{"id":"aef5989e-bec4-4543-8e68-be2e5f6da14f","arxiv_id":"1908.10175","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A tube-based nonlinear MPC law is shown to make an underactuated AUV track a 3D trajectory while respecting input and obstacle constraints under bounded disturbances.","lead":"This paper designs a corrective autopilot for torpedo-like underwater robots that can follow a 3D path while dodging newly detected obstacles. It pairs an online planner for the idealized robot with a backup feedback controller that keeps the real robot close to that plan despite currents and model errors.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Equation (20) misstates the yaw-rate term in the orientation-error dynamics, so the FHOCP and tube analysis in Theorem 1 are based on a model that differs from the actual AUV kinematics.","rationale":"The paper's goal is a tube-based NMPC law that guarantees ISS and obstacle avoidance for an underactuated AUV. The central condition for that guarantee is that the error dynamics (21) accurately describe the real closed-loop kinematics and that Lemma 1 renders the tube invariant. Equation (20) fails this condition: the yaw-rate term in \\dot e_o is wrong. Using e_o = (e_x/e_d) sinψ - (e_y/e_d) cosψ, the derivative contains (e_x cosψ + e_y sinψ)/e_d times r plus terms with denominator e_d^3; the printed formula has r multiplied by e_x e_y/e_d^2 and all terms over e_d^2. The discrepancy is not a matter of uncomputed constants or a missing reference; it is an internal inconsistency in the main derivation. Therefore the FHOCP (31) optimizes over trajectories of a different system, and the robustness and obstacle-avoidance claims in Theorem 1 and the abstract are unsupported as written. The reader's weakest assumption (uncomputed L1, L2, J, xi_bar) is real, but if those were supplied the model would still be wrong. I therefore keep the reader's CONDITIONAL verdict: the error appears identifiable and correctable, but the paper needs a re-derived (20)-(22) and a re-check of Lemma 1 and Theorem 1 with the corrected dynamics. Agreement with the reader's specific weakest_assumption is 'disagree' because the load-bearing issue is the differential equation itself, not the size of the tube constants.","tokens_in":13290,"tokens_out":18002,"duration_ms":168004,"concrete_test":"Independently re-derive (20) from (16) and (6) for the single case e_x=1, e_y=0, ψ=0, u=w=0, r=1, with zero disturbances and zero desired velocity. Direct differentiation gives \\dot e_o = 1, while (20) gives 0; equivalently, numerically integrate the full kinematics (6) and compare e_o(t) with the prediction of the nominal model (22). If the two disagree, equation (20) and the J-matrix in (21) need correction, and Lemma 1 and Theorem 1 must be re-verified with the corrected model.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Equation (20) and the third row of J(e,pd) in (21) are not the derivative of the orientation error defined in (16). From e_o = (e_x/e_d) sinψ − (e_y/e_d) cosψ and the kinematic equations (6), the coefficient of r in eo-dot is (e_x cosψ + e_y sinψ)/e_d, not e_x e_y/e_d^2, and the term coming from e_d-dot has denominator e_d^3. A direct test: with e_x=1, e_y=0, ψ=0, u=w=0, r=1 and zero disturbances/desired velocity, (16) gives e_o = sin ψ = 0 and eo-dot = r = 1, while (20) gives eo-dot = 0. Since (22), (25), Lemma 1, and the tightened constraints in (32) all use this erroneous J(e,pd), the FHOCP (31) is solved for a model that is not the actual error system (21), so Theorem 1's ISS claim and the obstacle-avoidance guarantee are not established for the real vehicle. The uncomputed Lipschitz constants flagged by the reader are a secondary issue: they concern the magnitude of the tube, whereas this concern invalidates the dynamics the tube is built on.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a robust nonlinear model predictive control (NMPC) scheme for an underactuated autonomous underwater vehicle (AUV) tracking a 3D trajectory in a workspace with obstacles and external disturbances. The controller combines a finite-horizon optimal control problem (FHOCP) solved for nominal error dynamics with an off-line state-feedback tube controller, aiming to keep the real trajectory in a hyper-tube around the nominal trajectory. The main theorem (Theorem 1) claims input-to-state stability of the closed loop and obstacle avoidance for all detected obstacles under the proposed feedback law. A MATLAB simulation study is included to illustrate the behavior of the controller.","tokens_in":13587,"tokens_out":6248,"duration_ms":57231,"significance":"If the theoretical guarantees were established, the proposed tube-based NMPC architecture would be a useful contribution to robust trajectory tracking for underactuated AUVs, particularly for the combination of input constraints, state constraints, and obstacle avoidance. The problem is well motivated and the control architecture is clearly presented. However, the main theorem is not proved in the paper, and the error dynamics used throughout the tube analysis contain a derivational error. As a result, the paper's central theoretical claims are currently unsupported. The simulation study is a useful illustration but does not substitute for a correct proof. The paper ships no machine-checked proofs or reproducible code, and the key design constants are left unspecified.","major_comments":[{"comment":"Equation (20) does not give the correct time derivative of the orientation error defined in (16). From e_o = (e_x/e_d) sin(psi) - (e_y/e_d) cos(psi) and the kinematic equation (6), the coefficient of r in e_o-dot is (e_x cos(psi) + e_y sin(psi))/e_d, not e_x e_y/e_d^2, and the derivative also contains a term proportional to e_d-dot/e_d^2 that is omitted. A direct test is: with e_x=1, e_y=0, psi=0, u=w=0, r=1 and zero disturbances and desired velocity, (16) gives e_o=0 and the true e_o-dot equals r=1, while (20) gives 0. Since (21), (22), (25), Lemma 1, and the tightened constraints in (32) all build on this erroneous J(e,pd), the FHOCP (31) is solved for a model that is not the actual error system, so Theorem 1's ISS and obstacle-avoidance claims are not established for the real vehicle.","section":"IV-A, Eq. (20)"},{"comment":"The proof of Theorem 1 is not provided. The text states that the feasibility proof follows [31], [35], [36] and is omitted, and the nominal asymptotic stability is invoked via a class KL bound without proving that the FHOCP (31) with terminal set (33) delivers this property. In particular, recursive feasibility of (31) under the terminal set and tightened constraints is a load-bearing step for the claimed ISS result and must be demonstrated or precisely imported from the cited works with all hypotheses verified.","section":"IV-C, Theorem 1"},{"comment":"The feasible error set E is defined in (17) as a subset of R4 (since x is in X subset of R4), but the FHOCP (31c) imposes ^e(s) in E with ^e in R3. The paper does not define the projection from the state space X to the error space [ed, ez, eo], nor does it explain how the constraints on x translate to constraints on e. Consequently, the tightened constraint sets in (32) are not well-defined in the space in which the FHOCP is solved.","section":"IV-A, Eq. (17) and IV-C, Eq. (31c)"},{"comment":"The constants L1, L2, J, and ~xi are asserted to exist but are never computed or bounded. The tube radius ~rho = ~xi/(sigma J - L1 - L2) and the tightened sets E and V in (32) depend directly on these values; without concrete bounds or a method to compute them from the vehicle parameters and the disturbance bound ~omega, the proposed controller cannot be instantiated and the claimed robustness guarantee is not constructive.","section":"IV-B, Lemma 1 and Remarks 1-2"}],"minor_comments":[{"comment":"The abstract contains a typo ('or 3D trajectories' for 'of 3D trajectories'), and the conclusion misspells 'Vehicles' as 'Vehciles'.","section":"Abstract and VI"},{"comment":"The notation for the sway velocity v and the disturbance bound ~v is confusing: in (4) ~v bounds the sway velocity, while in (13) ~V is the input bound; consider renaming one of these to avoid ambiguity. Also, the description of the current parameters delta_x, delta_y, delta_z and delta_c is unclear and should be restated.","section":"III-A"},{"comment":"In (31c), 'for all delta t in [t_k, t_k+T]' should presumably read 'for all s in [t_k, t_k+T]', and (31a) writes '^u(s)' while the control input is denoted ^v elsewhere. In the sentence after (32), 'the real error states e and the real control input v satisfy the state and input constraints E, P and U' should refer to the sets E and V, not 'P and U'.","section":"IV-C, Eq. (31c) and surrounding text"},{"comment":"The symbol P is overloaded: it denotes the error-tube set in Lemma 1, a positive definite matrix in the cost function (31a), and the terminal set F in (33) uses ||^e||_P. This overloading makes the text hard to follow; distinct notations should be used.","section":"IV-B and IV-C"},{"comment":"The simulation study does not report the values of the design constants L1, L2, J, ~xi, sigma, or the resulting tube radius ~rho, making it impossible to verify that the simulated controller actually corresponds to the proposed scheme and to the condition (28).","section":"V"}],"recommendation":"major_revision","confidential_remarks":"The paper leans heavily on the authors' own prior work for the proof of the main theorem ([35], [36]) and for parts of the model ([27], [37]); a more self-contained treatment would be appropriate. Given the incorrect error dynamics in Eq. (20) and the omitted proof of Theorem 1, the manuscript currently does not support its central claims, but the underlying tube-based MPC approach is a standard and plausible framework, so a major revision that corrects the derivations and supplies the missing proof could make the contribution sound."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper applies a known tube-based NMPC framework to trajectory tracking for an underactuated AUV with obstacle avoidance. The combination is arguably new: the error model, online obstacle detection, and a sensing-range-limited horizon are not in the cited prior work. The spherical-world setup and the tightened-constraint idea are standard but fit the problem well. A simulation shows the controller avoiding two obstacles while tracking a circle, with error and input plots. That part is fine.\n\nThe trouble is the dynamics. Equation (16) defines e_o = (e_x/e_d) sinψ − (e_y/e_d) cosψ. Differentiating that using (6), the coefficient of yaw rate r is (e_x cosψ + e_y sinψ)/e_d, not e_x e_y/e_d² as written in (20) and in the third row of J(e,p_d) in (21). A direct test: e_x=1, e_y=0, ψ=0, u=w=0, r=1 gives e_o=0 and the true e_ȯ=1, while (20) gives 0. So the FHOCP in (31) is solved for a model that differs from the actual error kinematics. Theorem 1's ISS claim and the obstacle-avoidance guarantee are therefore not established for the real vehicle. This is not a minor typo; the whole tube calculation in Lemma 1 and the tightened constraints in (32) use the wrong Jacobian.\n\nThere are other soft spots of varying size. The set E is defined in (17) as a subset of R⁴ via x∈X, but (31c) uses it for the R³ error vector; that's a dimension mismatch, fixable by redefining E properly. The constants J, L₁, L₂, ξ̄ that determine the tube radius ρ̄ are asserted to exist (Remarks 1–2) but never computed or bounded, so the guarantee is not constructively instantiated. Theorem 1's proof is explicitly delegated to the authors' prior papers, and the recursive feasibility argument is omitted. The simulation has no baseline or comparison, which limits its evidential value, though that is a minor issue.\n\nIs the central idea sound? Possibly. The tube-based machinery is well established, and the application is sensible. But the load-bearing model error means the paper, as it stands, does not deliver what it claims. I would not cite it until the error dynamics are fixed and the constants are actually instantiated.\n\nWould I send it to peer review? Yes. The topic is relevant, the framework is standard, and the flaws are specific and fixable. A serious referee should catch the yaw-rate issue and request a corrected derivation. It deserves referee time, not desk rejection, but it needs substantial revision before it is publishable.","headline":"Tube-based NMPC applied to underactuated AUV tracking, but a wrong yaw-rate term in the error dynamics (eq. 20) breaks the model that Theorem 1 relies on.","tokens_in":14118,"tokens_out":2408,"would_cite":false,"duration_ms":25985,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Tube-based NMPC makes underactuated AUVs input-to-state stable despite currents and model error.","keywords":["underactuated autonomous underwater vehicle","trajectory tracking","tube-based model predictive control","robust control invariant set","input-to-state stability","obstacle avoidance","external disturbances","finite horizon optimal control"],"falsifier":"Compute $L_1$, $L_2$, $J$, and $\\tilde{\\xi}$ for a concrete AUV model over the operating envelope. If $\\sigma J - L_1 - L_2 \\le 0$ for any admissible $\\sigma$, or if the tightened state set $E \\ominus \\mathcal{P}$ is empty, then the hyper-tube invariant set $\\mathcal{P}$ does not exist and the ISS guarantee of Theorem 1 cannot hold. Alternatively, run the simulation with a disturbance sequence whose norm touches $\\tilde{\\xi}$ and check whether $\\|e(t) - \\hat{e}(t)\\|$ ever exceeds $\\tilde{\\rho}$; any violation falsifies the robust invariance claim.","tokens_in":13067,"feed_emoji":"🌊","tokens_out":6640,"duration_ms":58117,"temperature":0.7,"pith_summary":"This paper tackles the problem of steering an underactuated autonomous underwater vehicle (actuated only in surge, heave, and yaw) along a 3D reference trajectory while respecting input limits and avoiding obstacles in a partially known, dynamic workspace. The authors propose a robust nonlinear model predictive controller in which an online optimizer plans for the nominal (disturbance-free) error dynamics, while an off-line linear feedback law keeps the real trajectory inside a hyper-tube of radius $\\tilde{\\rho}$ around the nominal one. The central claim is that if the finite-horizon optimal control problem is feasible at time zero, the closed-loop system is input-to-state stable with respect to currents, waves, and unmodeled dynamics, and every detected obstacle is avoided by construction. The value of the result is that robustness is achieved without solving a stochastic or min-max optimization online; the disturbance rejection is encoded in the off-line tube and the tightened constraints.","feed_headline":"Tube-based MPC keeps underactuated AUVs on track despite currents","feed_subtitle":"Real trajectories stay in a hyper-tube around the planned path, so disturbances and new obstacles can't break tracking or safety.","key_machinery":"The load-bearing mechanism is tube-based model predictive control built around a robust control invariant set. For the error dynamics (21), the deviation $\\rho$ between the real error and the nominal error evolves as (25). A linear feedback $\\kappa(e,\\hat{e}) = -\\sigma(e - \\hat{e})$ is chosen so that the time derivative of $\\Lambda(\\rho) = \\frac{1}{2}\\|\\rho\\|^2$ is negative outside the ball of radius $\\tilde{\\rho} = \\tilde{\\xi}/(\\sigma J - L_1 - L_2)$, where $J$ is a lower bound on the symmetrized Jacobian of $J(e,p_d)$, $L_1$ and $L_2$ are Lipschitz constants of $h$ and $\\zeta$, and $\\tilde{\\xi}$ is the bound on the disturbance term $\\xi(e,\\omega)$. The nominal trajectory is generated by an FHOCP solved for the nominal error dynamics (22), with state and input constraints tightened by the tube: $\\tilde{E} = E \\ominus \\mathcal{P}$ and $\\tilde{V} = V \\ominus [-\\sigma \\circ \\mathcal{P}]$. The prediction horizon is capped by $T \\le \\bar{R}/(\\bar{V} + \\tilde{\\xi})$ so that any newly detected obstacle lies within the sensing range during the horizon. The proof of the theorem follows the quasi-infinite horizon NMPC stability framework and a tube-based MPC proof with a robust control invariant set for Lipschitz nonlinear systems.","core_discovery":"The paper's central claim is Theorem 1: under feasibility of the FHOCP (31a)-(31d) at $t=0$, the control law $v = \\hat{v}(\\hat{e}) - \\sigma(e - \\hat{e})$ renders the closed-loop system input-to-state stable with respect to the disturbance vector $\\omega$, for every initial condition in the nominal error set $\\hat{e}(0)\\in\\tilde{E}$. The proof combines two facts: the nominal error dynamics are asymptotically stabilized by the receding-horizon solution, and the deviation $\\rho(t) = e(t) - \\hat{e}(t)$ remains inside the ball $\\mathcal{P} = \\{\\rho: \\|\\rho\\| \\le \\tilde{\\rho}\\}$ because the feedback law makes the Lyapunov function $\\Lambda(\\rho) = \\frac{1}{2}\\|\\rho\\|^2$ decrease whenever $\\|\\rho\\|$ exceeds $\\tilde{\\xi}/(\\sigma J - L_1 - L_2)$. Consequently, the real trajectory is the nominal trajectory inflated by a tube of radius $\\tilde{\\rho}$, so tightening the state and input constraint sets by that tube suffices to guarantee obstacle avoidance and input saturation despite disturbances.","pith_inferences":["The same tube construction would apply to multi-AUV scenarios if each vehicle treats the others as an additional bounded disturbance, though the required coupling bounds are not analyzed here.","An implementation would first need an identification step to compute or conservatively estimate $L_1$, $L_2$, $J$, and $\\tilde{\\xi}$ from the actual vehicle; the paper's guarantee is conditional on these constants satisfying the strict inequality.","There is an implicit trade-off between disturbance rejection and feasibility: increasing $\\sigma$ shrinks $\\tilde{\\rho}$ but shrinks the tightened input set $\\tilde{V}$, which may make the FHOCP infeasible for aggressive maneuvers.","The spherical-world representation means the obstacle guarantee is conservative: any convex obstacle could be handled by inflating it by the tube radius, a straightforward extension the paper does not state."],"forward_implications":["Feasibility of the nominal FHOCP at startup is the only online condition needed: once it holds, the same controller keeps tracking and safety for all future disturbances bounded by $\\bar{\\omega}$.","The real trajectory never leaves the hyper-tube around the nominal one, so obstacle avoidance and input saturation reduce to an offline geometric tightening of the constraint sets, not an online robustness computation.","Because the FHOCP solves only the nominal dynamics, the computational cost of the robust controller matches that of a standard NMPC, making real-time implementation plausible.","The horizon constraint $T \\le \\bar{R}/(\\bar{V} + \\tilde{\\xi})$ ensures that any obstacle entering the sensing region is seen early enough to be incorporated into the re-planned nominal trajectory, so the obstacle-avoidance guarantee covers newly detected obstacles, not just pre-known ones."],"supporting_citations":[{"why":"Supplies the result that sway velocity can be treated as a bounded perturbation, justifying the nominal kinematic model and the inner disturbance bound.","marker":"[28]"},{"why":"Provides the quasi-infinite horizon NMPC framework with terminal cost and terminal constraint that enforces stability of the nominal FHOCP.","marker":"[31]"},{"why":"Gives the tube MPC construction of a robust control invariant set for Lipschitz nonlinear systems, the basis for Lemma 1 and the tightened constraints.","marker":"[34]"},{"why":"The authors' previous tube-based MPC work whose proof arguments are directly reused for the feasibility and ISS proof of Theorem 1.","marker":"[35]"},{"why":"Defines the robust control invariant set and supplies the Lyapunov/ISS tools used in the proof of Lemma 1.","marker":"[24]"}],"fun_headline_variants":["Tube-based NMPC steers AUVs through obstacles and currents","Tube-MPC makes AUVs robust to waves and obstacles","AUVs navigate tricky waters with tube-based MPC","Robust tube MPC for AUVs dodging obstacles","AUVs track paths robustly via tube-based NMPC"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The guarantee collapses if the Lipschitz constants $L_1$, $L_2$, the Jacobian lower bound $J$, and the disturbance bound $\\tilde{\\xi}$ are not known finite numbers that satisfy $\\sigma J - L_1 - L_2 > 0$, since the tube radius and the tightened constraint sets are built from them.","fun_headline_variants_meta":{"raw":{"variants":["Tube-based NMPC steers AUVs through obstacles and currents","Tube-MPC makes AUVs robust to waves and obstacles","AUVs navigate tricky waters with tube-based MPC","Robust tube MPC for AUVs dodging obstacles","AUVs track paths robustly via tube-based NMPC"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001642,"raw_usage":{"total_tokens":6590,"prompt_tokens":1078,"completion_tokens":5512,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":694,"completion_tokens_details":{"reasoning_tokens":5426}},"tokens_in":694,"tokens_out":5512,"duration_ms":35312,"temperature":1.0,"reasoning_tokens":5426,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T10:50:05.857302+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute $L_1$, $L_2$, $J$, and $\\tilde{\\xi}$ for a concrete AUV model over the operating envelope. If $\\sigma J - L_1 - L_2 \\le 0$ for any admissible $\\sigma$, or if the tightened state set $E \\ominus \\mathcal{P}$ is empty, then the hyper-tube invariant set $\\mathcal{P}$ does not exist and the ISS guarantee of Theorem 1 cannot hold. Alternatively, run the simulation with a disturbance sequence whose norm touches $\\tilde{\\xi}$ and check whether $\\|e(t) - \\hat{e}(t)\\|$ ever exceeds $\\tilde{\\rho}$; any violation falsifies the robust invariance claim.","supporting_citations":[{"cited_title":"Control of underactuated sys- tems with viability constraints,","cited_arxiv_id":null,"evidence_quote":"Supplies the result that sway velocity can be treated as a bounded perturbation, justifying the nominal kinematic model and the inner disturbance bound."},{"cited_title":"Tube MPC Scheme Based on Robust Control Invariant Set with Application to Lipschitz Nonlinear Systems,","cited_arxiv_id":null,"evidence_quote":"Gives the tube MPC construction of a robust control invariant set for Lipschitz nonlinear systems, the basis for Lemma 1 and the tightened constraints."},{"cited_title":"Decentralized Tube-based Model Predictive Control of Uncertain Nonlinear Multi-Agent Systems,","cited_arxiv_id":null,"evidence_quote":"The authors' previous tube-based MPC work whose proof arguments are directly reused for the feasibility and ISS proof of Theorem 1."},{"cited_title":"Nonlinear Systems,","cited_arxiv_id":null,"evidence_quote":"Defines the robust control invariant set and supplies the Lyapunov/ISS tools used in the proof of Lemma 1."}],"review_version":1}