{"id":"03fe1860-51b5-4af8-b70b-bf23633b471c","arxiv_id":"2607.28353","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":5,"one_line_summary":"A precomputed ISS-CLF/robust-CBF pair yields a real-time QP that approximates robust NMPC with proven ISS and constraint satisfaction for nonlinear systems.","lead":"The paper gives a tiny quadratic program that approximates robust nonlinear MPC using a precomputed ISS-CLF and robust CBF pair, with proofs of robust constraint satisfaction and input-to-state stability. It is aimed at real-time spacecraft control on limited hardware where full robust NMPC is too slow.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified","rationale":"The reader correctly isolates Theorem 1 as the strongest claim and correctly flags certificate synthesis as the practical bottleneck rather than a flaw in the argument. Because the theorem is only an implication from certificates to closed-loop properties, and those certificates are exhibited for the numerical plants, there is no load-bearing technical concern that would move the verdict away from ACCEPT. The suggested concrete test simply reconfirms the already-reported certificates; it is a verification step, not a potential falsifier of the theory.","tokens_in":17817,"tokens_out":412,"duration_ms":7689,"concrete_test":"Re-run the nonconvex SOS program (17) for the 6-D attitude system of §5.2 with the same monomial basis and disturbance bound; verify that the returned (V̂, ĥ) still satisfy the SDP-relaxed inequalities (17b)–(17g) to numerical tolerance and that closed-loop trajectories under the resulting QP remain inside ĥ ≤ eta for the same 20 initial conditions. Success leaves the theorem and the examples intact.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (Theorem 1) is a short, standard certificate argument: if a compatible pair (V̂, ĥ) meets Definition 2, then the QP (15) inherits robust forward invariance of the sublevel set and the ISS-Lyapunov decrease. The proof is direct from the worst-case bound on the disturbance term and the fact that the QP minimizes the same dissipation expression that the terminal controller already satisfies. No internal inconsistency, missing hypothesis, or gap in the implication appears. The reader’s weakest assumption (existence and usefulness of modest-degree polynomial certificates via nonconvex SOS) is a genuine practical limitation, openly discussed in §5.3, but it is not load-bearing for the correctness of the theorem itself; the two spacecraft examples already supply concrete certificates that make the claim hold on those plants.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper proposes ISS-∂MPC, a robust approximate nonlinear MPC law obtained from a compatible pair of an ISS control Lyapunov function and a robust control barrier function. Building on a prior nominal infinitesimal-horizon scheme, the online law reduces to a small quadratic program (15) that enforces a worst-case CBF inequality and minimizes a CLF dissipation rate plus stage cost. Theorem 1 states that if the pair satisfies the robust terminal conditions of Definition 2, the closed loop is robustly feasible on the approximate safe set and input-to-state stable, recovering asymptotic stability when disturbances vanish. A nonconvex sum-of-squares program (17) is given for polynomial synthesis. Two constrained spacecraft examples (rate damping and large-angle attitude control) compare the method to zoRo-RTI, the synthesized polynomial feedback, and a standard rate controller, reporting sub-millisecond solve times and constraint satisfaction.","tokens_in":18052,"tokens_out":1271,"duration_ms":41944,"significance":"Real-time robust NMPC under tight compute budgets remains a genuine bottleneck; reducing the online problem to a small QP while retaining ISS and robust constraint certificates is a useful contribution. The argument that minimizing only the current dissipation (rather than a min-max over future disturbance signals) yields ISS rather than merely practical stability is cleanly made and distinguishes the scheme from classical min-max MPC. Supplementary code and synthesis details support reproducibility. The practical limitation—that useful low-degree polynomial certificates must exist and are equilibrium-specific—is stated openly in §5.3 and does not undermine the certificate theorem itself. If the result holds as claimed, the method is a credible option for embedded nonlinear constrained control.","major_comments":[{"comment":"Theorem 1 and Definition 2 are stated for the continuous-time closed loop under the feedback κ̂_∂. The implemented controller is sampled-data with zero-order hold (§5.1.1, 10 Hz; §5.3). The paper notes that inter-sample effects can destroy invariance (citing Breeden et al., 2022) but only reports that violations were not observed numerically. For the load-bearing claim of robust constraint satisfaction and ISS under the actual law, either a sampled-data certificate (e.g., a tightened CBF condition or maximum sampling period) or an explicit restriction of the guarantees to the continuous-time idealization should be added; the current gap sits between the theorem and the reported experiments.","section":"Theorem 1, §5.3"},{"comment":"In the reduction from (14) to (15) and in the proof of Theorem 1, the disturbance effect is bounded by an operator-norm term δ(x). The proof writes ∇ĥ(x)p(x)w ≤ ||∇ĥ(x)p(x)||_2 w̄, while (13) allows a general p-norm and the text defines δ via the induced operator norm. The argument is correct for p=2 (the case used in SOS and in the examples), but the write-up should state the norm consistently and note that the QP form (15) inherits the same p as the bound; otherwise the feasibility claim is not fully aligned with the general disturbance set (13).","section":"§3.2, Eq. (13)–(15), Theorem 1 proof"}],"minor_comments":[{"comment":"Abstract and title use “Realtime”; the body mixes “real-time” and “realtime”. Prefer a single spelling (real-time).","section":"Title, Abstract"},{"comment":"Eq. (17e) uses multiplier (s2−a) and class-K gain a in γ(r)=ar; a short sentence that a>0 is fixed (or optimized) would clarify the decision variables of (17).","section":"§4.2, Eq. (17)"},{"comment":"Table 1 “RMS ratio” is defined only in the text (§5.1.2) as RMS/RMS_ISS-∂MPC on [100,250] s; add the definition to the table caption.","section":"Table 1"},{"comment":"Figure 1 y-axis is log-scale RMS; state units (deg/s) in the caption. Figures 2–5 would benefit from stating that all 20 runs are overlaid.","section":"§5 Figures"},{"comment":"Reference ApS (2026) MOSEK URL is still “YYY”; replace with the working link or drop the placeholder.","section":"References"},{"comment":"Typos / wording: “constraint systems” → “constrained systems” (Abstract); “augment a recently introduced” spacing; “theclosed-loop” and similar missing spaces appear in the proof paragraph of Theorem 1 and in §5.3.","section":"Abstract, Theorem 1, §5.3"},{"comment":"The comparison baseline zoRo-RTI uses a 20 s horizon chosen “heuristically” so that the first step is feasible. A one-sentence sensitivity note (or a second horizon) would strengthen the fairness claim in Table 1.","section":"§5.1.1"}],"recommendation":"minor_revision","confidential_remarks":"The central certificate (Theorem 1) is standard once Definition 2 is granted and matches the reader/skeptic assessment: no internal contradiction. The work is a direct robust extension of the authors’ own nominal ∂MPC preprint; novelty is incremental but legitimate for a applications/methods journal in eess.SY. Fit is appropriate. I would not block on the SOS practicality limitation, which is already disclosed. Minor revision on the sampled-data gap and norm consistency is enough."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"Punchline: this is a solid, incremental method paper that turns their nominal infinitesimal-horizon NMPC into a robust ISS version without blowing up the online cost. Theorem 1 is the real payload and it holds.\n\nWhat is new is the combination, not any single ingredient. They take a compatible ISS-CLF / robust-CBF pair that meets their robust terminal conditions (Definition 2), fold the worst-case disturbance into an operator-norm bound δ(x), and get a small QP that minimizes stage cost plus the CLF decrease subject only to the robust CBF inequality. No slack variable, stage cost kept, and they recover asymptotic stability when w vanishes—stronger than the ISpS you usually get from min-max MPC. The SOS program (17) that synthesizes the pair is concrete and they ship the comparison setup. Online times on the spacecraft examples are orders of magnitude below zoRo-RTI while matching stage cost and beating RMS; that is the practical point.\n\nSoft spots are real but openly stated and not load-bearing for the theorem. Existence of a useful modest-degree polynomial certificate is plant-dependent; the feasible set is only for one equilibrium; inter-sample effects are waved at; the SOS is nonconvex. None of that breaks the implication “if the pair exists, the QP gives robust invariance + ISS.” The two numerical plants already supply such pairs, so the claim is demonstrated where it matters for the paper’s scope. Citation pattern is normal for the subfield; self-citation of the nominal precursor is necessary and not circular.\n\nThis is for people who need embedded robust NMPC on nonlinear plants and are willing to do offline SOS work. Math and experiments look honest. I would send it to referees; it deserves a careful read, not a desk reject. Engage if you care about real-time robust nonlinear control; skip if you only want tube or system-level synthesis advances.","headline":"Clean, usable extension of their own infinitesimal-horizon MPC to true ISS with a tiny online QP; the theorem is short and correct, the practical value hangs on getting good polynomial certificates.","tokens_in":18668,"tokens_out":491,"would_cite":true,"duration_ms":16106,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"A tiny quadratic program from an ISS Lyapunov function and a robust barrier function delivers real-time robust nonlinear MPC with input-to-state stability.","keywords":["Robust Model Predictive Control","Input-to-state Stability","Control Barrier Function","Control Lyapunov Function","Sum-of-squares Optimization","Infinitesimal-horizon MPC","Real-time feasibility","Spacecraft attitude control"],"falsifier":"On the spacecraft rate-damping or large-angle slew examples, either the online QP becomes infeasible, a closed-loop trajectory exits the claimed safe set under a disturbance inside the design bound, or the Lyapunov function fails to decrease to a neighborhood sized by that bound.","tokens_in":18708,"feed_emoji":"🚀","tokens_out":915,"duration_ms":24994,"temperature":0.7,"pith_summary":"The paper shows that robust nonlinear model predictive control need not solve a long-horizon min-max problem online. Instead, one precomputes a compatible pair of an input-to-state control Lyapunov function and a robust control barrier function, then evaluates a single small quadratic program at each sample. That program enforces safety against bounded disturbances and drives the state toward a neighborhood of the origin whose size shrinks with the disturbance; when disturbances vanish, ordinary asymptotic stability is recovered. The same guarantees that classical robust terminal conditions give for min-max MPC are obtained at the computational cost of a few-millisecond QP, small enough for embedded spacecraft hardware. Numerical comparisons on constrained attitude and rate control confirm that the scheme stays feasible, respects torque and rate limits, and runs orders of magnitude faster than a contemporary robust real-time-iteration method while matching its closed-loop cost.","feed_headline":"Tiny QP gives real-time robust nonlinear MPC with ISS","feed_subtitle":"Precomputed Lyapunov and barrier certificates turn safety and stability into one embedded quadratic program","key_machinery":"The ISS infinitesimal-horizon MPC quadratic program: at each measured state minimize the quadratic stage cost plus the directional derivative of the ISS-CLF, subject to a single robust CBF inequality that already accounts for the worst-case disturbance magnitude. Compatibility of the CLF/CBF pair supplies both feasibility and the ISS decrease.","core_discovery":"If a pair of functions (approximate value function and approximate feasible-set barrier) satisfies the robust stabilizing terminal conditions—an ISS control Lyapunov inequality together with a robust control-barrier inequality—then the feedback obtained by minimizing the stage cost plus the Lie derivative of the Lyapunov function subject only to the barrier inequality renders the approximate safe set forward-invariant under every admissible disturbance and makes the Lyapunov function an ISS-Lyapunov function for the closed loop. Consequently the controller is defined for all future time, state constraints are never violated, and the origin is input-to-state stable; asymptotic stability is re","pith_inferences":["Reference-dependent certificates would be the natural next step to handle changing set-points without re-solving the full SOS program for every equilibrium.","The same infinitesimal-horizon reduction should apply to other certificate pairs (e.g., contraction metrics plus barriers) whenever the certificates already encode robustness.","Inter-sample invariance under zero-order hold remains an open practical gap; a sampled-data barrier correction could close it without changing the online QP size."],"forward_implications":["Embedded platforms that cannot run multi-step robust NMPC can still obtain certified robust constraint satisfaction and ISS by solving one small QP per sample.","When disturbances disappear the same controller recovers ordinary asymptotic stability, unlike min-max schemes that only guarantee practical stability.","The offline SOS synthesis produces both the Lyapunov and barrier certificates and a polynomial fallback law that can be used without any online optimization.","Because only the present disturbance bound enters the QP, the method can later be combined with a disturbance observer to reduce conservatism without enlarging the online problem."],"fun_headline_variants":["Small QP yields real-time ISS-stable robust nonlinear MPC","ISS-CLF and CBF pair makes approximate NMPC embedded-feasible","Tiny quadratic programs give robust NMPC with proven ISS","Certificates reduce robust nonlinear MPC to one real-time QP","Approximate NMPC stays safe and ISS via barrier-Lyapunov QP"],"cache_read_input_tokens":128,"weakest_assumption_plain":"A compatible polynomial ISS-Lyapunov function and robust barrier of modest degree can be found offline by sum-of-squares optimization and still leave a usefully large safe set around the chosen equilibrium.","fun_headline_variants_meta":{"raw":{"variants":["Small QP yields real-time ISS-stable robust nonlinear MPC","ISS-CLF and CBF pair makes approximate NMPC embedded-feasible","Tiny quadratic programs give robust NMPC with proven ISS","Certificates reduce robust nonlinear MPC to one real-time QP","Approximate NMPC stays safe and ISS via barrier-Lyapunov QP"]},"model":"grok-4.5","effort":"low","cost_usd":0.004026,"raw_usage":{"total_tokens":1154,"prompt_tokens":670,"num_sources_used":0,"completion_tokens":74,"cost_in_usd_ticks":40264000,"prompt_tokens_details":{"text_tokens":670,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":410,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":670,"tokens_out":74,"duration_ms":7834,"temperature":1.0,"reasoning_tokens":410,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T10:22:59.413919+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"On the spacecraft rate-damping or large-angle slew examples, either the online QP becomes infeasible, a closed-loop trajectory exits the claimed safe set under a disturbance inside the design bound, or the Lyapunov function fails to decrease to a neighborhood sized by that bound.","supporting_citations":[],"review_version":1}