{"id":"4673fa4d-3533-48e2-8688-986645377145","arxiv_id":"2505.09987","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Using phase-plane analysis, the paper shows that Newell, IDM, and Gipps car-following models each violate at least one stated safety or human-like principle in stationary-lead problems.","lead":"The paper tests familiar car-following models against a checklist of safety and human-like driving rules, and finds that each model fails at least one rule. It gives automated-vehicle researchers a structured way to identify which existing models cannot support provably safe following.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The backward-travel claim for the IDM is proven only locally near equilibrium; the severity of the central critique depends on whether those negative-speed excursions are physically relevant.","rationale":"I agree with the reader that the paper is conditionally acceptable: the phase-plane analysis is a useful framework and most analytical results are plausible. My strongest concern is the local-to-global gap in the IDM backward-travel argument because it is the most novel and most cited limitation, and the phrasing 'regardless of the initial conditions' (Section 4.1) is stronger than what the linearization supports. The safe-stopping-distance critique for IDM is independent, and the BDA-Newell numerical inconsistency noted by the reader is separate; neither changes my overall verdict. The proposed grid-integration test would settle whether the backward-travel phenomenon is real at operational scales, and the conclusion can then be adjusted without altering the paper's core methodological contribution.","tokens_in":15525,"tokens_out":1384,"duration_ms":12328,"concrete_test":"Integrate the full nonlinear IDM (Eq. 20) for a grid of initial conditions spanning the plausible spacing range z(0) in [7, 400] m and speeds v(0) in [0, 33] m/s toward a stationary leader, and record the minimum v(t) attained. Use the paper's parameters (zeta = 7 m, zeta' = 5 m, tau = 1.6 s, alpha = 0.73 m/s^2, beta = 1.67 m/s^2, mu = 33.33 m/s, delta = 4). If every trajectory dips below v = 0 by a non-negligible margin (say less than -0.1 m/s for at least 0.1 s), the forward-traveling violation is global; if some or many trajectories stop with v staying nonnegative except in an arbitrarily small neighborhood, the claim should be weakened to a local artifact.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's most consequential negative claim is that IDM produces backward travel near equilibrium for the stationary lead-vehicle problem (Section 4.1, Eqs. 20-21). The proof linearizes around (v, z) = (0, ζ) and shows the linearized system is a stable spiral, hence v(t) becomes negative for generic nearby initial conditions. This establishes a local instability of the forward-travel constraint, but the central conclusion is that IDM violates the forward-traveling principle and is therefore not human-like. The omitted step is global: the stable spiral argument guarantees sign changes only in a small neighborhood of the equilibrium, and a trajectory that starts far away could enter that neighborhood with a speed that remains positive, or with dynamics that, after entering the linear regime, never crosses v = 0 in a way that produces observable backward travel within finite time. The numerical simulation replicates one published scenario, but it does not establish the claimed 'regardless of initial conditions' result, nor does it quantify the magnitude or duration of backward travel. If the backward-travel episodes are microscopic (e.g., centimeters over milliseconds within a linearization region), then the human-like critique loses force; if they are macroscopic at realistic scales, it lands.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a hierarchy of zeroth-, first-, and second-order principles for safe and human-like car following, derives Newell's simplified model from a subset of those principles, and then uses speed-spacing phase-plane analysis to evaluate bounded-acceleration/deceleration extensions of Newell, the Intelligent Driver Model (IDM), and a simplified Gipps model. It concludes that none of these standard models can simultaneously satisfy all stated principles, citing IDM backward travel and excessive braking distance, Gipps ill-defined spacings and fundamental-diagram inconsistency, and BDA-Newell violations of jam-spacing and forward-travel principles. Part 2 is announced as the construction of a new model that resolves these limitations.","tokens_in":15741,"tokens_out":11753,"duration_ms":122586,"significance":"The axiomatic framework is a genuine strength: the principles are stated explicitly, Newell's model is derived transparently from a subset of them, and the BDA-Newell counterexample is concrete and reproducible. The paper avoids fitted-parameter circularity, and the numerical replications of published IDM and Gipps scenarios are useful diagnostic illustrations. If the IDM critiques can be placed on rigorous footing, the paper would be a valuable reference for car-following model evaluation. The main limitation is that 'human-like' is operationalized through a specific parameter set from ISO 15622 and Treiber et al.; the negative conclusions are therefore conditional on that normative choice, a point the paper acknowledges only in passing.","major_comments":[{"comment":"The statement that the IDM 'travels backward regardless of the initial conditions' is not established by the analysis given. Linearizing at the equilibrium and finding a stable spiral controls only trajectories in a sufficiently small neighborhood of (v, z) = (0, ζ); the paper provides no global argument, such as a basin-of-attraction or Lyapunov-function proof, to show that every trajectory eventually enters that neighborhood and then crosses v = 0. The numerical example in Section 4.2 is a single initial condition. Moreover, the conclusion in Section 6 correctly says 'backward travel near equilibrium,' which is weaker than the Section 4.1 claim. The authors should either supply a global proof or weaken the claim to hold for generic initial conditions sufficiently close to the equilibrium and for the simulated scenario.","section":"Section 4.1, Eqs. (20)-(21)"},{"comment":"The claim that the IDM 'violates the safe stopping distance principle' is not supported by Eq. (10) as stated. Eq. (10) is a formula for the distance required to stop from speed v; it is not an upper bound on when braking may begin. Initiating braking at approximately 1000 m when the required stopping distance is about 366 m is conservative but does not violate any principle listed in Section 2.2. To retain this critique, the authors should introduce an explicit principle such as 'braking should not commence substantially earlier than the safe stopping distance' or present the finding as a human-likeness issue rather than a safety-principle violation. As written, the conclusion that the IDM violates the safe stopping distance principle is overstated.","section":"Section 4.1 and Section 4.2"},{"comment":"Asserting that the simplified Gipps model's fundamental diagram is 'inconsistent with human-driven vehicles' because the wave speed uses τ' rather than τ is an empirical or normative claim, not a consequence of the paper's stated principles. The values τ' = 1 s and τ = 1.6 s are drawn from different sources, and Section 2.2 itself notes variability in acceleration parameters. The mathematical derivation of the triangular fundamental diagram is correct, but the label 'inconsistent with human-driven vehicles' requires empirical support or should be recast as a conditional sensitivity result. Since the conclusion lists this as a limitation of the Gipps model, the claim should be either justified or softened.","section":"Section 5.1, Eq. (28)"}],"minor_comments":[{"comment":"The abstract states that 'numerical simulations and empirical observations validate the theoretical insights,' but the manuscript presents no empirical observations; only numerical replications of published simulations appear in Sections 3-5. Please revise the abstract and introduction to say 'numerical simulations.'","section":"Abstract and Section 1"},{"comment":"The wording that the BDA-Newell model 'prohibits initial spacings smaller than the comfort jam spacing' is imprecise: the model actually computes a negative speed for such initial states, thereby violating the forward-traveling and jam-spacing principles. The distinction matters because the model's failure mode is a computed trajectory, not a hard prohibition.","section":"Section 3.3, Eq. (18)"},{"comment":"The section heading promises 'braking profiles' as a second-order principle, but no braking-profile principle is formally stated; Eq. (10) is a scalar stopping-distance formula rather than a profile. Either define the profile principle or remove that phrase from the list.","section":"Section 2.2.3"},{"comment":"The time-gap definition τ(t) = (z(t) - ζ)/v(t + ε) is undefined when v(t + ε) = 0; please add a convention or restrict the definition to positive planned speeds.","section":"Section 2.1, Eq. (3)"},{"comment":"Eq. (31) sets the initial spacing equal to the safe stopping distance from v(0); this is a specific initial condition for the analytical braking solution, not a general property of the simplified Gipps model. Please state this explicitly to avoid confusion.","section":"Section 5.1, Eq. (31)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript depends on a companion Part 2 that is cited only as a working paper; the editor may wish to confirm its availability before publication. Also, the abstract's 'empirical observations' claim should be corrected. These points do not change my technical recommendation, which is driven by the local-to-global gap in the IDM backward-travel argument and the mischaracterization of Eq. (10) as a violated upper bound."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a look if you care about car-following model evaluation. The paper builds a normative checklist of safe/human-like driving principles, derives Newell's model from those principles, and then audits IDM, Gipps, and bounded-deceleration Newell against the checklist. The phase-plane organization is the real contribution: it makes each model's failure mode visible and comparable rather than scattered across separate derivations.\n\nWhat is genuinely new: the BDA-Newell analysis showing that adding a deceleration bound to Newell's model can violate minimum jam spacing and forward travel is a nice cautionary result. The simplified Gipps analytical solution near the comfort spacing is also clean and useful. The derivations are transparent, the notation is heavy but consistent, and there is no hidden curve-fitting or circular parameter estimation.\n\nNow the soft spots, in proportion. The biggest one is the IDM backward-travel claim. The paper linearizes around the stopped equilibrium, finds a stable spiral, and concludes that speed becomes negative \"regardless of the initial conditions.\" That does not follow from a local linear analysis. The linearization proves sign changes for trajectories that enter a sufficiently small neighborhood of the equilibrium; it says nothing about trajectories that start far away unless you add a global argument. The numerical replication of the Treiber et al. scenario does show backward travel in a realistic case, so the practical conclusion survives, but the universal claim is unsupported and should be softened or proven.\n\nThe abstract says \"numerical simulations and empirical observations validate the theoretical insights.\" There are no empirical observations in this paper. That sentence needs to go or be backed with data. Minor but irritating.\n\nAlso, the BDA-Newell numerical example has an internal inconsistency: it assigns beta = 2 m/s^2, but the stopping-distance formula and the reported stop time of 18.0 s only make sense with beta = 1.67 m/s^2. This looks like a typo, but it is exactly the kind of thing a referee will catch.\n\nThe normative principles themselves are asserted rather than empirically grounded. The parameter values come from ISO 15622 and Treiber et al., which is defensible, but if human drivers brake later or use different comfort bounds, some of the critique loses force. The paper acknowledges this in passing but does not engage with it seriously.\n\nOverall, the central framework holds up. The paper is aimed at traffic-flow theorists and ACC/ADAS control designers. It is not a field-shifter, but it is a solid, citable analysis with several genuinely useful observations. It deserves a serious referee, not a desk reject. My recommendation: send it to review, and ask the authors to fix the global IDM overclaim, correct the abstract, and sort out the BDA-Newell numbers.","headline":"A useful phase-plane audit of car-following models, but the IDM backward-travel claim overreaches from a local linearization to a global conclusion.","tokens_in":16261,"tokens_out":3677,"would_cite":true,"duration_ms":38287,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Standard car-following models fail safe and human-like tests","keywords":["safe driving principles","human-like car-following","speed-spacing phase plane","Newell simplified car-following model","Intelligent Driver Model","Gipps model","bounded acceleration and deceleration","stationary lead-vehicle problem"],"falsifier":"Concretely, simulate the Intelligent Driver Model with the paper's parameters approaching a stationary leader from free-flow speed; if speed never becomes negative and braking begins no earlier than the safe stopping distance, the paper's two IDM claims are wrong. Similarly, if any parameter set within the stated comfort ranges lets the BDA-Newell model stop at a spacing above the minimum jam spacing without moving backward, the claim that bounded deceleration breaks Newell's model fails.","tokens_in":1773,"feed_emoji":"🚗","tokens_out":3235,"duration_ms":119229,"temperature":0.7,"pith_summary":"This paper builds a multi-phase dynamical-systems test for car-following models and uses it to ask whether standard models can be both provably safe and human-like. The answer it defends is no: the Intelligent Driver Model brakes far too early near a stopped leader and then briefly moves backward; Newell's model, when given the comfort deceleration bound, can drive through the stop and reverse; and the Gipps model loses meaning for spacings between the minimum and comfort jam gaps. The test is a phase portrait in the speed-spacing plane for the stationary lead-vehicle problem, and it shows exactly which safety or human-likeness constraint each model drops. A companion paper uses the same phase analysis to build a model that meets all the principles.","feed_headline":"Standard car-following models fail safe and human-like tests","feed_subtitle":"Phase-plane analysis shows the Intelligent Driver Model travels backward, Gipps fails on some spacings, and Newell extensions can collide.","key_machinery":"The central machinery is the speed-spacing phase plane, $(v,z)$, applied to the stationary lead-vehicle problem. A model's update rule is seen as a min/max choice among constraints, and each active constraint defines a phase: bounded acceleration, equilibrium cruising, equilibrium acceleration, equilibrium deceleration, or bounded deceleration. Drawing the phase portrait and the vector field shows whether trajectories stay above the comfort jam spacing $\\zeta$ and minimum jam spacing $\\zeta'$, whether speed stays nonnegative, and whether the braking profile respects the safe stopping distance $B=v\\tau' + v^2/(2\\beta)$. This turns each model's safety behavior into a geometric property of its phase portrait.","core_discovery":"On its own terms, the discovery is that the stationary lead-vehicle problem, in which one vehicle approaches a leader that never moves, is a sharp discriminating test for car-following models. In the speed-spacing plane, the model's update rule splits into phases (bounded acceleration, equilibrium cruising, equilibrium acceleration, equilibrium deceleration, and bounded deceleration), and the phase portrait determines whether the trajectory respects the comfort jam spacing, the minimum jam spacing, and forward travel. For the Intelligent Driver Model, linearization around the stopped equilibrium gives a stable spiral, so the speed crosses below zero for every initial condition, and at free-flow speed the model brakes immediately even from arbitrarily long distances, giving stopping distances roughly 2.7 times the safe stopping distance. For the bounded-acceleration-and-deceleration extension of Newell's model (BDA-Newell), the comfort deceleration bound makes the follower overshoot the jam spacing unless the bound is raised to about 9.375 m/s², above the comfort range, so the model can violate the minimum jam spacing and forward-travel principles. For the Gipps model, the safe-stopping derivation keeps speed nonnegative and deceleration bounded, but the model is undefined for spacings between the comfort and minimum jam gaps, and aligning its fundamental diagram with the safe-stopping reaction time creates an inconsistent wave speed. The paper concludes that no model in this family simultaneously satisfies all the stated principles.","pith_inferences":["The same phase-plane test could be applied to other standard car-following models, and hidden violations of forward travel or braking distance are likely to appear wherever the model's update rule leaves one phase unconstrained.","The framework suggests a practical acceptance test for adaptive cruise control: record the phase trajectory during an emergency approach, and reject any model whose trajectory enters the negative-speed region or begins braking earlier than the safe stopping distance.","The empirical parameter values matter: if typical drivers comfortably use decelerations above 1.67 m/s², the 'human-like' bound could be relaxed and the BDA-Newell collision example would be mitigated, but then the model's safety margin shrinks.","The stationary-lead-vehicle analysis is the most conservative safety test; extending it to moving leaders that brake would make the safe stopping distance a function of the leader's speed and deceleration as well."],"forward_implications":["A car-following model intended for automated driving should be checked in the speed-spacing plane for the stationary lead-vehicle problem before deployment; the phase portrait shows whether it can collide or reverse direction.","Adaptive cruise control systems built on the Intelligent Driver Model will inherit its early braking and backward-creep behavior, which can create dilemma-zone conflicts at signalized intersections designed for human braking distances.","Adding a bounded-deceleration constraint to Newell's model is not enough by itself: the constraint must be coordinated with the jam-spacing and forward-travel principles, or the model stops too late and reverses.","The Gipps model's safe-stopping formula cannot be combined with a comfort jam spacing and a standard fundamental diagram without either ill-defined states or inconsistent shock-wave speeds.","A model that provably satisfies the paper's principles must manage braking as a multi-phase process rather than a single formula; the companion paper develops such a model."],"supporting_citations":[{"why":"Defines the Intelligent Driver Model and supplies the parameter values used throughout the paper's simulations and phase analysis.","marker":"Treiber et al., 2000"},{"why":"Derives the Gipps model from the safe stopping distance principle; the paper analyzes a simplified form of it.","marker":"Gipps, 1981"},{"why":"Introduces the simplified car-following model that the paper derives from its zeroth- and first-order principles.","marker":"Newell, 2002"},{"why":"Provides the bounded-acceleration formulation whose bounded-deceleration extension the paper tests and finds unsafe.","marker":"Jin and Laval, 2018"},{"why":"Earlier identification of IDM limitations, including backward travel at spacings below the comfort jam spacing, which the paper sharpens to all initial conditions.","marker":"Albeaik et al., 2022"},{"why":"Source of the reaction time and the safe-stopping-distance reasoning behind the paper's safety principles.","marker":"Gazis et al., 1960"},{"why":"Reference for the IDM's braking strategy and for the discrete simplified Gipps model used as a comparison baseline.","marker":"Treiber and Kesting, 2013"},{"why":"Documents the ISO 15622 parameter values adopted for comfort spacing, time gap, and acceleration/deceleration bounds.","marker":"Hiraoka et al., 2005"}],"fun_headline_variants":["Phase-plane test exposes car-following model flaws","Car-following models flunk stationary-leader test","IDM goes backward, Gipps undefined, Newell collides","No standard car-following model meets all safety principles"],"cache_read_input_tokens":18432,"weakest_assumption_plain":"The evaluation treats the paper's stated behavioral principles, with parameter values from ISO 15622 and the IDM literature (comfort spacing 7 m, minimum spacing 5 m, time gap 1.6 s, comfort deceleration 1.67 m/s²), as the definition of safe and human-like driving; if typical drivers brake later than the safe-stopping formula $B=v\\tau' + v^2/(2\\beta)$ or accept different comfort bounds, the IDM and Gipps critiques lose force.","fun_headline_variants_meta":{"raw":{"variants":["Phase-plane test exposes car-following model flaws","Car-following models flunk stationary-leader test","IDM goes backward, Gipps undefined, Newell collides","No standard car-following model meets all safety principles"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000611,"raw_usage":{"total_tokens":2942,"prompt_tokens":1144,"completion_tokens":1798,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":760,"completion_tokens_details":{"reasoning_tokens":1730}},"tokens_in":760,"tokens_out":1798,"duration_ms":13928,"temperature":1.0,"reasoning_tokens":1730,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T21:19:48.396653+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Concretely, simulate the Intelligent Driver Model with the paper's parameters approaching a stationary leader from free-flow speed; if speed never becomes negative and braking begins no earlier than the safe stopping distance, the paper's two IDM claims are wrong. Similarly, if any parameter set within the stated comfort ranges lets the BDA-Newell model stop at a spacing above the minimum jam spacing without moving backward, the claim that bounded deceleration breaks Newell's model fails.","supporting_citations":[{"cited_title":", author Hennecke, A","cited_arxiv_id":null,"evidence_quote":"Defines the Intelligent Driver Model and supplies the parameter values used throughout the paper's simulations and phase analysis."},{"cited_title":", year 1981","cited_arxiv_id":null,"evidence_quote":"Derives the Gipps model from the safe stopping distance principle; the paper analyzes a simplified form of it."},{"cited_title":", year 2002","cited_arxiv_id":null,"evidence_quote":"Introduces the simplified car-following model that the paper derives from its zeroth- and first-order principles."},{"cited_title":", author Laval, J","cited_arxiv_id":null,"evidence_quote":"Provides the bounded-acceleration formulation whose bounded-deceleration extension the paper tests and finds unsafe."},{"cited_title":", author Bayen, A","cited_arxiv_id":null,"evidence_quote":"Earlier identification of IDM limitations, including backward travel at spacings below the comfort jam spacing, which the paper sharpens to all initial conditions."},{"cited_title":", author Herman, R","cited_arxiv_id":null,"evidence_quote":"Source of the reaction time and the safe-stopping-distance reasoning behind the paper's safety principles."},{"cited_title":", author Kesting, A","cited_arxiv_id":null,"evidence_quote":"Reference for the IDM's braking strategy and for the discrete simplified Gipps model used as a comparison baseline."},{"cited_title":", author Kunimatsu, T","cited_arxiv_id":null,"evidence_quote":"Documents the ISO 15622 parameter values adopted for comfort spacing, time gap, and acceleration/deceleration bounds."}],"review_version":1}