{"id":"ce7fa03b-509a-4629-8674-a36e49315e62","arxiv_id":"2505.22352","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"A saturated barrier-Lyapunov adaptive controller for uncertain Euler-Lagrange systems is claimed to guarantee state and input constraints, and includes a feasibility condition on the input bound.","lead":"This paper proposes an adaptive controller for uncertain robotic systems that promises to keep both joint positions and control torques inside user-specified limits while tracking a reference path. It adds a checkable feasibility condition that links the allowed state bounds, the input bound, and the disturbance size.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The unsaturated control case is not analyzed in Theorem 1: with Delta_tau=0, inequality (25) permits Vdot>0 for small ||r|| whenever dbar > lambda_min(K1)||r||, so Lemma 1's hypothesis Vdot<0 on Omega_r is unverified and the invariant set ||r||<kappa is unsupported.","rationale":"The reader's weakest-assumption identification is exactly the load-bearing concern. The theorem's guarantee of state constraint satisfaction hinges on the invariant set ||r||<kappa, established via Lemma 1. Lemma 1 requires Vdot<0 on the entire admissible region. In the unsaturated case, however, the disturbance term in (25) is not dominated for small ||r||, so the lemma's hypothesis is not established. This is not a cosmetic omission: the paper's own simulation values (dbar=5, lambda_min(K1)=1, kappa=0.04) put the entire set ||r||<kappa inside the region where the upper bound on Vdot is positive, and a worst-case disturbance can make Vdot positive at t=0. The proof therefore does not show that the BLF cannot grow to its boundary, and the state constraint conclusion (37)-(41) does not follow. The feasibility condition C1 is a useful contribution and the saturated-case analysis is mostly sound, but because the main theorem is unsupported, the preprint needs revision before acceptance. I would keep the reader's reject/revise verdict; no separate issue is needed to explain the decision.","tokens_in":11587,"tokens_out":10056,"duration_ms":107949,"concrete_test":"Re-derive Case 1 of the proof by substituting Delta_tau=0 into (25) and attempting to verify Lemma 1's condition Vdot<0 at all ||r||<kappa. Using the paper's parameter values, take lambda_min(K1)=1, dbar=5, kappa=0.04, choose r(0)=0.01 e_2, theta_hat(0)=theta, and d(0)=5 e_2 (equivalently d=dbar r/||r||), with initial conditions arranged so ||u(0)||<=taubar. Since theta_tilde(0)=0, the exact Vdot at t=0 is positive, contradicting the proof's claim (35). If the authors instead complete Case 1 with a new argument or an added robustifying term, verify that Theorem 1's assumptions and controller (8), (9), (20) remain unchanged.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Theorem 1 has a decisive gap at the Case 1 / Case 2 split. Case 1 only states that ||Delta_tau||=0; it does not show Vdot<0. Indeed, with Delta_tau=0, (25) becomes Vdot <= (||r||/(kappa_m^2 - mbar ||r||^2)) (dbar - lambda_min(K1)||r||). This upper bound is positive for every ||r|| < dbar/lambda_min(K1). The manuscript's own simulation parameters give dbar=5, lambda_min(K1)=1, kappa=0.04, so the bound is positive throughout Omega_r. Lemma 1's hypothesis (18) requires Vdot<0 on the whole set Omega_r, not merely near the boundary; hence the lemma cannot be applied. The assertion after (34) that Vdot(0)<0 and 'repeating the argument' yields Vdot(t)<0 is not justified: Case 1 is admissible at t=0 whenever the unsaturated controller is active, and with theta_hat(0)=theta and d = dbar r/||r|| one obtains Vdot(0)>0 exactly. Since the proof's derivation of ||r(t)||<kappa, and therefore of the state constraints (40)-(41), depends entirely on this invalid Lemma 1 invocation, the central state-constraint theorem is unsupported. A related compounding issue is that the saturated-regime bound (29), derived under ||u||>taubar, is treated as if it could be substituted into (25) in Case 1; this is not legitimate.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an adaptive tracking controller for uncertain Euler-Lagrange systems with user-defined state and input constraints. The design combines a barrier Lyapunov function (BLF) for state constraints with a saturated control law for input constraints, and uses a projection-based adaptive update law for parametric uncertainty and bounded disturbances. The main contribution is a claimed verifiable feasibility condition (C1) that guarantees state and input constraint satisfaction and boundedness of all closed-loop signals. The claims are supported by a Lyapunov analysis and a simulation on a two-link robotic manipulator.","tokens_in":11959,"tokens_out":11599,"duration_ms":102866,"significance":"If the main theorem were correct, the result would be a useful optimization-free alternative to MPC and CBF methods for constrained Euler-Lagrange systems, with the advantage of a verifiable feasibility condition. The paper includes a simulation comparison with a classical robust adaptive controller, showing constraint satisfaction with the proposed method. However, the proof of Theorem 1 has serious gaps that undermine the central claim, and the feasibility condition derivation contains an algebraic error. The core theoretical contribution is therefore not established in the present form.","major_comments":[{"comment":"In the unsaturated case, Eq. (25) with Delta_tau = 0 gives Vdot <= (||r||/(kappa_m^2 - mbar||r||^2)) (dbar - lambda_min(K1)||r||). This upper bound is positive for all ||r|| < dbar/lambda_min(K1). The feasibility condition C1 does not imply lambda_min(K1)*kappa > dbar; indeed, with the simulation parameters dbar=5, lambda_min(K1)=1, kappa=0.04, the bound is positive on the entire set Omega_r. Therefore Lemma 1's hypothesis (18), which requires Vdot < 0 on the whole set Omega_r, is not verified, and the conclusion ||r(t)|| < kappa, and hence the state constraints (40)-(41), are unsupported.","section":"Section III, Theorem 1 proof, Case 1 (after Eq. (25))"},{"comment":"Inequality (29) is derived under the saturation condition ||u|| > taubar (Case 2). In Case 1, Delta_tau = 0, so (29) is not applicable. The proof substitutes (29) into (25) without regard to which case is active, and then concludes Vdot < 0 in (31). This mixing of the two cases invalidates the derivation of the negative-definiteness bound and the subsequent invariant-set argument.","section":"Section III, Theorem 1 proof, Cases 1 and 2 (Eqs. (29)-(31))"},{"comment":"The algebraic step from (32) to (33) is incorrect. Substituting the definitions of Psi, xi, and kappa = EV - alpha EQ into (32) yields the lower bound taubar > theta_bar(alpha^2 EQ + alpha EQ + Vd + alpha3 + 2) + dbar + (EV - alpha EQ)(theta_bar(alpha + 2) + lambda_max(K1) - lambda_min(K1)), not the expression in (33), which has theta_bar(2alpha + 3) instead of theta_bar(alpha + 2). Moreover, the right-hand side of (34) is larger than the right-hand side of (33), so the claimed implication (33) => (34) has the wrong direction. While C1 may be a sufficient condition if proved directly, the derivation as written does not establish it.","section":"Section III, Theorem 1 proof, Eqs. (32)-(34)"},{"comment":"The claim that Vdot(0) < 0 and 'repeating the argument' yields Vdot(t) < 0 for all t is not justified. At t=0 the controller may be unsaturated, and the admissible initial condition ||r(0)|| < kappa permits ||r(0)|| < dbar/lambda_min(K1). For an adversarial disturbance d = dbar r/||r|| and theta_hat(0) = theta, the exact expression (24) gives Vdot(0) > 0 when ||r(0)|| < dbar/lambda_min(K1). Thus (36) is false under the stated assumptions, and the proof of the invariant set fails.","section":"Section III, Theorem 1 proof, Eqs. (35)-(36)"}],"minor_comments":[{"comment":"The set Omega'_r is defined as a subset of R, but it should be R^n; the norm ||r|| is n-dimensional. This is likely a typo but should be corrected.","section":"Section III, Lemma 1 and Theorem 1 setup"},{"comment":"The caption for Fig. 1 is missing; it currently reads only 'Caption'.","section":"Section IV, Fig. 1"},{"comment":"The constants delta and omega3 are described as positive, but delta is never defined anywhere in the paper.","section":"Section III, Eq. (34)"},{"comment":"The simulation states that the control input satisfies ||tau(t)|| < 30, while the input constraint in Section II is ||tau(t)|| <= taubar. Consistency in the inequality direction would be clearer.","section":"Section V, simulation parameters"}],"recommendation":"reject","confidential_remarks":"The paper addresses a relevant problem and the simulation suggests the approach may work in practice, but the theoretical proof of the main theorem has a decisive gap in the unsaturated control case, and the feasibility condition derivation contains an algebraic error. These are not merely presentation issues; they affect the central claim. Even though the result might be repairable with additional assumptions or a different analysis, the current manuscript does not provide a valid proof of its main contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper offers a genuinely useful feasibility check for constrained adaptive tracking of uncertain Euler-Lagrange systems: the condition C1 and the gain bound alpha < EV/EQ give designers a way to pick state and input limits without running an optimization. That is a real step beyond the authors' earlier ACC paper, which required full mass-matrix knowledge and had no feasibility analysis. The saturated-control analysis in Case 2 is internally coherent, and the simulation comparison against a standard robust adaptive controller is honest.\n\nThe soft spot is load-bearing. The proof of Theorem 1 claims the filtered tracking error stays inside the set ||r|| < kappa for all time by invoking Lemma 1. But Lemma 1 requires Vdot < 0 on the whole set Omega_r. The proof only establishes negativity in the saturated regime, where ||u|| > taubar. In the unsaturated regime (Case 1), Delta_tau = 0, and inequality (25) reduces to Vdot <= ||r||(dbar - lambdamin(K1)||r||)/(kappa_m^2 - mbar||r||^2), which is positive for ||r|| < dbar/lambdamin(K1). With the paper's own simulation values (dbar=5, lambdamin(K1)=1, kappa=0.04), that upper bound is positive on the entire set. The authors try to avoid this by substituting the saturated-case bound (29) into (25) in Case 1, but that bound is not valid when the control is unsaturated. So the central state-constraint theorem is unsupported. The leap from Vdot(0)<0 to \"repeating the argument gives Vdot(t)<0 for all t\" is not justified.\n\nThis is a decisive gap, not a minor presentation issue. The feasibility condition and the architecture are worth preserving, but the main theorem needs either an additional assumption (e.g., dbar < lambdamin(K1)kappa, which would contradict the chosen parameters) or a modified claim about uniform ultimate boundedness with a tunable residual set. The paper also has small problems: undefined symbols, a mislabeled figure reference, and the \"first result\" claim is worded too strongly given the proof gap.\n\nWho should read this? Researchers working on constraint satisfaction for Euler-Lagrange systems will find the feasibility trade-off plots and the adaptive-saturation architecture useful, even if the theorem is not yet proven. A serious editor should send this to peer review, not desk-reject it, because the idea is new and the gap might be fixable. But it should not be accepted in its current form.\n\nRecommendation: send to review with the expectation of major revision; the authors need to close the unsaturated-case analysis or weaken the claim.","headline":"Promising feasibility condition and a real gap: the unsaturated-case analysis never establishes Vdot<0, so the main state-constraint theorem is unsupported.","tokens_in":12485,"tokens_out":2437,"would_cite":false,"duration_ms":23410,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["93C40","93C10","93D30","93C85"],"pacs":[],"model":"deepseek-v4-flash","headline":"An adaptive, saturation-based controller can certify state and input constraint satisfaction for uncertain Euler-Lagrange systems, provided an explicit feasibility inequality holds.","keywords":["adaptive tracking control","Euler-Lagrange systems","barrier Lyapunov function","input saturation","state constraints","feasibility condition","projection-based adaptation","robotic manipulator"],"falsifier":"Run the Section V simulation with the disturbance set to its allowed bound $\\bar{d}=5$ from $t=0$ while the control remains unsaturated, and record the filtered tracking error $\\|r(t)\\|$; if it ever reaches $\\kappa=E_V-\\alpha E_Q=0.04$, the theorem's invariant-set claim fails. A cheaper check is to evaluate the right side of (25) at small $\\|r\\|$ with $\\|d\\|=\\bar{d}$: a positive value shows the Lyapunov argument alone does not certify invariance.","tokens_in":1822,"feed_emoji":"🤖","tokens_out":4665,"duration_ms":120721,"temperature":0.7,"pith_summary":"This paper tries to establish that safety-critical tracking for uncertain Euler-Lagrange systems, such as robotic manipulators, does not require online optimization or exact model knowledge. It proposes an adaptive controller that combines a barrier function for state limits, saturation for actuator limits, and projection-based parameter estimation for disturbance robustness, and it states an explicit feasibility inequality that tells a user when a chosen set of limits can actually be met. If the theorem is right, an engineer can choose position, velocity, and torque bounds, check two inequalities, and run the controller with a guarantee that those bounds hold at every time. The paper also works out how the three bounds trade off against one another, and demonstrates the controller on a two-link manipulator with injected disturbances.","feed_headline":"Adaptive controller keeps robot inside state and torque limits","feed_subtitle":"A verifiable torque-limit condition certifies one optimization-free controller for constrained Euler-Lagrange tracking.","key_machinery":"The central mechanism is a barrier Lyapunov function defined on the filtered tracking error, $V(r)=\\frac{1}{2}\\log\\frac{\\kappa_m^2}{\\kappa_m^2-\\bar{m}\\|r\\|^2}$, which goes to infinity as $\\|r\\|$ approaches $\\kappa$; keeping $V$ bounded forces $\\|r\\|<\\kappa$, and the relations among $e$, $\\dot{e}$, and $r$ then transfer that bound to the position and velocity errors. A saturation block clamps each control component $\\tau_i$ to $\\frac{\\bar{\\tau}}{\\|u\\|}u_i$ when the auxiliary control exceeds the torque limit, and the feasibility condition C1 bounds the resulting saturation error $\\Delta\\tau=\\tau-u$ so that it can be absorbed in the Lyapunov analysis. Projection in the update law keeps $\\hat{\\theta}$ inside the known ball, which is what makes the disturbance and saturation terms manageable.","core_discovery":"The paper's central claim is Theorem 1: for the uncertain Euler-Lagrange system (1) with bounded disturbance $\\|d(t)\\|\\le \\bar{d}$, unknown parameters drawn from a known ball $\\|\\theta\\|<\\bar{\\theta}$, and a reference trajectory whose position and velocity stay strictly inside the state bounds, the saturated feedback controller (8)--(9) together with the projection-based update law (20) guarantees that $\\|q(t)\\|<\\bar{Q}$, $\\|\\dot{q}(t)\\|<\\bar{V}$, and $\\|\\tau(t)\\|\\le \\bar{\\tau}$ for all $t\\ge 0$, and that all closed-loop signals remain bounded. The guarantee is conditional on two explicit user-checkable inequalities: the gain condition $\\alpha<E_V/E_Q$ and the feasibility condition $\\bar{\\tau}>\\omega_1+\\omega_2\\bar{V}-\\omega_3\\bar{Q}$. The proof routes the state constraints through a filtered tracking error $r=\\dot{e}+\\alpha e$ and maintains $\\|r\\|<\\kappa=E_V-\\alpha E_Q$.","pith_inferences":["The feasibility condition can be read as an offline actuator-sizing rule: a user could check whether a candidate torque ceiling is large enough for a given trajectory envelope and known disturbance bound before deployment, although the paper does not develop that certification workflow.","A natural test of the theorem is to run the two-link example with the disturbance at its allowed bound $\\bar{d}=5$ while the control is unsaturated; watching whether the filtered error $\\|r\\|$ stays below $\\kappa=0.04$ would probe the proof's treatment of the unsaturated case.","The saturated-BLF construction may extend to per-coordinate state constraints or to underactuated Lagrangian systems, but each generalization would need its own feasibility condition because the norm-based derivation in C1 depends on the specific vector bound."],"forward_implications":["If Theorem 1 is correct, the controller keeps $\\|q(t)\\|<\\bar{Q}$, $\\|\\dot{q}(t)\\|<\\bar{V}$, and $\\|\\tau(t)\\|\\le \\bar{\\tau}$ for every $t$, including transients, so the safety guarantee is not merely asymptotic.","Condition C1 directly quantifies the trade-off between actuator authority and state limits: tightening the position bound $\\bar{Q}$ increases the required torque bound $\\bar{\\tau}$, while increasing the velocity bound $\\bar{V}$ also demands more control authority.","The gain condition $\\alpha<E_V/E_Q$ gives a design range for the filter gain, preventing either the position error or the velocity error from dominating and violating the allowed envelope.","Because no quadratic program or model predictive control optimization is run online, the computational cost per step is comparable to standard adaptive control, which matters for fast or resource-limited systems."],"supporting_citations":[{"why":"Supplies the barrier Lyapunov function lemma (Lemma 1) that the state-constraint argument relies on.","marker":"[13]"},{"why":"The authors' earlier adaptive controller for the same problem, which assumed the control effort always stays below saturation; the present paper removes that assumption and adds a feasibility condition.","marker":"[29]"},{"why":"Provides the standard Euler-Lagrange properties—bounded inertia, skew symmetry, linear parameterization—used throughout the Lyapunov proof.","marker":"[30]"},{"why":"Defines the projection operator used in the adaptive update law (20) to keep parameter estimates within their known bound.","marker":"[31]"},{"why":"The classical robust adaptive controller used as the comparison baseline in the simulations and as justification for the bounded-parameter Assumption 3.","marker":"[1]"},{"why":"A zeroing control barrier function approach for E-L systems with state and input constraints that requires full model knowledge and online quadratic programming; the present method positions itself as an optimization-free alternative.","marker":"[28]"}],"fun_headline_variants":["Verifiable feasibility for adaptive constrained tracking","First certifiable controller for state and torque limits","Adaptive tracking with proven state and input bound guarantees","Robust adaptive control with explicit feasibility conditions","Constrained Euler-Lagrange tracking made certifiable"],"cache_read_input_tokens":14464,"weakest_assumption_plain":"The proof relies on the barrier-function derivative being negative at every time the filtered tracking error is inside its safe interval, but the negative sign is only demonstrated in the saturated-control case; in the unsaturated case the bound in equation (25) allows the derivative to be positive, so the invariant set $\\|r\\|<\\kappa$ is asserted rather than proved.","fun_headline_variants_meta":{"raw":{"variants":["Verifiable feasibility for adaptive constrained tracking","First certifiable controller for state and torque limits","Adaptive tracking with proven state and input bound guarantees","Robust adaptive control with explicit feasibility conditions","Constrained Euler-Lagrange tracking made certifiable"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000181,"raw_usage":{"total_tokens":1272,"prompt_tokens":874,"completion_tokens":398,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":490,"completion_tokens_details":{"reasoning_tokens":327}},"tokens_in":490,"tokens_out":398,"duration_ms":4720,"temperature":1.0,"reasoning_tokens":327,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T13:10:26.368857+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the Section V simulation with the disturbance set to its allowed bound $\\bar{d}=5$ from $t=0$ while the control remains unsaturated, and record the filtered tracking error $\\|r(t)\\|$; if it ever reaches $\\kappa=E_V-\\alpha E_Q=0.04$, the theorem's invariant-set claim fails. A cheaper check is to evaluate the right side of (25) at small $\\|r\\|$ with $\\|d\\|=\\bar{d}$: a positive value shows the Lyapunov argument alone does not certify invariance.","supporting_citations":[{"cited_title":"Adaptive tracking control of uncertain euler- lagrange systems with state and input constraints,","cited_arxiv_id":null,"evidence_quote":"The authors' earlier adaptive controller for the same problem, which assumed the control effort always stays below saturation; the present paper removes that assumption and adds a feasibility condition."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the standard Euler-Lagrange properties—bounded inertia, skew symmetry, linear parameterization—used throughout the Lyapunov proof."},{"cited_title":"Adaptive robust control of a class of uncertain nonlinear systems with unknown sinusoidal disturbances,","cited_arxiv_id":null,"evidence_quote":"The classical robust adaptive controller used as the comparison baseline in the simulations and as justification for the bounded-parameter Assumption 3."},{"cited_title":"Correct-by-design control barrier functions for euler-lagrange systems with input constraints,","cited_arxiv_id":null,"evidence_quote":"A zeroing control barrier function approach for E-L systems with state and input constraints that requires full model knowledge and online quadratic programming; the present method positions itself as an optimization-free alternative."}],"review_version":1}