{"id":"b41f5277-026d-4530-aa4a-ffc42bf763d4","arxiv_id":"2607.04213","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":3,"one_line_summary":"Tube-based robust receding-horizon games guarantee recursive feasibility under additive disturbances and drive each agent’s nominal state to a steady-state variational GNE, with actual states entering an mRPI neighborhood.","lead":"This paper gives multi-agent controllers a way to keep competing under shared limits even when each agent’s dynamics are hit by unknown bounded noise. It matters for fleets, grids, and robots that must stay safe without a central planner or full model sharing.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified","rationale":"The Reader correctly isolates the tracking-cost hypothesis as the weakest (and openly acknowledged) assumption and correctly judges that it does not undermine the theorems inside the stated scope. My own pass found no deeper load-bearing flaw: the tube tightening, privacy-preserving scalar broadcast, decoupled terminal sets, and potential Lyapunov argument are standard pieces assembled carefully, with every constraint of the shifted candidate checked. The absence of numerical examples is a presentational limitation, not a correctness risk for a pure math.OC paper. Consequently the Reader’s ACCEPT / high-confidence verdict needs no adjustment.","tokens_in":15651,"tokens_out":480,"duration_ms":5779,"concrete_test":"Independently re-derive the one-step decrease of V from the shifted candidate (Definition 3) through equations (33)–(37) without invoking the potential-game property a priori; confirm that the only place the tracking form (19) is used is the cancellation supplied by the DARE identity (14)/Lemma 3, and that the shared-constraint verification in Step 4 of Theorem 1 continues to hold under the min-share rule (16). If both checks succeed, the theorems stand as written.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claims (recursive feasibility of the tube-based GNEP for every bounded disturbance, and asymptotic convergence of nominal states to the unique steady-state vGNE with actual states to the mRPI neighborhood) rest on standard tube-MPC constructions plus a potential-game Lyapunov argument that is fully spelled out. Theorem 1 constructs an explicit shifted candidate and verifies every constraint, including the shared one via the offline share allocation (16)–(17) and positive invariance of S_i^f. Theorem 2 uses the potential V = sum J_i (Lemma 2) together with the DARE identity (14)/Lemma 3 to obtain the one-step decrease (31); the rest is a standard telescoping + positive-definiteness argument. The tracking-cost restriction that makes the game potential is stated explicitly (Remark 3) and is the same structural hypothesis the Reader flags; it is not hidden. No internal inconsistency, missing step, or unstated assumption that would invalidate the theorems under the paper’s own hypotheses was found.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper develops a robust receding-horizon game (RHG) for N agents with linear dynamics under bounded additive disturbances, private state/input constraints, and shared coupling constraints. Each agent solves a tube-based finite-horizon GNEP with tightened private and shared constraints (only scalar worst-case contributions d_j are broadcast), a DARE-based terminal cost, a prestabilizing tube feedback, and a decoupled positively invariant terminal set obtained by offline share allocation of the tightened coupling bound. Theorem 1 proves recursive feasibility of the joint GNEP for every disturbance realization by an explicit shifted candidate. Exploiting the potential-game structure induced by pure tracking stage costs, Theorem 2 shows that the joint potential decreases, so each agent’s nominal state converges to the unique steady-state variational GNE while the actual state converges to the mRPI neighborhood of that equilibrium.","tokens_in":15902,"tokens_out":1116,"duration_ms":30437,"significance":"The work closes a clear gap between deterministic receding-horizon games and tube-based robust (cooperative) distributed MPC. The combination of privacy-preserving scalar tightening, offline share allocation that decouples a coupled terminal constraint, and a fully spelled-out potential-game Lyapunov argument yielding both recursive feasibility and asymptotic convergence under additive uncertainty appears to be new. The proofs of Theorems 1–2 and Lemmas 1–3 are complete under the stated assumptions; the tracking-cost hypothesis that makes the game potential is stated explicitly (Remark 3) rather than hidden. These are solid, machine-checkable theoretical contributions of genuine interest to the multi-agent MPC and game-theoretic control communities.","major_comments":[{"comment":"The stability argument (Theorem 2, Lemma 2, Remark 3) rests entirely on pure tracking stage costs (19) that render the finite-horizon GNEP a potential game. While the authors acknowledge this restriction and list general economic/coupled costs as future work, the abstract and title present the framework more broadly as “Robust Receding Horizon Games.” A short clarifying sentence in the abstract (or a dedicated remark early in Section V) stating that the Lyapunov decrease (31) is specific to tracking costs would prevent over-reading of the scope.","section":null},{"comment":"Section IV and Remark 2 permit a polytopic outer approximation of the mRPI set Z_i^∞. Recursive feasibility (Theorem 1, Step 1) and robust constraint satisfaction (Remark 4) require that the set used in place of Z_i^∞ itself be robustly positively invariant under the prestabilizing dynamics. The manuscript should state this requirement explicitly (the algorithm of [25] produces such invariant outer approximations, so the fix is only textual).","section":null},{"comment":"The paper contains no numerical example. While the theorems are self-contained, a minimal two- or three-agent illustration (showing recursive feasibility under a nontrivial disturbance sequence, the evolution of the potential V, and convergence of actual states into the mRPI neighborhood) would substantially increase confidence that the offline share allocation (16)–(17) and terminal-set computation are practical, and is standard for this class of contribution.","section":null}],"minor_comments":[{"comment":"Notation: the same symbol b_i appears for the original shared bound and, after tightening, as the right-hand side of (9); introducing b̄_i earlier (already done) and consistently using it in (2d) versus (3b)/(20d) would reduce momentary confusion.","section":null},{"comment":"Definition 2 defines S_i^f as “the largest set” satisfying the three conditions; a one-line remark that any positively invariant subset containing s_i^* would also work for the subsequent proofs would be helpful for readers who compute only an inner approximation.","section":null},{"comment":"In Algorithm 1 the online step cites [24], [29] for distributed vGNE computation; a brief note that these methods are assumed to return an exact (or sufficiently accurate) vGNE at each sampling instant would align the algorithmic claim with the exact-equilibrium analysis of Theorems 1–2.","section":null},{"comment":"Typos / polish: “Mignoniet al.” → “Mignoni et al.”; “Conteet al.” → “Conte et al.”; “Stewartet al.” → “Stewart et al.”; “Trodden and Richards [19]” is fine but the surrounding list is missing spaces before “et al.”","section":null},{"comment":"The communication graph is required only to be connected (Assumption 1); a short remark on whether the share-allocation protocol (17) needs only neighborhood broadcasts (yes) would make the privacy claim fully self-contained.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The novelty claim relative to cooperative tube-MPC and deterministic RHG is credible; the related-work discussion is fair. The absence of numerics is the main practical weakness for a broad control-systems audience, but it is not a correctness issue. Fit for a theory-oriented venue (TAC, Automatica, SIAM J. Control Optim.) is good."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"This is the first clean certificate package for receding-horizon games under additive disturbances: tube tightening that only broadcasts a scalar, DARE terminal cost, and a share-allocation trick that decouples the terminal set so each agent can certify invariance locally. Theorems 1 and 2 are written out carefully. The shifted candidate works for the shared constraint because of the offline α shares (16)–(17), and the Lyapunov decrease follows directly from the potential V = sum J_i plus the DARE identity. That is real progress relative to the deterministic RHG papers (Hall et al.) and the cooperative robust-MPC literature they cite.\n\nWhat they do well is keep the privacy claim honest and the assumptions transparent. Agents never exchange dynamics or predicted trajectories—only the scalar d_j and the steady-state contributions needed for the shares. The tracking-cost restriction that makes the game potential is stated up front (Remark 3) and is exactly the same structural hypothesis that limits the result; they do not hide it. The mRPI outer-approximation is standard and does not break the arguments.\n\nSoft spots are minor and proportional. There are no numerical examples, so we do not yet see how conservative the tubes and shares become in practice. Free parameters (Q_i, R_i, K_p, H) are the usual ones; nothing is fitted and then re-sold as a prediction. The circularity burden is essentially zero. The paper is pure math.OC inside its stated scope.\n\nThis is for people who already work on multi-agent MPC or GNEPs and need a robust baseline they can cite. It is not a field-reorganizing result, but it is a legitimate methodological advance that a serious referee should see. I would accept it for peer review, bring it to reading group if we are doing robust multi-agent control that month, and expect to cite the feasibility/stability theorems when the setting matches.","headline":"Solid first robust RHG with full recursive-feasibility and potential-game stability proofs; tracking-cost restriction is explicit and the missing numerics are the only real soft spot.","tokens_in":16479,"tokens_out":481,"would_cite":true,"duration_ms":6435,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["93B45","91A10","93D20","49N10"],"pacs":[],"model":"grok-4.5","headline":"Selfish agents with private data and shared constraints can still drive uncertain linear systems to a common equilibrium neighborhood under receding-horizon play.","keywords":["receding horizon games","robust model predictive control","generalized Nash equilibrium","potential games","tube-based constraint tightening","additive uncertainty","variational GNE"],"falsifier":"Replace the tracking stage cost by a general economic cost (or couple the agents’ costs) while keeping the same tube tightening and terminal ingredients; check whether the joint potential still decreases and whether nominal trajectories still converge to the steady-state vGNE for a simple two-agent linear example.","tokens_in":16552,"feed_emoji":"🎯","tokens_out":622,"duration_ms":6675,"temperature":0.7,"pith_summary":"When several agents each run their own model-predictive controller on a linear plant that is hit by bounded noise, and they must also obey shared coupling constraints without revealing their dynamics or costs, ordinary cooperative distributed MPC does not apply. This paper shows that a tube-based tightening of both private and shared constraints, together with a simple DARE terminal cost and a carefully split terminal set, keeps the finite-horizon generalized Nash game recursively feasible for every disturbance. Because the stage costs are pure tracking costs, the game is a potential game; the joint potential then decreases at every step and forces every agent’s nominal trajectory to the unique steady-state variational equilibrium while the true state remains inside a minimal robust neighborhood of that equilibrium. The result gives the first rigorous closed-loop guarantees for competitive multi-agent MPC under additive uncertainty and privacy constraints.","feed_headline":"Uncertain agents reach equilibrium without sharing models","feed_subtitle":"Tube tightening and a potential Lyapunov function keep competitive MPC feasible and convergent under noise.","key_machinery":"The potential function formed by summing the individual tracking costs: because each agent’s cost depends only on its own predicted trajectory, the pseudo-gradient of the GNEP coincides with the gradient of this joint potential, which therefore serves as a common Lyapunov function whose one-step decrease is exactly the sum of the stage costs.","core_discovery":"A tube-based receding-horizon generalized Nash equilibrium problem, closed by a DARE terminal cost and a resource-allocation terminal set that decouples the shared terminal constraint, is recursively feasible for every bounded disturbance and drives every agent’s nominal state to the unique steady-state variational GNE while the actual state converges to the corresponding minimal robust positively invariant neighborhood.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["Tube-based GNEPs keep multi-agent MPC feasible under additive noise","Nominal states converge to steady-state vGNE despite disturbances","DARE terminals and tubes yield recursive feasibility for uncertain agents","Potential games drive robust receding-horizon agents to mRPI neighborhoods","Decoupled terminal sets ensure GNEP feasibility for every bounded disturbance"],"cache_read_input_tokens":128,"weakest_assumption_plain":"The entire Lyapunov argument collapses if the stage costs are not pure tracking costs of the special quadratic form that makes the finite-horizon game a potential game.","fun_headline_variants_meta":{"raw":{"variants":["Tube-based GNEPs keep multi-agent MPC feasible under additive noise","Nominal states converge to steady-state vGNE despite disturbances","DARE terminals and tubes yield recursive feasibility for uncertain agents","Potential games drive robust receding-horizon agents to mRPI neighborhoods","Decoupled terminal sets ensure GNEP feasibility for every bounded disturbance"]},"model":"grok-4.5","effort":"low","cost_usd":0.004706,"raw_usage":{"total_tokens":1291,"prompt_tokens":664,"num_sources_used":0,"completion_tokens":94,"cost_in_usd_ticks":47060000,"prompt_tokens_details":{"text_tokens":664,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":533,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":664,"tokens_out":94,"duration_ms":4927,"temperature":1.0,"reasoning_tokens":533,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-11T20:54:04.943164+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Replace the tracking stage cost by a general economic cost (or couple the agents’ costs) while keeping the same tube tightening and terminal ingredients; check whether the joint potential still decreases and whether nominal trajectories still converge to the steady-state vGNE for a simple two-agent linear example.","supporting_citations":[],"review_version":1}