indirect solution of the 6DOF powered descent guidance problem, with high-accuracy solutions satisfying the necessary conditions of optimality. That claim depends on the problem being a physically meaningful rigid-body landing problem. The final boundary condition q_f = [0, 0, 0.01, 1] has norm 1.00005, so it is not a unit quaternion; since quaternion norm is conserved by Eq. (2), the entire state history is a non-unit-quaternion trajectory, the DCM is not orthonormal, and the tilt-angle constraint is computed from a non-rotation matrix. The authors acknowledge the modification is made to avoid the singularity in Eq. (15) and assert 'no impact,' but they offer no derivation or numerical sensitivity study. This is the most load-bearing weakness because it is a concrete, internally checkable inconsistency rather than a matter of requiring a rigorous convergence proof: the penalty-method and Conjecture 1 limitations are explicitly acknowledged and could be accepted as numerical practice, whereas a non-unit final orientation is an outright mismatch between the stated problem and the solved problem. A single re-run with a normalized quaternion settles the issue: if the results change materially, the central numerical conclusions apply only to a modified problem; if they do not, the conditional acceptance can proceed with a clarifying remark. The reader's weakest assumption already noted the modified final boundary conditions without pinpointing the unit-norm violation, so agreement is partial; the proposed concern reinforces rather than overturns the reader's CONDITIONAL verdict.","tokens_in":25765,"tokens_out":16181,"duration_ms":149828,"concrete_test":"Re-solve the fuel-optimal case with the same parameters and continuation path, replacing q_f by the unit-norm quaternion [0, 0, 0.01, sqrt(1 - 0.01^2)] (or equivalently normalize [0, 0, 0.01, 1] to unit length). Compare final mass, time-of-flight, thrust-switch count, gimbal-switch intervals, and tilt-angle constraint activity against the reported values. If any of these change by more than the stated convergence tolerances, the 'no impact' assertion fails and the headline results are artifacts of the non-physical boundary condition; if they are unchanged, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Table 2 sets q_f = [0, 0, 0.01, 1], whose Euclidean norm is sqrt(1.0001) ≈ 1.00005, not 1. The quaternion kinematics in Eq. (2) conserve q^T q, so the entire trajectory is a non-unit quaternion history; the direction cosine matrix in Eq. (1) is then not orthonormal, and the tilt angle computed by Eq. (8) is not a physical rotation angle. The paper states this 'slight modification' has 'no impact' but provides no normalization, no sensitivity analysis, and no justification that the optimal switching structure and costate jumps are preserved under the modification. Because the central claim is a high-accuracy solution of the 6DOF PDG problem, a boundary condition that is not a valid rotation means the problem actually solved is not the stated rigid-body landing problem. The reported final mass, thrust-switch counts, and optimality checks are therefore for a non-physical problem unless unit-norm is restored or the no-impact assertion is substantiated.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a regularized indirect method for fuel- and time-optimal six-degree-of-freedom powered descent guidance (6DOF PDG) with state-only path inequality constraints on tilt angle, glideslope angle, and angular velocity magnitude, plus control constraints on thrust magnitude and gimbal angle. First-order necessary conditions are derived from the Hamiltonian, the thrust and steering controls are obtained in closed form under a gimbal-angle constraint, and the state path constraints are enforced through secant penalty functions embedded in a numerical continuation over smoothing parameters. The resulting multipoint boundary-value problems are solved with an indirect multiple-shooting scheme, and the solutions are compared with independent DIDO solutions and with the analytic Breakwell problem. The paper reports that the fuel-optimal trajectory has two thrust-magnitude switches and the time-optimal trajectory has three, and claims this is the first application of indirect methods to the full 6DOF PDG problem.","tokens_in":26075,"tokens_out":7381,"duration_ms":67330,"significance":"If the results hold, the paper makes a useful contribution: it demonstrates that a carefully regularized indirect method can produce high-accuracy extremal solutions for a constrained 6DOF rocket landing problem, with closed-form control expressions and an a posteriori recovery of state-constraint multipliers. The validation strategy is a clear strength: the indirect solutions are compared against independent DIDO solutions and against the analytic Breakwell benchmark, and the derivation of the necessary conditions is standard and transparent. However, the paper currently contains a load-bearing boundary-condition issue (non-unit final quaternion), an unsupported assertion that modified boundary conditions have no impact, and an unproven convergence claim for the secant penalty method. These issues must be resolved before the central quantitative claims can be accepted.","major_comments":[{"comment":"The final orientation boundary condition q_f = [0, 0, 0.01, 1]^T in Table 2 has Euclidean norm sqrt(1.0001), not 1. Since the quaternion kinematics in Eq. (2) conserve q^T q, the entire solved trajectory is a non-unit quaternion history. Consequently, the direction cosine matrix in Eq. (1) is not orthonormal, and the tilt angle computed from Eq. (8) is not the physical angle between the body and inertial vertical axes. This means the boundary-value problem actually solved is not the stated rigid-body landing problem, and the reported thrust-switch counts, final mass, and optimality checks pertain to that modified problem. The assertion that this slight modification has no impact is not substantiated; please re-solve with a unit-norm final quaternion (for example, by normalizing q_f or by imposing an explicit unit-norm constraint) and report whether the switching structure, objective values, and costate profiles change.","section":"Section VI, Table 2; Eqs. (1), (2), (8)"},{"comment":"The final boundary conditions are modified from Ref. [13]: r_z(t_f) is changed to 0.01 LU and q_f is changed to [0, 0, 0.01, 1]. The first modification is motivated by avoiding a glideslope-angle singularity, but no sensitivity analysis shows that the solution is representative of the original r_z(t_f) = 0 problem; the second modification is likewise asserted to have no impact. Please provide a convergence study as the perturbation sizes tend to zero, or solve the original boundary conditions with a robust formulation (for example, an atan2-based glideslope angle and a normalized quaternion), and quantify changes in the final mass, time of flight, and switching structure.","section":"Section VI, Table 2; Section IV, remark after Eq. (15)"},{"comment":"The formula for the approximate Lagrange multiplier is presented as an equality between the direct-adjoining Hamiltonian and the penalty Hamiltonian, but the algebra does not support the sign: with S_i ≤ 0 and rho_i sec(pi/2 P_i) ≥ 0, the term eta_i S_i in the direct adjoining approach is non-positive while the penalty term is non-negative, so the negative sign in Eq. (16) is imposed to satisfy complementarity rather than derived. Moreover, no asymptotic argument is given for the limit rho_i → 0; the ratio in Eq. (16) could in principle depend on how rho_i and S_i approach zero. Since this conjecture is one of the three stated contributions, it needs either a rigorous asymptotic derivation or a clear reframing as an empirical approximation with convergence evidence beyond the two test cases.","section":"Section IV, Conjecture 1, Eq. (16)"},{"comment":"The central claim that high-accuracy solutions satisfy the necessary conditions rests on the secant penalty functions in Eq. (12) converging to exact state-only path constraint satisfaction as rho_i → 0. The paper explicitly states that this is not rigorously shown, and the cited convergence results for interior penalty methods (Refs. 37 and 42) do not cover the singular secant penalty used here. Please provide a convergence argument adapted to this penalty, or at least a systematic error study (constraint violation, final mass, Hamiltonian constancy as functions of rho_i) to quantify how close the reported solutions are to the exact constrained optimum.","section":"Section V, Eqs. (12)-(13); Section VI"}],"minor_comments":[{"comment":"The text reads '6DOG PDG problems'; this should be '6DOF PDG problems'.","section":"Section VI, first paragraph"},{"comment":"The abstract states that an empirical relation is 'derived' for the Lagrange multipliers, but the paper itself presents the relation as a conjecture without rigorous proof; the wording should be softened to 'proposed and empirically validated'.","section":"Abstract and Section IV, Conjecture 1"},{"comment":"The discrepancy in lambda_omega_z is attributed to the roll degree of freedom being uncontrollable, but no verification is given that the offset is a gauge freedom. Please state explicitly whether the roll dynamics and costate equation are invariant under the observed offset, or provide a numerical check that the offset does not affect the other variables.","section":"Section VI.A, Fig. 10(d)"},{"comment":"No code or data availability statement is provided; for reproducibility, the continuation schedules, shooting-segment counts, and solver settings would be valuable as supplementary material.","section":"Section V and VI"}],"recommendation":"major_revision","confidential_remarks":"The non-unit final quaternion boundary condition is the main correctness concern; if normalizing q_f changes the solution materially, the paper's quantitative claims would not survive. The paper would be strengthened by a direct comparison of the penalty solution against a direct transcription that enforces quaternion normalization and the original boundary conditions. The empirical multiplier formula should also be labeled more cautiously. The novelty claim of being the first indirect method for 6DOF PDG should be checked carefully against the planar attitude-coupled work in Ref. [17] and related literature. Overall, the numerical machinery and validation approach are promising, but the boundary-condition validity and penalty convergence issues need to be resolved before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nQuick take: this is the first credible indirect-method treatment of 6DOF powered descent guidance with state-path constraints. The authors derive the necessary conditions, solve the resulting BVP with a multiple-shooting/continuation pipeline, and validate against DIDO plus the analytic Breakwell problem. That is real work and a real increment.\n\nThe new closed-form thrust-steering expression under the gimbal constraint (Eq. 30) is a nice piece of algebra. The empirical multiplier formula (Conjecture 1) is honestly labeled as a conjecture and tested against DIDO; it is plausible and useful, though not proven. The validation is genuinely independent: no part of the solution is fitted to DIDO's output. The thrust-switch counts and costate profiles match well. This is a solid paper, not a revolutionary one, and the authors are appropriately measured in their claims.\n\nSoft spots, in order of severity:\n\n1. The final quaternion boundary condition q_f=[0,0,0.01,1] is not unit norm (norm ≈ 1.00005). Since the quaternion kinematical equations preserve q^T q, the entire trajectory is non-unit. The DCM in Eq. (1) is then not orthonormal, and the tilt angle computed from Eq. (8) is not strictly a physical rotation angle. The paper says this 'slight modification' has 'no impact' but does not show a sensitivity test. This is a legitimate referee question. It is not fatal: the perturbation is tiny, and DIDO is run with the same boundary condition, so the comparison is consistent. But the authors should either normalize q_f (e.g., q_f=[0,0,0.01,sqrt(1-0.0001)] or an exact representation) and recompute, or provide a sensitivity analysis showing negligible change in switching structure and costate jumps. Right now the solved problem is technically a different, slightly non-physical problem.\n\n2. The secant penalty convergence is asserted with references to interior penalty proofs, but those proofs do not obviously cover the secant function. This matters because the whole method relies on the limit rho -> 0. The authors should at least discuss what conditions are needed and provide numerical evidence of convergence (e.g., a table of rho values vs. constraint violation).\n\n3. Conjecture 1 is empirically supported but not proven. That is acceptable if presented as a conjecture, but it should be clearly separated from the rest of the theory.\n\n4. No code or data. For reproducibility, sharing the shooting/continuation implementation would be a plus. The error assessment vs. DIDO is qualitative; a table of max differences would be more convincing.\n\nBottom line: the paper deserves serious peer review. It is a genuine first application of indirect methods to this problem class, the derivation is careful, and the validation approach is sound. The referee should ask for the quaternion issue to be resolved and for more rigorous or at least systematic treatment of the penalty convergence. Not a desk reject.\n\nRecommended action: send to a good aerospace or control journal (JGCD, Automatica, or similar), with a revision request addressing the quaternion normalization/sensitivity and penalty convergence discussion. I would cite this paper and would bring it to the reading group.","headline":"First credible indirect solutions to constrained 6DOF powered descent guidance, with a small but real boundary-condition normalization issue and some unproven penalty claims.","tokens_in":26538,"tokens_out":3173,"would_cite":true,"duration_ms":28902,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49K15","49M05","70Q05"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that a regularized indirect method can solve the constrained six-degree-of-freedom powered descent guidance problem for both minimum fuel and minimum time, producing extremal solutions with two and three thrust magnitude…","keywords":["powered descent guidance","six-degree-of-freedom dynamics","indirect optimization","state-only path constraints","interior penalty method","multiple shooting","fuel-optimal","time-optimal"],"falsifier":"Reduce the tilt-angle penalty weight well below $10^{-12}$ in the fuel-optimal case and check whether the initial tilt angle approaches exactly $90^\\circ$ while the Hamiltonian stays constant; if the constraint boundary is violated, the multiplier diverges, or the Hamiltonian drifts, then the penalty-family convergence that the method depends on does not hold.","tokens_in":25542,"feed_emoji":"🚀","tokens_out":9771,"duration_ms":81095,"temperature":0.7,"pith_summary":"The paper claims a regularized indirect optimization method can solve the constrained six-degree-of-freedom (6DOF) powered descent guidance problem in both fuel-optimal and time-optimal forms, with inequality constraints on thrust magnitude, gimbal angle, tilt angle, glideslope angle, and angular velocity. The payoff is that the method returns solutions satisfying the first-order necessary conditions of optimality, so the structure of the optimal control, in particular the number and timing of thrust-magnitude switches, comes out of the calculation rather than being guessed. For the studied parameters, the fuel-optimal solution has two thrust switches and the time-optimal solution has three, and the paper validates these against an independent pseudospectral solver. If the method holds, it gives a practical verification path for 6DOF landing guidance, a problem so far dominated by direct and convex methods and whose optimal control structure has been regarded as open.","feed_headline":"Indirect method solves 6DOF rocket landing guidance","feed_subtitle":"Both objectives satisfy optimality conditions, with two vs three thrust switches.","key_machinery":"The central machinery is a regularized indirect method assembled from four parts. First, control constraints are directly adjoined to the Hamiltonian, producing closed-form piecewise controls: the thrust magnitude $T^*$ through a switching function $S_T$, and the steering direction $\\hat{\\boldsymbol{\\alpha}}^*$ through a gimbal switching function $S_\\delta$. Second, state-only path inequality constraints are enforced with secant penalty functions $\\tilde S_i = \\sec(\\frac{\\pi}{2} P_i)$ whose weights $\\rho_i$ are driven toward zero by continuation, so a multipoint boundary-value problem becomes a smooth family of two-point boundary-value problems. Third, indirect multiple shooting with propagated sensitivity matrices keeps the numerically unstable integration tractable. Fourth, Conjecture 1 gives $\\tilde\\eta_i = -\\rho_i \\sec(\\frac{\\pi}{2}P_i)/S_i$ as an a posteriori approximation of the direct-adjoining Lagrange multiplier for each active constraint, allowing complementarity to be checked without solving the constrained problem directly.","core_discovery":"The central claim, stated on the paper's own terms, is that indirect methods can be applied to the full 6DOF powered descent guidance problem and yield high-accuracy extremal solutions without prior knowledge of the active-constraint sequence. The paper derives closed-form expressions for the thrust magnitude and the thrust-steering direction under the gimbal-angle constraint, enforces the three state-only path inequality constraints (SOPICs) with secant penalty functions, and solves the resulting boundary-value problems with an indirect multiple-shooting method and numerical continuation. It also proposes an empirical relation, Conjecture 1, that recovers the direct-adjoining Lagrange multipliers of the active path constraints from penalty solutions, and it validates the entire approach by comparing states, controls, costates, and multipliers against an independent pseudospectral solver and against a benchmark problem with an analytic solution.","pith_inferences":["Beyond the paper, if the penalty-family convergence is rigorously established, the approach would give verification-quality extremal solutions for other 6DOF guidance problems, including those with state-triggered or attitude-dependent constraints, without the need for convexification.","The paper's own observation that hundreds of shooting segments are needed when path constraints are active suggests that an adaptive or mesh-refined segment placement would materially reduce cost; implementing such a scheme is a natural testable extension.","Because the model leaves roll uncontrollable, the observed constant offset in the angular-velocity costate points to a gauge freedom; fixing a roll-costate condition would make the multiplier comparison with direct methods fully one-to-one.","The empirical multiplier relation, if proven, would close the gap between penalty methods and exact constrained necessary conditions; a first test is to apply it to a problem with a third-order state constraint, where the impulse structure differs."],"forward_implications":["For the studied parameter set, the fuel-optimal thrust magnitude is bang-bang with exactly two switches and the time-optimal with three, giving a concrete control structure that direct methods can be checked against.","The closed-form gimbal-constrained steering expression removes the steering control from the set of numerical unknowns, shrinking the boundary-value problem and removing a source of discretization error.","The secant-penalty continuation solves the state-only path constraints without requiring a priori knowledge of which constraints are active or when, so the method applies to problems where the active-constraint sequence is unknown.","The a posteriori multiplier recovery from Conjecture 1 lets practitioners verify complementarity slackness and detect whether an active constraint has been missed.","The same machinery transfers to simpler constrained optimal control problems, as demonstrated on the benchmark problem with an analytic solution."],"supporting_citations":[{"why":"Supplies the 6DOF dynamics and attitude constraint formulation the paper adapts.","marker":"[10]"},{"why":"Supplies the problem parameters, boundary conditions, and the convex baseline solution used for comparison.","marker":"[13]"},{"why":"Introduces the secant-penalty and trigonometrization machinery for constrained optimal control problems.","marker":"[32]"},{"why":"Provides the direct-adjoining necessary conditions and maximum-principle framework for state constraints.","marker":"[33]"},{"why":"Demonstrates interior penalty methods for constrained optimal control, the convergence context for the secant penalties.","marker":"[37]"},{"why":"Provides the Hamiltonian invariance principle on which the approximate multiplier formula is built.","marker":"[39]"},{"why":"Independent pseudospectral solver used to validate states, controls, costates, and multipliers.","marker":"[40]"},{"why":"Textbook source for the necessary conditions, transversality, and the benchmark problem with analytic solution.","marker":"[41]"},{"why":"Formulates the direct-adjoining multipliers and impulse behavior for state constraints used in the conjecture's justification.","marker":"[45]"},{"why":"Supplies the L2-norm regularization of bang-bang controls used for thrust and gimbal switching.","marker":"[51]"}],"fun_headline_variants":["Indirect method solves constrained 6DOF rocket landing","Penalty-embedded indirect guide cracks 6DOF powered descent","No sequence guess: indirect solves 6DOF rocket landing","Indirect optimization unlocks constrained 6DOF powered descent","Fuel-time optimal 6DOF landing via regularized indirect method"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the secant penalty solutions converge to exact path-constraint satisfaction as the penalty weights go to zero, and that the empirical multiplier formula recovers the true direct-adjoining multipliers; the paper explicitly says neither is rigorously proven.","fun_headline_variants_meta":{"raw":{"variants":["Indirect method solves constrained 6DOF rocket landing","Penalty-embedded indirect guide cracks 6DOF powered descent","No sequence guess: indirect solves 6DOF rocket landing","Indirect optimization unlocks constrained 6DOF powered descent","Fuel-time optimal 6DOF landing via regularized indirect method"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000623,"raw_usage":{"total_tokens":2883,"prompt_tokens":937,"completion_tokens":1946,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":553,"completion_tokens_details":{"reasoning_tokens":1860}},"tokens_in":553,"tokens_out":1946,"duration_ms":13389,"temperature":1.0,"reasoning_tokens":1860,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T15:19:25.471973+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Reduce the tilt-angle penalty weight well below $10^{-12}$ in the fuel-optimal case and check whether the initial tilt angle approaches exactly $90^\\circ$ while the Hamiltonian stays constant; if the constraint boundary is violated, the multiplier diverges, or the Hamiltonian drifts, then the penalty-family convergence that the method depends on does not hold.","supporting_citations":[{"cited_title":"Successive Convexification for Fuel-Optimal Powered Landing with Aerodynamic Drag and Non-Convex Constraints,","cited_arxiv_id":null,"evidence_quote":"Supplies the 6DOF dynamics and attitude constraint formulation the paper adapts."},{"cited_title":"Six-Degree-of-Freedom Rocket Landing Optimization via Augmented Convex–Concave Decomposition,","cited_arxiv_id":null,"evidence_quote":"Supplies the problem parameters, boundary conditions, and the convex baseline solution used for comparison."},{"cited_title":"Generalized Vectorized Trigonometric Regularization for Solving Optimal Control Problems with Complex Solution Structures,","cited_arxiv_id":null,"evidence_quote":"Introduces the secant-penalty and trigonometrization machinery for constrained optimal control problems."},{"cited_title":"Fuel Optimization for Continuous-Thrust Orbital Rendezvous with Collision Avoidance Constraint,","cited_arxiv_id":null,"evidence_quote":"Demonstrates interior penalty methods for constrained optimal control, the convergence context for the secant penalties."},{"cited_title":"E., and Ho, Y.-C.,Applied optimal control: optimization, estimation, and control, rev","cited_arxiv_id":null,"evidence_quote":"Textbook source for the necessary conditions, transversality, and the benchmark problem with analytic solution."},{"cited_title":"M.,A primer on Pontryagin’s principle in optimal control, second edition ed., Collegiate Publisher, San Francisco,","cited_arxiv_id":null,"evidence_quote":"Formulates the direct-adjoining multipliers and impulse behavior for state constraints used in the conjecture's justification."}],"review_version":1}