{"id":"cd3cd658-6f4c-45ee-8899-91f95f064805","arxiv_id":"2508.00175","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"An immersion-and-invariance adaptive observer estimates velocity and friction parameters for a stiction-plus-Coulomb system with position-only measurement, enabling globally convergent tracking.","lead":"An adaptive observer estimates velocity and unknown friction parameters using only position measurements. If correct, this would be the first globally convergent solution for a standard stiction-plus-Coulomb friction model, which matters for servo drives without velocity sensors.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Global tracking claim rests on an unverified, state-dependent Assumption 1; the step-plus-ramp simulation may violate it, and the perturbation argument importing [26] is not justified.","rationale":"The reader's weakest-assumption analysis correctly identifies Assumption 1 as the load-bearing condition for the central global tracking claim. My stress-test agrees: without a verifiable excitation condition, Proposition 2 does not substantiate the abstract's 'first globally convergent solution' claim. The false tanh strong monotonicity assertion is a genuine mathematical error but is not the most load-bearing issue, since the weaker monotonicity of tanh still yields Hdot ≤ -γ1 ϑ x̃^2, which is enough for the observer part. The more serious gap is the imported proof of Proposition 2: the paper neither proves Assumption 1 for any reference nor establishes that convergence of the unforced part of (22) persists under the stabilizing σ-term. The step-plus-ramp simulation illustrates the issue because the regressor becomes asymptotically rank-one, suggesting Assumption 1 is not satisfied; if so, the simulation is only evidence of robustness, not of the theorem's hypotheses. This does not warrant rejection, because the observer construction itself is plausible, the local errors are fixable, and the tracking conclusion may hold under stronger or more explicit excitation conditions. The appropriate disposition remains CONDITIONAL, with the condition being a rigorous, self-contained proof of the tracking theorem and verification of Assumption 1 on at least one example, or a reformulation of the excitation condition directly in terms of the reference signal.","tokens_in":12394,"tokens_out":11446,"duration_ms":121256,"concrete_test":"Re-run the step-plus-ramp simulation of Section V with the stated initial conditions and reference. From the simulated closed-loop trajectory, extract φ(t) = [x̂2(t), tanh(100 x̂2(t))]^T. Choose a sequence of intervals [t_k, t_k+T_k] with bounded T_k and compute λ_k = λ_min(∫ φ φ^T dt). Check whether ∑ λ_k^2 diverges. If it is finite, Assumption 1 is violated for the paper's own example, so Proposition 2 does not cover the demonstrated scenario. A secondary check is to recompute the inequality at (14) without the strong monotonicity step and confirm the resulting bound Hdot ≤ -γ1 ϑ x̃^2 is still sufficient to prove the observer convergence claimed in Proposition 1.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The advertised 'globally convergent tracking controller' (Proposition 2) is conditional on Assumption 1, but Assumption 1 is a condition on the closed-loop regressor φ(t) = [x̂2(t), tanh(ϑ x̂2(t))]^T, not on the reference r(t) alone. It is therefore not checkable a priori, and the paper never verifies it for any simulation. In fact, for the step-plus-ramp example of Section V, once the ramp is reached the regressor tends to the constant vector [v*, tanh(ϑ v*)] as x̂2 → v*, so the integral in (20) becomes asymptotically rank-one and λ_k decays at the rate of the error decay. Unless that decay is slow enough to make ∑ λ_k^2 diverge, Assumption 1 fails, placing the paper's own demonstration outside the theorem's hypotheses. Additionally, the proof of Proposition 2 is imported from [26] and the step from global asymptotic stability of the unforced part of (22) to convergence of the forced system with the σ(t) term is asserted without the uniformity or robustness argument required for time-varying systems. The false strong monotonicity claim in (14) is real but localized: tanh's derivative is not uniformly positive on R, so the stronger decay bound is invalid, although the weaker inequality Hdot ≤ -γ1 ϑ x̃^2 still suffices for Proposition 1. The decisive open question is whether the central global tracking claim holds under Assumption 1 for any reference that is actually used to demonstrate the method.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript treats a single-degree-of-freedom mechanical system with linear viscous damping plus a smooth Coulomb/stiction friction term, with only the position measured and the friction parameters unknown. The authors propose an immersion-and-invariance adaptive observer that estimates the velocity and the two friction parameters, and they prove convergence of the velocity estimate. A certainty-equivalent tracking controller is then added, and global tracking is claimed under an excitation assumption (Assumption 1) imposed on the closed-loop regressor. The paper also adapts the observer to a third-order hydro-mechanical model and presents simulations with a LuGre friction model, including a step-plus-ramp reference signal. The abstract advertises this as the first globally convergent solution to adaptive friction compensation without velocity measurement.","tokens_in":12703,"tokens_out":8925,"duration_ms":87425,"significance":"If the tracking result is made fully rigorous, this would be a valuable contribution: the observer is a transparent I&I construction with direct Lyapunov cancellations and no fitted parameters, and the simulations suggest robustness to the LuGre model mismatch. However, the global tracking theorem is conditional on a state-dependent excitation condition whose verification is not supplied, and its proof is delegated to a conference paper rather than derived in the manuscript. The false strong-monotonicity inequality in the observer proof is localized and repairable, but the tracking claim needs additional work before the advertised result is supported.","major_comments":[{"comment":"The mapping tanh does not satisfy the strong-monotonicity Definition 1 on all of R, since tanh'(s)=sech^2(s) tends to 0 as |s| tends to infinity. Consequently the inequality −θ2(ϑx̂2−ϑx2)[tanh(ϑx̂2)−tanh(ϑx2)] ≤ −θ2ϑ^2 x̃2^2, used to obtain (14), is not valid as written. The weaker inequality −θ2(ϑx̂2−ϑx2)[tanh(ϑx̂2)−tanh(ϑx2)] ≤ 0 is available, which still yields ᵊċ ≤ −γ1ϑx̃2^2. That weaker bound is enough, together with boundedness of ᵋdx̃2 from (6), to conclude x̃2∈L^2∩L∞ and hence x̃2(t)→0 via Barbalat's lemma. The proof of Proposition 1 should be corrected to state this weaker bound and to adjust the final argument accordingly.","section":"Section II-B, Definition 1 and Eq. (14)"},{"comment":"Assumption 1 is a condition on the closed-loop regressor φ(t)=[x̂2(t), tanh(ϑx̂2(t))]^T, not a condition on the reference r(t) alone, and it is never verified for any of the simulations. In the step-plus-ramp example of Section V, once the ramp is reached x̂2 tends to a constant v*, so φ(t) tends to the constant vector [v*, tanh(ϑv*)]^T; the integral in (20) then becomes asymptotically rank-one and the λk will decay at the rate of the error convergence. Whether (21) holds is not demonstrated, so the paper's own step-plus-ramp demonstration may lie outside the hypotheses of Proposition 2. The authors should either verify Assumption 1 for the examples or explicitly state that those simulations are heuristic illustrations rather than instances of the theorem.","section":"Section III-B, Assumption 1 and Proposition 2"},{"comment":"The proof of Proposition 2 is not self-contained and contains a nontrivial gap. It is asserted that because σ(t) is bounded and converges to zero, and the unforced part of (22) is globally asymptotically stable, it is sufficient to ensure χ(t)→0. For time-varying systems this is not automatic: a vanishing perturbation can destroy global asymptotic stability unless a uniformity or robustness property is established. The manuscript should either prove this perturbation step directly or state the precise theorem from [26] that justifies it and verify that its hypotheses (including any boundedness and regularity conditions on φ) hold in the present closed loop. In addition, Proposition 1 assumes u is such that the state remains bounded, but Proposition 2 does not prove this boundedness for the closed loop (17); a bootstrap argument is needed before the observer result can be invoked.","section":"Section III-B, proof of Proposition 2"},{"comment":"The Lyapunov function in (31) appears to be missing a square: it should be U = H + (1/2)α1 x̃3^2 rather than U = H + (1/2)α1 x̃3. More importantly, the derivative in (33) contains the term −α1 a3 x̃3^2, but the displayed inequality later contains −α1 a1 x̃3^2 and the constant α3 is defined as α1 a1−1. Unless a3=a1 in the system (26), condition (32) with a1 does not ensure α3>0 as written. This affects the stated result of Proposition 3 and needs to be corrected.","section":"Section IV, Proposition 3 and Eq. (31)-(33)"}],"minor_comments":[{"comment":"The heading contains a typo: “Adaptive obsesrver” should be “Adaptive observer”.","section":"Section II-B heading"},{"comment":"The caption says “PE condition under a cosine reference signal” but the vertical axis is not labeled and no scalar measure of excitation is defined; please clarify what quantity is plotted.","section":"Section V, Fig. 5"},{"comment":"The proof of Proposition 2 is delegated to [26] without stating the assumptions or the theorem; for a journal paper the key argument should be reproduced or at least precisely stated, especially since [26] is not a standard textbook reference.","section":"Section III-B and Section V"},{"comment":"The concluding remarks state that this is the “first solution” of the problem, but the global tracking claim is conditional on Assumption 1; the conclusion should be qualified accordingly.","section":"Section VI"},{"comment":"The notation in (35) reuses the symbols θ̃1 and θ̃2 for errors relative to the LuGre parameters σ2 and FC, after these symbols were defined for the stiction-plus-Coulomb model; this overloading should be made explicit to avoid confusion.","section":"Eq. (35)"}],"recommendation":"major_revision","confidential_remarks":"The central tracking proof relies almost entirely on [26], a conference paper co-authored by one of the authors. This is not improper, but the present manuscript should state the needed conditions and proof rather than referring the reader to an external paper. The novelty claim of being the first globally convergent solution should be checked carefully against [15]-[17] and the adaptive friction-compensation literature before the final version is accepted; the current proof gap in Proposition 2 makes that claim premature. If the authors can supply a complete proof of the perturbation step and verify or weaken Assumption 1, the contribution would be solid enough for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new piece is the adaptive speed observer of Proposition 1, which extends the authors' I&I line to include an unknown Coulomb coefficient θ2. That is a real step beyond their earlier stiction-only observers, and the construction is elegant: the tanh term is absorbed by θ2P(x̂2) = −(1/k1) log(cosh(ϑx̂2)). The Lyapunov argument nearly works, but there is a concrete mistake: tanh is not strongly monotone on all of R in the sense of Definition 1, so the bound in (14) is false as written. The weaker bound Ḣ ≤ −γ1ϑx̃2² still gives x̃2 → 0 via Barbalat's lemma, so the observer claim stands with a minor repair.\n\nThe tracking result is another matter. Proposition 2 is lifted from the co-author's [26] and depends on Assumption 1, a state-dependent condition on the regressor φ(t) that the paper itself calls 'rather cryptic.' The stress-test observation is on point: for the step-plus-ramp simulation, once the ramp is reached φ(t) tends to a constant vector, so the integral in (20) becomes asymptotically rank-one and the λk² sum may fail to diverge. That places the paper's own demonstration outside the theorem's hypotheses. Also, the passage from the unforced system's global asymptotic stability to the forced system with the σ(t) term is asserted without a uniformity argument. These are fixable but matter: the headline claim of a globally convergent tracking controller is conditional, and the actual scope is narrower than the abstract suggests.\n\nCredit where due: the paper is clearly written, honest about the cryptic excitation condition, and the simulations—especially the LuGre robustness test—are useful. The citation pattern is fine; [26] is a co-authored reference, but the import is explicitly flagged and the observer result does not depend on it.\n\nThis deserves a serious referee. The observer result is worth publishing after the monotonicity slip is corrected; the tracking proposition needs either a verifiable sufficient condition or a clearer statement of what remains open. I would suggest a major revision, not a desk reject.","headline":"A solid I&I observer for stiction-plus-Coulomb friction without velocity measurement, with a local proof error and a tracking claim that is conditional on an unverified excitation assumption.","tokens_in":13224,"tokens_out":2090,"would_cite":true,"duration_ms":20515,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["93C40","93B07","93C10"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that friction in a mechanical system can be compensated adaptively using only position measurements, with the speed estimate converging for all initial conditions.","keywords":["adaptive friction compensation","immersion and invariance","velocity observer","Coulomb friction","stiction","global convergence","tracking control","LuGre model"],"falsifier":"Run the closed-loop system with a constant reference r, which makes the estimated regressor φ(t)=[x̂2(t), tanh(ϑx̂2(t))]^T non-exciting under the paper's own discussion. If parameter estimates stay bounded but do not converge and the tracking error does not vanish, the excitation assumption is load-bearing for the global tracking claim; conversely, if tracking still succeeds, the need for Assumption 1 is weakened and the global claim may hold in greater generality.","tokens_in":12197,"feed_emoji":"⚙️","tokens_out":5609,"duration_ms":49380,"temperature":0.7,"pith_summary":"This paper takes on a long-standing control problem: canceling friction in a mechanical system without any velocity measurement and without knowing the friction parameters. It proposes an immersion-and-invariance adaptive observer that estimates both the speed and the two unknown friction coefficients from position measurements alone, and proves the speed estimate converges to the true velocity for all initial conditions whenever the state remains bounded. On top of this observer, a certainty-equivalent position controller is shown to achieve global tracking of a desired reference, provided a state-dependent excitation condition on the estimated regressor holds. If the claims stand, servo drives and robotic joints could drop velocity sensors and still get precise motion control under stiction-plus-Coulomb friction.","feed_headline":"First adaptive friction control proven without velocity sensor","feed_subtitle":"An I&I observer estimates speed and friction parameters from position data alone, then drives global tracking.","key_machinery":"The engine is an immersion-and-invariance construction that splits each estimate into a proportional plus an integral part: x̂2=x2I+k1x1, θ̂1=θ1I-(ϑ/(2k1))x̂$2^{2}$, θ̂2=θ2I-(1/k1)log($\\cosh$(ϑx̂2)). These choices make the estimation-error dynamics admit the Lyapunov function H=½(ϑx̃$2^{2}$+θ̃$1^{2}$+θ̃$2^{2}$), whose time derivative is dominated by a strictly negative term in x̃2 because the tanh function is strongly monotone, and a standard convergence lemma yields x̃2→0. For the tracking result, the same error system is rearranged into a linear time-varying form whose unforced part is globally asymptotically stable under the excitation condition adapted from the cited stability theorem.","core_discovery":"The central claim is that, for the plant ẋ1=x2, ẋ2=-θ1x2-θ2 tanh(ϑx2)+u where only x1 is measured and ϑ is known, the proposed I&I observer guarantees lim_{t→∞}(x̂2(t)-x2(t))=0 for all initial conditions, with all signals bounded, making it the first globally convergent adaptive friction compensator without velocity measurement for this model. The tracking extension states that with the same observer feeding a certainty-equivalent control law, and under Assumption 1 (non-summable excitation of the estimated regressor [x̂2, tanh(ϑx̂2)]^T), both parameter estimates and the tracking error converge to zero over time.","pith_inferences":["Because the proof only uses strong monotonicity of tanh, the observer should generalize to any friction nonlinearity that is strongly monotone in velocity, such as other smooth approximations of the sign function.","The state-dependent excitation condition could be turned into a practical design tool: before commissioning, simulate the reference trajectory and check numerically whether the interval sums in (20)-(21) grow, which would give engineers a verifiable certificate for the tracking claim.","The concluding remarks leave dynamic friction models (LuGre, Dahl, Stribeck) as an open problem; this paper's LuGre simulations suggest robustness to model mismatch, but no proof, so a natural next test is the same observer against the full LuGre model with uncertainty in its internal state dynamics.","The 'global' qualifier should be read as global in initial conditions under an input that keeps the state bounded; the observer proof itself assumes boundedness of the state, so the design does not yet handle inputs that drive the system to infinity."],"forward_implications":["A servo drive or robot joint could run the proposed controller with only a position encoder, avoiding the cost, wiring, and noise of tachometers.","The observer result holds for all initial conditions, not just local or small-error configurations, when the state remains bounded.","The tracking result requires a weaker excitation condition than classic persistent excitation, so references with interspersed quiet intervals may still satisfy it.","The same observer shape extends to a three-state hydro-mechanical system with an additional pressure state, requiring only an upper bound on one friction parameter."],"supporting_citations":[{"why":"Supplies the immersion-and-invariance methodology used to construct the adaptive observer.","marker":"[18]"},{"why":"Provides the global asymptotic stability condition for the unforced error system, cited as the proof of Proposition 2.","marker":"[26]"},{"why":"Theorem 8.4 therein is invoked to conclude convergence of the speed estimation error.","marker":"[27]"},{"why":"Defines the LuGre dynamic friction model used in the simulation validation and supplies the parameter values in Table 1.","marker":"[2]"},{"why":"Frames the adaptive friction compensation problem in the model reference adaptive control context and the need for excitation conditions.","marker":"[6]"},{"why":"Earlier I&I adaptive speed observer for the stiction-only model that this paper extends to stiction plus Coulomb friction.","marker":"[15]"}],"fun_headline_variants":["Adaptive friction fix works with no velocity sensor","Global convergence in friction control without speed measurement","First globally convergent friction compensator, no velocity data","Adaptive observer learns friction without velocity measurement","No velocity sensor needed for adaptive friction control"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The global tracking claim rests entirely on Assumption 1, a state-dependent excitation condition on the estimated speed signal that the paper itself calls 'rather cryptic' and never verifies for any concrete reference; if the condition fails, parameter estimates are not proven to converge and the tracking guarantee disappears.","fun_headline_variants_meta":{"raw":{"variants":["Adaptive friction fix works with no velocity sensor","Global convergence in friction control without speed measurement","First globally convergent friction compensator, no velocity data","Adaptive observer learns friction without velocity measurement","No velocity sensor needed for adaptive friction control"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000421,"raw_usage":{"total_tokens":2139,"prompt_tokens":895,"completion_tokens":1244,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":511,"completion_tokens_details":{"reasoning_tokens":1175}},"tokens_in":511,"tokens_out":1244,"duration_ms":9002,"temperature":1.0,"reasoning_tokens":1175,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T10:20:07.594079+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the closed-loop system with a constant reference r, which makes the estimated regressor φ(t)=[x̂2(t), tanh(ϑx̂2(t))]^T non-exciting under the paper's own discussion. If parameter estimates stay bounded but do not converge and the tracking error does not vanish, the excitation assumption is load-bearing for the global tracking claim; conversely, if tracking still succeeds, the need for Assumption 1 is weakened and the global claim may hold in greater generality.","supporting_citations":[{"cited_title":"Astolfi, D","cited_arxiv_id":null,"evidence_quote":"Supplies the immersion-and-invariance methodology used to construct the adaptive observer."},{"cited_title":"Barabanov and R","cited_arxiv_id":null,"evidence_quote":"Provides the global asymptotic stability condition for the unforced error system, cited as the proof of Proposition 2."},{"cited_title":"Khalil, Nonlinear Systems , Prentice-Hall, NJ, 3rd ed, 2001","cited_arxiv_id":null,"evidence_quote":"Theorem 8.4 therein is invoked to conclude convergence of the speed estimation error."},{"cited_title":"Canudas, H","cited_arxiv_id":null,"evidence_quote":"Defines the LuGre dynamic friction model used in the simulation validation and supplies the parameter values in Table 1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Frames the adaptive friction compensation problem in the model reference adaptive control context and the need for excitation conditions."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Earlier I&I adaptive speed observer for the stiction-only model that this paper extends to stiction plus Coulomb friction."}],"review_version":1}