pith. sign in

arxiv: 2605.01431 · v1 · submitted 2026-05-02 · 📡 eess.SY · cs.SY

Point-to-Cloud NMPC with Smooth Avoidance Constraints

Brener G. Ferreira , Vinicius M. Gon\c{c}alves , Marcelo A. Santos , Guilherme V. Raffo This is my paper

Pith reviewed 2026-05-09 18:16 UTC · model grok-4.3

classification 📡 eess.SY cs.SY
keywords nonlinear model predictive controlobstacle avoidancecontrol barrier functionspoint-to-cloud distanceset-point trackingaerial robotsmooth constraintsnonconvex geometries
0
0 comments X

The pith

A smooth point-to-cloud distance metric lets NMPC enforce differentiable safety constraints for reliable avoidance in nonconvex spaces while tracking changing set-points.

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

This paper develops a nonlinear model predictive control approach for set-point tracking that builds in obstacle avoidance. It relies on a specially constructed distance measure from the controlled point to a cloud of obstacle points that stays smooth and differentiable even when the obstacles have complicated shapes. This property lets the designer add safety rules through control barrier functions without breaking the math the optimizer needs. Stationary artificial variables keep the problem solvable when the target moves. A reader would care because many real robots must navigate messy environments without crashing or getting the optimizer stuck.

Core claim

The paper presents a finite-horizon NMPC formulation for set-point tracking that incorporates avoidance by means of a smooth point-to-cloud distance metric. This metric produces continuously differentiable and numerically well-conditioned gradients for arbitrary nonconvex geometries, allowing safety constraints to be expressed consistently through control barrier functions. Stationary artificial variables are added to the optimal control problem to preserve feasibility when set-points change. Closed-loop simulations of an aerial robot confirm accurate tracking together with smooth avoidance behavior in complex settings.

What carries the argument

The smooth point-to-cloud distance metric, which supplies a continuously differentiable approximation of clearance to the nearest point in an obstacle cloud and thereby supports gradient-based safety constraints.

If this is right

  • Safety constraints can be written consistently and differentiably using control barrier functions.
  • The closed-loop system produces reliable avoidance behavior even when obstacle geometry is nonconvex.
  • Feasibility of the optimization is maintained when set-points change by introducing stationary artificial variables.
  • Numerical tests on an aerial robot demonstrate both accurate set-point tracking and smooth obstacle avoidance.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same distance construction could be paired with real-time sensor point clouds to handle partially observed environments.
  • Similar smoothness techniques might transfer to other optimization-based motion planners that currently rely on nonsmooth clearance functions.
  • Hardware experiments with time-varying obstacles would test whether the differentiability continues to support fast re-planning.

Load-bearing premise

The smooth point-to-cloud distance metric stays numerically stable and actually guarantees avoidance for every possible nonconvex obstacle shape without creating new local minima or solver problems.

What would settle it

A simulation run on a highly nonconvex obstacle geometry in which the NMPC either collides, the solver fails to converge, or the computed gradients become ill-conditioned would show the claim is false.

Figures

Figures reproduced from arXiv: 2605.01431 by Brener G. Ferreira, Guilherme V. Raffo, Marcelo A. Santos, Vinicius M. Gon\c{c}alves.

Figure 1
Figure 1. Figure 1: UAV flight trajectory (black line) to safely execute a task view at source ↗
Figure 2
Figure 2. Figure 2: Projection of the trajectory onto the XZ-plane. view at source ↗
Figure 3
Figure 3. Figure 3: Projection of the trajectory onto the XY-plane. view at source ↗
Figure 4
Figure 4. Figure 4: Time evolution of the thrust lift forces applied to the quadrotor view at source ↗
Figure 5
Figure 5. Figure 5: Absolute output tracking error, illustrating the relationship between view at source ↗
Figure 8
Figure 8. Figure 8: Control inputs applied to the propellers, exhibiting high-frequency view at source ↗
read the original abstract

This paper proposes a finite-horizon optimal control strategy for set-point tracking using a nonlinear model predictive control framework with integrated avoidance capabilities. The formulation employs a smooth point-to-cloud distance metric that ensures continuously differentiable and numerically well-conditioned gradients, even in the presence of regions with complex and nonconvex geometries. This smoothness allows safety constraints to be formulated consistently and differentiably through control barrier functions, resulting in a reliable avoidance behavior for the closed-loop system. Additionally, stationary artificial variables are introduced in the optimal control problem to preserve feasibility under changing set-points. The proposed approach is validated through numerical experiments of an aerial robot, demonstrating accurate tracking and smooth obstacle avoidance in complex environments.

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

2 major / 2 minor

Summary. The paper proposes a finite-horizon NMPC for set-point tracking that integrates avoidance via a smooth point-to-cloud distance metric. The metric is constructed to be continuously differentiable with well-conditioned gradients, enabling its use inside CBF-based safety constraints even for nonconvex geometries. Stationary artificial variables are added to the OCP to maintain feasibility when set-points change. The approach is demonstrated via numerical experiments on an aerial robot performing tracking and avoidance in complex environments.

Significance. If the smoothness property and closed-loop avoidance can be shown to hold with quantifiable margins, the method would offer a practical route to differentiable, nonconvex obstacle avoidance inside NMPC without resorting to non-smooth min-distance formulations that degrade solver conditioning.

major comments (2)
  1. [Abstract] Abstract: the claim that the smooth metric 'results in a reliable avoidance behavior for the closed-loop system' is not supported by any error analysis. For standard smooth approximations (log-sum-exp, softmin, etc.) the sublevel set {x | d_smooth(x) > 0} strictly contains the true safe set; a feasible NMPC solution can therefore satisfy the CBF constraint while the robot still intersects the obstacle. No analytic bound on the penetration depth or worst-case distance error for arbitrary point clouds is supplied.
  2. [Numerical validation] The numerical validation section: the reported aerial-robot experiments demonstrate tracking and avoidance on selected scenarios, but provide no quantitative assessment of minimum true distance to the point cloud, no stress tests on dense or highly nonconvex clouds, and no comparison against a non-smooth baseline that would quantify the practical impact of the smoothing error.
minor comments (2)
  1. [Abstract] The abstract states that 'stationary artificial variables are introduced ... to preserve feasibility under changing set-points,' but does not indicate how these variables are incorporated into the cost or constraints or whether they affect the CBF formulation.
  2. [Method] Notation for the smooth distance function and the CBF h(x) should be introduced with explicit definitions of the smoothing parameter and its relation to the true minimum distance.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive comments highlighting the need to qualify claims about avoidance reliability and to strengthen the numerical evidence. We address each major comment below and will incorporate revisions accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the smooth metric 'results in a reliable avoidance behavior for the closed-loop system' is not supported by any error analysis. For standard smooth approximations (log-sum-exp, softmin, etc.) the sublevel set {x | d_smooth(x) > 0} strictly contains the true safe set; a feasible NMPC solution can therefore satisfy the CBF constraint while the robot still intersects the obstacle. No analytic bound on the penetration depth or worst-case distance error for arbitrary point clouds is supplied.

    Authors: We agree that the abstract claim of 'reliable avoidance behavior' lacks supporting error analysis and overstates the guarantee, since the smoothing enlarges the effective safe set. We will revise the abstract to state that the formulation 'enables smooth obstacle avoidance via CBF constraints' without the reliability qualifier. A short discussion will be added noting that the approximation error can be reduced by tuning the smoothing parameter, but we do not supply a general analytic bound on penetration depth for arbitrary point clouds. revision: partial

  2. Referee: [Numerical validation] The numerical validation section: the reported aerial-robot experiments demonstrate tracking and avoidance on selected scenarios, but provide no quantitative assessment of minimum true distance to the point cloud, no stress tests on dense or highly nonconvex clouds, and no comparison against a non-smooth baseline that would quantify the practical impact of the smoothing error.

    Authors: We acknowledge the absence of quantitative minimum-distance metrics and limited scenario coverage. In the revision we will add time-series plots of the true (non-smooth) minimum distance to the point cloud and include additional experiments with denser and more nonconvex clouds. A direct non-smooth baseline comparison is difficult because such formulations frequently cause solver failures or ill-conditioning; we will add a brief discussion of this practical motivation but will not include full comparative runs that prove unreliable. revision: yes

standing simulated objections not resolved
  • Deriving a general analytic bound on penetration depth or worst-case distance error for arbitrary point clouds without additional assumptions on point density and geometry.

Circularity Check

0 steps flagged

No circularity: new metric construction with external validation

full rationale

The paper defines a smooth point-to-cloud distance metric explicitly to achieve continuous differentiability, then incorporates it into CBF-based safety constraints inside an NMPC formulation, and validates the closed-loop behavior via numerical experiments on aerial-robot scenarios. No load-bearing step reduces by construction to a fitted parameter, a self-referential definition, or an unverified self-citation chain; the derivation chain is self-contained and relies on the explicit functional construction plus independent simulation results rather than tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard nonlinear optimal control assumptions plus the novel smoothness property of the distance metric; no free parameters or invented entities are mentioned in the abstract.

axioms (2)
  • domain assumption The robot dynamics are known and the system is controllable under the stated constraints.
    Implicit in any NMPC formulation for set-point tracking.
  • domain assumption Control barrier functions can be used to enforce safety when the distance metric is continuously differentiable.
    Standard CBF theory invoked to justify the avoidance constraints.

pith-pipeline@v0.9.0 · 5420 in / 1326 out tokens · 51293 ms · 2026-05-09T18:16:08.989972+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

16 extracted references · 16 canonical work pages

  1. [1]

    E. F. Camacho, C. Bordons, E. F. Camacho, and C. Bordons,Model predictive controllers. Springer, 2007

    work page 2007

  2. [2]

    Nonlinear mpc for tracking piece-wise constant reference signals,

    D. Limon, A. Ferramosca, I. Alvarado, and T. Alamo, “Nonlinear mpc for tracking piece-wise constant reference signals,”IEEE Transactions on Automatic Control, vol. 63, no. 11, pp. 3735–3750, 2018

    work page 2018

  3. [3]

    The distance function in the presence of an obstacle,

    P. Albano, V . Basco, and P. Cannarsa, “The distance function in the presence of an obstacle,”Calculus of Variations and Partial Differential Equations, vol. 61, no. 13, 2022

    work page 2022

  4. [4]

    Constrained model predictive control: Stability and optimality,

    D. Q. Mayne, J. B. Rawlings, C. V . Rao, and P. O. Scokaert, “Constrained model predictive control: Stability and optimality,”Au- tomatica, vol. 36, no. 6, pp. 789–814, 2000

    work page 2000

  5. [5]

    A penalty method based approach for autonomous navigation using nonlinear model predictive control,

    B. Hermans, P. Patrinos, and G. Pipeleers, “A penalty method based approach for autonomous navigation using nonlinear model predictive control,”IFAC-PapersOnLine, vol. 51, no. 20, pp. 234–240, 2018

    work page 2018

  6. [6]

    Nonlinear model predictive control schemes for obstacle avoidance,

    M. A. Santos, A. Ferramosca, and G. V . Raffo, “Nonlinear model predictive control schemes for obstacle avoidance,”Journal of Control, Automation and Electrical Systems, vol. 34, no. 5, pp. 891–906, 2023

    work page 2023

  7. [7]

    Control barrier function based quadratic programs for safety critical systems,

    A. D. Ames, X. Xu, J. W. Grizzle, and P. Tabuada, “Control barrier function based quadratic programs for safety critical systems,”IEEE Transactions on Automatic Control, vol. 62, no. 8, pp. 3861–3876, 2016

    work page 2016

  8. [8]

    Efficient trajectory planning and control for quadrotor uavs in cluttered environments,

    W. Zhang, K. Li, X. Chen, and Y . Wang, “Efficient trajectory planning and control for quadrotor uavs in cluttered environments,”IEEE Transactions on Robotics, vol. 39, no. 4, pp. 1590–1605, 2023

    work page 2023

  9. [9]

    Ex- tended euclidean distance-based model predictive control for safety- critical dynamic obstacle avoidance,

    B. Zhang, J. Pan, Z. Huang, J. Wen, Z. Wu, and B. M. Chen, “Ex- tended euclidean distance-based model predictive control for safety- critical dynamic obstacle avoidance,”IEEE Transactions on Industrial Electronics, pp. 1–11, 2025

    work page 2025

  10. [10]

    Point cloud-based control barrier functions for model predictive control in safety-critical navigation of autonomous mobile robots,

    F. Liang, Y . Yang, and S.-L. Dai, “Point cloud-based control barrier functions for model predictive control in safety-critical navigation of autonomous mobile robots,” inProceedings of the IROS. IEEE, 2025, pp. 2060–2067

    work page 2025

  11. [11]

    Smooth dis- tances for second-order kinematic robot control,

    V . M. Gonc ¸alves, A. Tzes, F. Khorrami, and P. Fraisse, “Smooth dis- tances for second-order kinematic robot control,”IEEE Transactions on Robotics, vol. 40, pp. 2950–2965, 2024

    work page 2024

  12. [12]

    On the stability of a nonlinear MPC scheme for avoidance,

    M. A. Santos, A. Ferramosca, and G. V . Raffo, “On the stability of a nonlinear MPC scheme for avoidance,”IEEE Control Systems Letters, vol. 8, pp. 3285–3290, 2024

    work page 2024

  13. [13]

    Set-point tracking mpc with avoidance features,

    ——, “Set-point tracking mpc with avoidance features,”Automatica, vol. 159, p. 111390, 2024

    work page 2024

  14. [14]

    Motion planning with dynamic obstacles using convexified control barrier functions,

    V . Varun, A. P. Vinod, and S. Kolathaya, “Motion planning with dynamic obstacles using convexified control barrier functions,” in Proceedings of the ICC, 2021, pp. 81–86

    work page 2021

  15. [15]

    Nonlinear model predictive control on se (3) for quadrotor aggressive maneuvers,

    J. C. Pereira, V . J. Leite, and G. V . Raffo, “Nonlinear model predictive control on se (3) for quadrotor aggressive maneuvers,”Journal of Intelligent & Robotic Systems, vol. 101, pp. 1–15, 2021

    work page 2021

  16. [16]

    Algorithm 247: Radical-inverse quasi-random point sequence,

    J. H. Halton, “Algorithm 247: Radical-inverse quasi-random point sequence,”Communications of the ACM, vol. 3, no. 4, pp. 214–215, 1960

    work page 1960