{"id":"a40c2ada-01a3-4017-a97f-9a292681690f","arxiv_id":"2606.24316","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":1,"one_line_summary":"A set-membership data-driven min-max MPC approach is developed for unknown nonlinear systems, yielding recursive feasibility and stability guarantees via Lyapunov SDPs for noise-free and disturbed measurements.","lead":"The paper proposes a data-driven robust min-max MPC scheme for unknown nonlinear systems with disturbances, representing dynamics via basis functions and using set-membership from noisy data to derive stabilizing controllers via SDPs. If validated, this could enable safer control of complex real-world systems where models are unavailable or uncertain.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Stability and feasibility proofs require exact linear parametrization of the unknown nonlinear dynamics via a finite dictionary of basis functions.","rationale":"The reader's weakest assumption is precisely the load-bearing step; the concern is internal to the argument rather than external consensus. Because the full derivations are unavailable in the supplied abstract, the verdict remains conditional on that assumption holding exactly.","tokens_in":1696,"tokens_out":335,"duration_ms":18997,"concrete_test":"In the system-representation section, locate the equation that writes the dynamics as an equivalent linear form; re-derive the set-membership bounds assuming only that the residual between true f and the span of the basis is bounded by some \rho>0; check whether the recursive-feasibility and Lyapunov arguments still close when this residual term is propagated through the min-max cost and the SDP.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The strongest claim (recursive feasibility plus exponential/robust stability) is derived from a min-max MPC formulation whose uncertainty set is obtained by set-membership identification on the matrices of the equivalent linear form. This form is obtained only when the true vector field lies exactly in the span of the chosen basis functions; the set-membership description then correctly contains the true matrices. If the dictionary is incomplete, the true dynamics lie outside every matrix set consistent with the data, so the robust MPC problem solved online is not a valid over-approximation and the Lyapunov-based SDP certificates no longer apply to the real closed-loop system. The paper states the representation step without supplying verifiable conditions on the dictionary or a procedure to certify exactness from data.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper proposes a data-driven robust min-max MPC scheme for unknown nonlinear systems with process disturbances. It represents the dynamics in an equivalent linear form via a chosen dictionary of basis functions, derives a set-membership uncertainty description of the unknown matrices from noisy input-state data, formulates an online min-max MPC problem, and supplies Lyapunov-based SDP solutions for a stabilizing state-feedback controller in two measurement scenarios. The schemes are claimed to ensure recursive feasibility together with exponential stability (noise-free case) or robust stability (disturbed case).","tokens_in":1864,"tokens_out":553,"duration_ms":14554,"significance":"If the exact linear parametrization holds and the SDP certificates are valid, the approach would provide a tractable data-driven route to robust MPC for nonlinear systems under disturbances, extending existing linear data-driven methods while retaining stability guarantees via set-membership learning. The use of SDP-based controllers and explicit feasibility/stability proofs would be a concrete strength if the representation assumption can be certified.","major_comments":[{"comment":"Abstract and model-representation section: the central claim that the unknown nonlinear dynamics admit an 'equivalent linear form' with matrices fully characterized by set-membership sets derived from data requires that the true vector field lies exactly in the span of the chosen basis functions. No verifiable conditions on the dictionary, no procedure to certify exactness from data, and no discussion of the consequences when the assumption fails are supplied; without these the robust MPC problem is not guaranteed to be a valid over-approximation and the Lyapunov SDP certificates do not apply to the real closed-loop system.","section":"Abstract / model representation"},{"comment":"Stability and recursive-feasibility claims (abstract): these rest on the min-max MPC formulation whose uncertainty set is obtained under the exact-representation assumption. When the dictionary is incomplete the true dynamics lie outside every matrix set consistent with the data, so the derived SDP certificates and feasibility arguments no longer bound the actual nonlinear closed-loop behavior; the paper provides no sensitivity analysis or fallback when this occurs.","section":"Abstract / stability analysis"}],"minor_comments":[{"comment":"Notation for the basis-function dictionary and the resulting matrix sets should be introduced with explicit dimensions and an example in the main text rather than only in the appendix.","section":null},{"comment":"The two online scenarios (noise-free vs. disturbed measurements) are distinguished clearly in the abstract but the corresponding SDP formulations would benefit from a side-by-side comparison table of the decision variables and constraints.","section":null}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive comments on our manuscript. We address each major comment below and propose revisions where appropriate to clarify the scope and limitations of the approach.","responses":[{"response":"We acknowledge that the proposed framework assumes the nonlinear dynamics admit an exact linear parametrization in the chosen dictionary of basis functions. This assumption is standard in approximation-based nonlinear control but is indeed central to the validity of the uncertainty sets and subsequent guarantees. In the revised manuscript, we will add explicit discussion in the model representation section stating the assumption, noting that no general finite-data certification procedure is provided (as verifying exact span membership from noisy data alone is generally intractable without additional system knowledge), and outlining the consequences of violation (the true dynamics may lie outside the computed sets, rendering the robust MPC and SDP certificates inapplicable to the actual system).","revision_made":"yes","referee_comment":"[Abstract / model representation] Abstract and model-representation section: the central claim that the unknown nonlinear dynamics admit an 'equivalent linear form' with matrices fully characterized by set-membership sets derived from data requires that the true vector field lies exactly in the span of the chosen basis functions. No verifiable conditions on the dictionary, no procedure to certify exactness from data, and no discussion of the consequences when the assumption fails are supplied; without these the robust MPC problem is not guaranteed to be a valid over-approximation and the Lyapunov SDP certificates do not apply to the real closed-loop system."},{"response":"The stability and recursive feasibility results are derived under the exact-representation assumption, as stated in the problem formulation and analysis sections. We agree that no sensitivity analysis or fallback is provided for dictionary incompleteness. In revision, we will insert a remark in the stability analysis section explicitly conditioning the guarantees on the assumption and recommending dictionary selection via domain knowledge. Developing a general sensitivity analysis or fallback would require quantifying representation error, which lies outside the current set-membership framework focused on exact parametrization.","revision_made":"partial","referee_comment":"[Abstract / stability analysis] Stability and recursive-feasibility claims (abstract): these rest on the min-max MPC formulation whose uncertainty set is obtained under the exact-representation assumption. When the dictionary is incomplete the true dynamics lie outside every matrix set consistent with the data, so the derived SDP certificates and feasibility arguments no longer bound the actual nonlinear closed-loop behavior; the paper provides no sensitivity analysis or fallback when this occurs."}],"tokens_in":1433,"tokens_out":553,"duration_ms":23456,"standing_objections":["A general, verifiable procedure to certify exact dictionary representation from finite noisy data without additional prior knowledge of the system dynamics."]},"desk_editor":{"model":"grok-4.3","letter":"The core contribution is a data-driven robust MPC that turns noisy input-state samples into a set-membership description of unknown matrices after rewriting the nonlinear dynamics in an equivalent linear form via basis vector fields. It then solves a min-max problem with Lyapunov SDPs to get a stabilizing feedback, handling both noise-free and disturbed state measurements, and claims recursive feasibility plus exponential or robust stability.\n\nThis combination for nonlinear plants with disturbances looks new relative to the linear-system literature they cite. The two online scenarios and the SDP-based controller synthesis are laid out clearly enough to follow.\n\nThe load-bearing assumption is that the true vector field is exactly captured by the chosen dictionary; otherwise the matrix sets derived from the data do not contain the real system and the robust guarantees do not transfer. The abstract states the representation step without giving verifiable conditions on the dictionary or a data-driven test for exactness. That is the main soft spot.\n\nThe work is aimed at control researchers already using set-membership or data-driven MPC. It is worth sending to referees because the gap it targets is real and the technical route is concrete, even though the representation assumption will need explicit discussion and possibly weakening or certification in revision.","headline":"The paper gives a set-membership min-max MPC scheme for nonlinear systems but its stability and feasibility claims only hold if the unknown dynamics lie exactly in the span of a pre-chosen finite basis dictionary.","tokens_in":2331,"tokens_out":326,"would_cite":false,"duration_ms":19063,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Unknown nonlinear systems can be robustly controlled with data-driven min-max MPC that learns set-membership uncertainty bounds from noisy input-state data.","keywords":["data-driven MPC","robust control","nonlinear systems","set-membership learning","min-max optimization","Lyapunov stability","semidefinite programming","closed-loop stability"],"falsifier":"A simulation or experiment in which the true dynamics lie outside the computed set-membership bounds, causing the closed-loop trajectory to violate recursive feasibility or lose stability despite the SDP conditions being satisfied.","tokens_in":2618,"feed_emoji":"⚙️","tokens_out":655,"duration_ms":22704,"temperature":0.7,"pith_summary":"The paper develops a min-max model predictive control scheme for nonlinear systems whose dynamics are unknown but can be rewritten in equivalent linear form using a dictionary of basis functions. Noisy measurements are used to construct a set-membership description of the unknown matrices, which then defines the uncertainty set inside a min-max optimization. A Lyapunov-based semidefinite program computes a stabilizing state-feedback gain for two cases: noise-free state measurements and measurements affected by process disturbance. The resulting controller is shown to keep the optimization feasible at every step while delivering exponential stability without disturbances or robust stability with them.","feed_headline":"Data-driven min-max MPC stabilizes unknown nonlinear systems","feed_subtitle":"Set-membership bounds from noisy data deliver recursive feasibility plus exponential or robust stability.","key_machinery":"The set-membership description of the unknown system matrices obtained from noisy data, which supplies the uncertainty set for the min-max MPC problem solved by a Lyapunov-based semidefinite program.","core_discovery":"By representing the unknown nonlinear dynamics in an equivalent linear form whose matrices are characterized by a set-membership set derived from noisy input-state data, a min-max MPC problem can be solved via a Lyapunov-based semidefinite program to obtain a state-feedback controller that guarantees recursive feasibility of the optimization and either exponential or robust closed-loop stability depending on the presence of process disturbances.","pith_inferences":["The same set-membership construction could be used to incorporate partial prior knowledge by fixing some matrix entries while learning the rest.","Online SDP solves may restrict the method to systems whose sampling period allows sufficient computation time unless warm-starting or faster solvers are added.","The approach suggests a route to adaptive control by periodically refreshing the set-membership description as new data arrive."],"forward_implications":["The optimization remains recursively feasible at every time step.","Exponential stability of the closed loop is obtained when process disturbances are absent.","Robust stability of the closed loop is obtained when process disturbances are present.","The controller exhibits competitive performance relative to existing data-driven and model-based methods on benchmark examples."],"fun_headline_variants":["Robust min-max MPC for nonlinear systems via set-membership","Data-driven set-membership stabilizes unknown nonlinear systems","Min-max MPC guarantees stability from noisy nonlinear data","SDP computes robust controllers for data-driven nonlinear MPC"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The unknown nonlinear dynamics can be exactly represented in an equivalent linear form using vector fields built from a chosen dictionary of basis functions, allowing the unknown matrices to be fully characterized by a set-membership description derived from noisy input-state data.","fun_headline_variants_meta":{"raw":{"variants":["Robust min-max MPC for nonlinear systems via set-membership","Data-driven set-membership stabilizes unknown nonlinear systems","Min-max MPC guarantees stability from noisy nonlinear data","SDP computes robust controllers for data-driven nonlinear MPC"]},"model":"grok-4.3","cost_usd":0.005801,"raw_usage":{"total_tokens":2752,"prompt_tokens":648,"num_sources_used":0,"completion_tokens":59,"cost_in_usd_ticks":58012000,"prompt_tokens_details":{"text_tokens":648,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2045,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":648,"tokens_out":59,"duration_ms":13212,"temperature":1.0,"reasoning_tokens":2045,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-25T23:07:34.026147+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A simulation or experiment in which the true dynamics lie outside the computed set-membership bounds, causing the closed-loop trajectory to violate recursive feasibility or lose stability despite the SDP conditions being satisfied.","supporting_citations":[],"review_version":1}