pith. machine review for the scientific record. sign in

arxiv: 2605.09376 · v1 · submitted 2026-05-10 · 💻 cs.RO

Recognition: no theorem link

Mismatch-Aware Adaptive Constraint Tightening for Bicycle-Model Trajectory Optimization

Lingxue Lyu , Zihui Liu
Authors on Pith no claims yet

Pith reviewed 2026-05-12 04:49 UTC · model grok-4.3

classification 💻 cs.RO
keywords model mismatchadaptive constraint tighteningbicycle modeltrajectory optimizationautonomous vehiclessafety marginsmodel predictive controlvehicle dynamics
0
0 comments X

The pith

An adaptive safety margin that scales with speed squared and curvature guarantees full safety under model mismatch while using 84 percent less margin than fixed worst-case approaches.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that the kinematic bicycle model used for fast trajectory planning deviates from true dynamic behavior due to tire slip and yaw coupling, producing outward path errors that grow with velocity and turn sharpness. It derives a characteristic speed separating safe inward and critical outward mismatch regimes, then proves the maximum outward deviation follows a quadratic time-horizon scaling whose coefficient depends only on vehicle parameters and the planning horizon. From this it constructs a state-dependent tightening margin that stays near zero on straight or slow paths yet expands exactly enough for high-speed turns. Numerical tests on both linear and nonlinear bicycle models confirm the scaling, and closed-loop MPC comparisons show the method preserves 100 percent safety while cutting unnecessary margin substantially versus constant-margin or tube-MPC baselines.

Core claim

The peak outward deviation caused by dynamic mismatch follows a T-squared horizon scaling whose analytical coefficient transitions between transient and steady-state bounds and is given by one-half times (one minus the square of the ratio of characteristic speed to maximum speed) times the horizon squared. Multiplying this coefficient by velocity squared and absolute curvature produces the required safety margin in the proposed Mismatch-Aware Adaptive Constraint Tightening method, replacing any fixed worst-case value with a state-dependent quantity.

What carries the argument

Mismatch-Aware Adaptive Constraint Tightening (MACT) defined by the state-dependent margin ε(v, κ) = a₂ v² |κ|, where a₂ is the closed-form coefficient computed from vehicle parameters and planning horizon alone.

Load-bearing premise

The derived scaling law for peak outward deviation and the resulting analytical coefficient hold only under the linear tire model of the dynamic bicycle without unmodeled disturbances, nonlinear tire effects, or actuator limits that would change the mismatch regimes.

What would settle it

A measurement on a physical vehicle showing that actual outward deviation exceeds the predicted a₂ v² |κ| value at a chosen speed and curvature, or a trajectory that violates safety constraints despite using the MACT margin, would falsify the central claim.

Figures

Figures reproduced from arXiv: 2605.09376 by Lingxue Lyu, Zihui Liu.

Figure 1
Figure 1. Figure 1: At v=15 m/s and κ=0.015 rad/m, the dynamic model drifts 1.04 m outward from the kinematic plan in 1.5 s. 12 13 14 15 16 17 18 Speed v [m/s] 0 50 100 150 200 250 300 Max outward deviation [cm] Speed scaling law ( =0.015 rad/m) Numerical (dynamic model) Transient bound: 1 2 (v 2 v 2 c ) T 2 (Prop.~3) MACT safe bound: a anal 2 v 2 (a anal 2 =0.62, Cor.~2) vc =12 m/s [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Max outward deviation vs. speed (κ=0.015 rad/m). Dashed: transient bound 1 2 (v 2−v 2 c )κT 2 ; dash-dot: MACT safe bound a anal 2 v 2κ, which stays above the data over the whole range. outward. MACT with a2=0.404 yields ϵ=1.36 m, which gives a valid safety certificate. Exp. 2 – Speed scaling. ( [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 5
Figure 5. Figure 5: Per-scenario margin: MACT follows ε ∗ closely while the fixed margin overshoots at low speed/curvature. No margin Fixed margin MACT (ours) 0 20 40 60 80 100 120 140 Mean wasted margin [cm] 0% safe 0.0 cm 100% safe 119.0 cm 100% safe 18.8 cm Conservatism (lower is better) [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Mean wasted margin across the 20-scenario grid. Both MACT and fixed margin are 100% safe; MACT wastes 84% less. ics with state (ϕ, ϕ, δ ˙ ) and yaw ψ˙=v tan δ/(l cos ϕ). A balance controller ˙δ=K1(ϕ−ϕref)+K2ϕ˙+K3(δ−δtgt) with [K1, K2, K3]=[71, 21, −20] [9] stabilizes lean around the equilibrium ϕref= arctan(v 2κ/g). The bicycle starts upright and has to build lean before reaching the commanded curvature; d… view at source ↗
Figure 10
Figure 10. Figure 10: Per-scenario applied ε¯. Tube is flat at 5.8 cm; MACT scales as a cl 2 v 2κ and uses only 2.3–5.8 cm. 10 0 10 Cross-tra c k n [c m] No margin Tube (worst-case) Adaptive MACT (ours) Lane ±16 cm 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Time [s] 0 2 4 6 A p plie d [c m] [PITH_FULL_IMAGE:figures/full_fig_p006_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Closed-loop trajectory at v=15 m/s, κ=0.012 rad/m. Top: cross-track n(t); all methods stay within ±LANEhw. Bottom: applied ε(t). Tube is flat; adaptive ramps up slowly (misses the initial transient window); MACT is at its correct scenario-specific level from the first step. only 4.1 cm, which is smaller than the 5.3 cm drift that was actually observed — a safety shortfall that does not show up here only b… view at source ↗
Figure 12
Figure 12. Figure 12: Grid-averaged ε¯. MACT saves 34% vs. tube while keeping the same safety. The vc result. The characteristic speed isolates a direc￾tional transition in the initial response of the bicycle model. Prior work on bicycle stability [6] characterizes the self￾balancing limit but not this initial-response transition. The vc threshold immediately identifies the safety-critical regime and gives the sign of the requ… view at source ↗
read the original abstract

Trajectory optimization for autonomous vehicles usually relies on the kinematic bicycle model because of its computational simplicity. However, when the planned trajectory is executed under the true vehicle dynamics, which include lateral slip, tire stiffness and yaw-lateral coupling, safety constraints can be violated owing to the model mismatch. In this paper, we make three theoretical contributions. First, we derive a characteristic speed $v_c=\sqrt{C_\alpha L/M}$ which separates two different mismatch regimes: below $v_c$ the dynamic bicycle initially oversteers inward (safe); above $v_c$ it understeers outward (safety-critical). Second, we prove that the peak outward deviation $\varepsilon^*$ follows a $T^2$ horizon scaling whose coefficient transitions between a transient bound $\frac{1}{2}(v^2-v_c^2)\kappa$ and a steady-state bound. Third, we obtain a simulation-free analytical coefficient $a_2^{\mathrm{anal}}=\frac{1}{2}(1-v_c^2/v_{\max}^2)T^2$ that is computable from vehicle parameters and the planning horizon alone. Putting these together, we propose Mismatch-Aware Adaptive Constraint Tightening (MACT), $\epsilon(v,\kappa)=a_2 v^2|\kappa|$, which replaces a fixed worst-case margin by a state-dependent one that is large at high speed/curvature but nearly zero on gentle paths. Eight numerical experiments confirm the scaling laws. MACT reaches 100% safety with 84% less wasted margin than a fixed-margin baseline on the 2-DOF vehicle, extends to a nonlinear leaning bicycle, and in a closed-loop direct-shooting MPC comparison it cuts the applied margin by 34% compared with tube MPC while keeping the same safety.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The manuscript proposes Mismatch-Aware Adaptive Constraint Tightening (MACT) for bicycle-model trajectory optimization to handle model mismatch between kinematic planning and dynamic execution. It introduces a characteristic speed vc = sqrt(C_α L / M) to separate mismatch regimes, proves that the peak outward deviation ε* scales with T² with an analytical coefficient a2^anal computable from vehicle parameters, and defines the adaptive margin ε(v, κ) = a2 v² |κ|. Eight numerical experiments validate the scaling laws and demonstrate that MACT achieves 100% safety with 84% less wasted margin than fixed-margin baselines and 34% less margin than tube MPC in closed-loop comparisons, while extending to a nonlinear leaning bicycle model.

Significance. If the derivations hold under the stated assumptions, the work offers a meaningful contribution by replacing fixed worst-case margins with a state-dependent tightening that is analytically derived from vehicle parameters (Cα, L, M) and planning quantities (T, vmax) without post-hoc fitting. The simulation-free nature of a2^anal, the regime separation at vc, and the eight experiments confirming the predicted T² scaling are clear strengths that support more efficient yet safe trajectory optimization. The closed-loop MPC comparison and extension to a nonlinear model further indicate practical relevance, though applicability remains tied to the linear tire regime.

major comments (1)
  1. The three theoretical contributions (vc regime separation, T² scaling of ε*, and a2^anal = ½(1−vc²/vmax²)T²) are derived under the linear tire model assumptions of the dynamic bicycle model without unmodeled disturbances or actuator limits. While the abstract notes extension to a nonlinear leaning bicycle, the eight experiments and MPC comparison remain inside the linear model class; this makes the conservative T² coefficient and 100% safety claim load-bearing only within those assumptions, and a dedicated discussion of sensitivity to nonlinear tire forces or disturbances is needed to support broader claims.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback. The comment correctly identifies the scope of our assumptions and validation; we address it directly below and will incorporate the requested discussion.

read point-by-point responses

  1. Referee: The three theoretical contributions (vc regime separation, T² scaling of ε*, and a2^anal = ½(1−vc²/vmax²)T²) are derived under the linear tire model assumptions of the dynamic bicycle model without unmodeled disturbances or actuator limits. While the abstract notes extension to a nonlinear leaning bicycle, the eight experiments and MPC comparison remain inside the linear model class; this makes the conservative T² coefficient and 100% safety claim load-bearing only within those assumptions, and a dedicated discussion of sensitivity to nonlinear tire forces or disturbances is needed to support broader claims.

    Authors: We agree that the derivations of vc, the T² scaling, and a2^anal, together with the closed-loop MPC comparison and seven of the eight experiments, are performed under the linear tire model without disturbances or actuator limits. Experiment 8 applies the MACT margin (computed from the linear analysis) to a nonlinear leaning bicycle model and reports maintained safety, but this single experiment does not constitute a full sensitivity study. We will add a dedicated subsection in the Discussion that (i) analytically examines how moderate tire-force saturation perturbs the T² coefficient, (ii) presents additional simulation results with additive lateral disturbances, and (iii) explicitly states actuator limits and severe nonlinearity as current limitations of the 100% safety guarantee. This revision will qualify the broader applicability claims without altering the core contributions. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivations are first-principles from the linear bicycle model

full rationale

The three theoretical contributions (vc separation, T² scaling of ε*, and simulation-free a2^anal) are obtained by direct algebraic manipulation of the linear tire-force equations of the dynamic bicycle model, using only the stated parameters Cα, L, M, T, and vmax. The MACT formula follows immediately from these bounds without any post-hoc fitting or relabeling of inputs as outputs. No self-citations support the core claims, no uniqueness theorems are imported, and the eight experiments serve as independent numerical checks rather than the source of the coefficient. The derivation chain is therefore self-contained under the model's assumptions.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard vehicle dynamics models and the derivation of mismatch regimes from them; no new entities are postulated and no free parameters are fitted beyond standard planning inputs.

axioms (1)
  • domain assumption The dynamic bicycle model with linear tire forces accurately captures the true vehicle dynamics mismatch including lateral slip, tire stiffness, and yaw-lateral coupling.
    Invoked to derive the characteristic speed vc and the T^2 deviation bounds separating safe and safety-critical regimes.

pith-pipeline@v0.9.0 · 5654 in / 1358 out tokens · 49266 ms · 2026-05-12T04:49:50.778021+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages

  1. [1]

    Path planning for autonomous vehicles using model predictive control,

    C. Liu, S. Lee, S. Varnhagen, and H. E. Tseng, “Path planning for autonomous vehicles using model predictive control,” inIEEE Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 174–179

    work page 2017

  2. [2]

    Generation of minimum-snap trajectories for quadrotors by solving quadratic pro- grams,

    F. Augugliaro, A. P. Schoellig, and R. D’Andrea, “Generation of minimum-snap trajectories for quadrotors by solving quadratic pro- grams,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2012, pp. 1257–1263

    work page 2012

  3. [3]

    Trajectory planning for bertha—a local, continuous method,

    J. Ziegler, P. Bender, T. Dang, and C. Stiller, “Trajectory planning for bertha—a local, continuous method,”IEEE Intelligent Vehicles Symposium (IV), pp. 450–457, 2014

    work page 2014

  4. [4]

    A survey of motion planning and control techniques for self-driving urban vehicles,

    B. Paden, M. Cap, S. Z. Yong, D. Yershov, and E. Frazzoli, “A survey of motion planning and control techniques for self-driving urban vehicles,”IEEE Transactions on intelligent vehicles, vol. 1, no. 1, pp. 33–55, 2016

    work page 2016

  5. [5]

    Rajamani,Vehicle Dynamics and Control

    R. Rajamani,Vehicle Dynamics and Control. Springer Science & Business Media, 2011

    work page 2011

  6. [6]

    Towards a maximally-robust self-balancing bicycle without reaction-moment gy- roscopes or reaction wheels,

    A. M. Sharma, S. Wang, Y . Zhou, and A. Ruina, “Towards a maximally-robust self-balancing bicycle without reaction-moment gy- roscopes or reaction wheels,” inBicycle and Motorcycle Dynamics Symposium, 2016

    work page 2016

  7. [7]

    Robust model predictive control using tubes,

    W. Langson, I. Chryssochoos, S. V . Rakovic, and D. Q. Mayne, “Robust model predictive control using tubes,” inAutomatica, vol. 40, no. 1. Elsevier, 2004, pp. 125–133

    work page 2004

  8. [8]

    Robust model predic- tive control of constrained linear systems with bounded disturbances,

    D. Q. Mayne, M. M. Seron, and S. V . Rakovi ´c, “Robust model predic- tive control of constrained linear systems with bounded disturbances,” Automatica, vol. 41, no. 2, pp. 219–224, 2005

    work page 2005

  9. [9]

    Dynamic model derivation and controller design for an autonomous bicycle,

    S. Wang, “Dynamic model derivation and controller design for an autonomous bicycle,” Cornell University, MEng Report, 2014

    work page 2014

  10. [10]

    Borrelli, A

    F. Borrelli, A. Bemporad, and M. Morari,Predictive Control for Linear and Hybrid Systems. Cambridge University Press, 2017

    work page 2017

  11. [11]

    H. B. Pacejka,Tire and Vehicle Dynamics, 3rd ed. Elsevier, 2012

    work page 2012

  12. [12]

    Efficient online calibration for autonomous vehicle’s longitudinal dynamical system: A Gaussian model approach,

    S. Wanget al., “Efficient online calibration for autonomous vehicle’s longitudinal dynamical system: A Gaussian model approach,” inIEEE International Conference on Robotics and Automation (ICRA). Xi’an, China: IEEE, 2021

    work page 2021

  13. [13]

    Control barrier functions: Theory and applications,

    A. D. Ames, S. Coogan, M. Egerstedt, G. Notomista, K. Sreenath, and P. Tabuada, “Control barrier functions: Theory and applications,” European Control Conference (ECC), pp. 3420–3431, 2019

    work page 2019

  14. [14]

    Kinematic and dynamic vehicle models for autonomous driving control design,

    J. Kong, M. Pfeiffer, G. Schildbach, and F. Borrelli, “Kinematic and dynamic vehicle models for autonomous driving control design,” in IEEE Intelligent Vehicles Symposium (IV). IEEE, 2015, pp. 1094– 1099

    work page 2015

  15. [15]

    Real-time stabilization of a falling humanoid robot using hand contact: An optimal control approach,

    S. Wang and K. Hauser, “Real-time stabilization of a falling humanoid robot using hand contact: An optimal control approach,” inIEEE-RAS International Conference on Humanoid Robots (Humanoids). IEEE, 2017

    work page 2017

  16. [16]

    Control-limited differential dynamic programming,

    Y . Tassa, N. Mansard, and E. Todorov, “Control-limited differential dynamic programming,”IEEE International Conference on Robotics and Automation (ICRA), pp. 1168–1175, 2014

    work page 2014

  17. [17]

    Learning-based model predictive control: Toward safe learning in control,

    L. Hewing, K. P. Wabersich, M. Menner, and M. N. Zeilinger, “Learning-based model predictive control: Toward safe learning in control,”Annual Review of Control, Robotics, and Autonomous Sys- tems, vol. 3, pp. 269–296, 2020

    work page 2020

  18. [18]

    Unified multi-contact fall mitigation planning for humanoids,

    S. Wang and K. Hauser, “Unified multi-contact fall mitigation planning for humanoids,” inIEEE-RAS International Conference on Humanoid Robots (Humanoids). IEEE, 2018

    work page 2018

  19. [19]

    A fast, on-line corridor-constrained trajectory planner for autonomous ground vehicles,

    D. Zhouet al., “A fast, on-line corridor-constrained trajectory planner for autonomous ground vehicles,” inIEEE Intelligent Vehicles Sym- posium (IV). IEEE, 2017

    work page 2017

  20. [20]

    Real-time trajectory planning for autonomous urban driving: Framework, algorithms, and verifications,

    X. Li, Z. Sun, D. Cao, Z. He, and Q. Zhu, “Real-time trajectory planning for autonomous urban driving: Framework, algorithms, and verifications,” inIEEE/ASME Transactions on Mechatronics, vol. 21, no. 2. IEEE, 2016, pp. 740–753

    work page 2016

  21. [21]

    Realization of a real-time optimal control strategy to stabilize a falling humanoid robot with hand contact,

    S. Wang and K. Hauser, “Realization of a real-time optimal control strategy to stabilize a falling humanoid robot with hand contact,” in IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2018

    work page 2018