Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Robust Convex Model Predictive Control with collision avoidance guarantees for robot manipulators

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A convex MPC that provably avoids collisions for six-axis robots under model uncertainty

desk verdict Solid convex tube-MPC core with a genuinely useful state/input-dependent tube, but the advertised collision-avoidance guarantee is conditional on an uncertified learned SCDF. read the letter →

arxiv 2508.21677 v3 pith:XLANOHFM submitted 2025-08-29 cs.RO

Bernhard Wullt , Johannes Köhler , Per Mattsson , Mikeal Norrlöf , Thomas B. Schön This is my paper
classification cs.RO
keywords modelpredictivecontrolrobusttubeMPCcollisionavoidanceconfigurationspacesigneddistancefunctionrobotmanipulatorconvexoptimizationuncertainty
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper proposes a model predictive controller that is both convex and robust, aimed at industrial robot manipulators operating in cluttered environments with uncertain dynamics. The key claim is that by coupling a homothetic tube—whose size scales with the current state and input—with collision-free balls from a learned signed configuration distance function, one can guarantee collision avoidance for nonlinear 6-DOF robot dynamics, something the authors say no prior convex MPC has achieved. The resulting optimization is a second-order cone program that solves fast enough for real-time control. In simulation the method reaches goals quicker than a rigid-tube MPC and tolerates larger model uncertainties, while all tested trajectories passed an independent collision check.

What carries the argument

The mechanism carrying the argument is a state-and-input-dependent homothetic tube: an ellipsoid around the nominal predicted trajectory whose radius δ is governed by the inclusion δ⁺=(ρ+dLβ)δ+dβ(x̄,ā), guaranteeing that the true uncertain closed-loop state stays inside the tube despite any θ∈Θ. Paired with this, a learned signed configuration distance function r(q) supplies collision-free Euclidean balls B(c)={q:‖q−c‖≤r(c)} in configuration space; the MPC constraints place each nominal state and its tube inside one such ball, which preserves convexity while certifying obstacle avoidance.

What would settle it

For a fixed obstacle layout, compute the exact SCDF with a conventional collision checker and compare it to the learned network over a dense grid of configurations. If at any configuration r_learned(q) > r_true(q), then constraint (21) can admit a state that is actually in collision; executing the MPC from such a state and observing a collision would falsify the safety claim.

Watch

Extended reading notes

Core claim

The construction starts from feedback linearization, which turns the uncertain manipulator into a double integrator plus a state- and input-dependent model error Δθ(x,a) bounded by β(x,a)=a‖a‖+b‖q̇‖+c. Around the nominal prediction, an ellipsoidal homothetic tube is propagated with radius δ⁺=(ρ+dLβ)δ+dβ(x̄,ā), so the tube is wide only where the error bound demands. Nominal states and tubes are then constrained inside collision-free balls from a learned signed configuration distance function, making obstacle avoidance convex. Theorem 1 proves recursive feasibility, constraint satisfaction, and convergence to the goal; with a corridor planner feeding virtual goals, the MPC navigates non-convex

Load-bearing premise

The learned SCDF r(q) must never over-estimate the true clearance to obstacles; if it does, the ball constraints (21) can certify a collision state and the safety guarantee collapses.

Editorial extensions

If this is right

  • Robust MPC for manipulators becomes a convex second-order cone program, solvable at the few-tens-of-milliseconds scale needed for real-time control of a 6-DOF arm.
  • A flexible tube that shrinks along benign trajectories is less conservative than a fixed-radius tube, producing faster motion and tolerance of larger model uncertainties.
  • The corridor planner lets the convex MPC navigate non-convex obstacle layouts by chaining SCDF balls, so the guarantee extends beyond a single convex free region.
  • Theorem 1 guarantees recursive feasibility, constraint satisfaction, and convergence for any model parameter in the uncertainty set, given a feasible initial problem.
  • Because the SCDF is learned once for a robot and obstacle class, new obstacle placements only require re-querying the network, not re-learning.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The formal guarantee is only as conservative as the learned SCDF: the paper verifies safety empirically with an independent collision checker, but it does not prove that the deep network under-estimates r(q). An adversarial or poorly calibrated SCDF that over-estimates clearance would break the ball-containment argument, so a certification layer for r(q) would turn the empirical claim into a fully
  • The 6-DOF result is achieved by over-approximating the wrist axes with a sphere, effectively reducing the collision problem to 3 DOF; a testable extension is to train the SCDF network directly on the full 6-DOF configuration space and check whether the same guarantee holds.
  • The homothetic-tube-plus-corridor pattern could transfer to other nonlinear uncertain systems whose linearization yields a similar bound β(x,a), such as quadrotors or legged robots, whenever a configuration-space distance function is available.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes a robust model predictive control framework for robot manipulators with collision avoidance. The authors feedback-linearize the uncertain manipulator dynamics, bound the resulting state- and input-dependent model error (Prop. 1), and design a homothetic ellipsoidal tube whose scaling evolves according to Prop. 2. The MPC (18) is a convex second-order cone program. Obstacle avoidance is encoded through configuration-space balls obtained from a signed configuration distance function (SCDF), which is approximated offline by a deep network from prior work [11]; corridor planning assigns balls to the predicted trajectory and selects a virtual goal. The paper claims closed-loop constraint satisfaction, convergence, and collision avoidance guarantees (Theorem 1 and Section VI), and reports simulations for a 6-DOF industrial robot with parametric uncertainty, comparing against rigid-tube and nominal MPC.

Significance. If the advertised guarantees held, this would be a valuable result: a convex, real-time-feasible robust MPC that is less conservative than rigid-tube methods and scales to 6 DOF. The tube derivation in Section V is mostly sound, conditional on a valid model-error bound and a valid convex acceleration set, and the paper provides an open-source implementation and a useful experimental comparison. However, the headline collision-avoidance guarantee is not established as rigorously as claimed because the learned SCDF approximation has no certified conservatism, and because the model-error constants and the convex input set are obtained by sampling without formal verification. The paper's contribution is therefore best described as a promising algorithmic framework with strong simulation evidence, rather than a certified guaranteed-safety controller.

major comments (4)
  1. [Section VI-A, Eq. (20)-(21), Algorithm 1 line 9] The collision-avoidance guarantee relies on the implication: true state in tube, nominal state in tightened ball, and B(c) subset of free space imply collision-free motion. The last step requires that the SCDF value r(c) used to define B(c) is a conservative lower bound on the true distance from c to the obstacle boundary. The paper only states that the SCDF is 'approximated' by deep learning [11] and gives no formal approximation error bound, no Lipschitz/interval certificate, and no conservative margin. If r(c) overestimates clearance at any used ball center, (21) can admit points in the obstacle region even though Theorem 1's other conditions hold. The conventional collision check in Section VII-A4 verifies only the simulated trajectories, not all configurations inside the balls. This is a load-bearing gap for the claimed guarantee; it must be fixed by a certified lower bound on r, or
  2. [Section V-A, Prop. 1, Eq. (7)-(9)] Proposition 1 defines a, b, c as exact maxima over the continuous parameter and state sets. The text states these constants 'can be well approximated by sampling,' but if the sampled estimates are not verified upper bounds on the true maxima, then beta(x,a) in (6) is not a valid bound on the model error. Since the tube recursion (18e) and Theorem 1 use beta directly, undershooting the constants invalidates the robust tube guarantee. The paper needs either a rigorous over-approximation (e.g., interval arithmetic, Lipschitz-based bounds with sampling error certificates) or an explicit statement that the robust guarantees are conditional on the sampled constants.
  3. [Section V-C, Eq. (17)] The convex acceleration set A is obtained by uniformly sampling states and vertices, shrinking the set until the sampled condition is satisfied. No guarantee is provided that condition (17) holds for all (q, qdot) in X, so the resulting A is not certified to map into the torque constraint set U under the feedback linearization (2). This is another sampling-based certificate gap: the closed-loop input constraint satisfaction claim in Theorem 1 presupposes that A is valid for all states. A finite verification over the continuous state set or an explicit conservatism argument is needed.
  4. [Section VI-B, 'Convergence analysis'] The global convergence argument for the corridor planner is informal. Theorem 1 provides convergence only for a constant intermediate goal. In Algorithm 1, the virtual goal changes when the trajectory approaches it, and the paper states that advancement is ensured by 'a fine enough discretization,' but gives no precise condition relating ball radii, discretization step, the tightening eq(epsilon+delta_f), and the switching logic. The claim that the overall scheme 'successfully navigates non-convex obstacles and reaches the goal' is therefore not a proven guarantee; at most it is supported by simulation. Since the problem formulation includes reaching a goal as a requirement, this gap should be acknowledged and either formalized or explicitly left as an empirical property.
minor comments (5)
  1. [Section V-A, notation] The constants in (7)-(9) are denoted a, b, c, which collide with the acceleration vector a and the input matrix B. This makes expressions such as 'a ||a||' in (6) hard to read. Consider renaming the constants, e.g., alpha, beta, gamma.
  2. [Throughout] There are several typos: 'Mikeal Norrlof' should likely be 'Mikael'; 'the identify matrix' should be 'the identity matrix'; in Algorithm 1 line 9, 'C, B' uses B for the ball set, which conflicts with the input matrix B. The code link contains spaces ('rob cvx mpc rob man'), probably a formatting artifact.
  3. [Section VII-A3, Table I] Table I reports SCDF training time as 32000 s for both DOF cases; it would help to state the hardware/software environment and whether this is for the 3-DOF or 6-DOF model, since the SCDF is 'currently limited to 3 DOF' and the 6-DOF wrist is over-approximated by a sphere.
  4. [Section VI-A, Eq. (19)] The definition of the SCDF is for the exact signed distance. The transition to the learned approximation could be made explicit at the point of definition, e.g., by introducing a separate notation r_hat(q) for the network output and clearly stating where the exact r is replaced.
  5. [Section IV, Eq. (1)] The problem formulation states that q lies in a hyperbox C and qdot in V. The feedback linearization and the model error derivation assume the robot dynamics are given by (1); a brief statement on the validity of this model for the industrial robot (e.g., joint torques) would help.

Circularity Check

0 steps flagged · score 2.0 of 10

No construction-level circularity: the tube-MPC derivation is self-contained, and the SCDF dependence on prior work is a verification gap, not a circular reduction.

full rationale

Walking the derivation chain: Proposition 1 upper-bounds the feedback-linearization error using constants (7)-(9) computed from the parameter set Θ; Proposition 2 derives tube recursion (13) from that bound and the contraction condition (11); Theorem 1's recursive feasibility, constraint satisfaction, and convergence follow from the candidate solution and standard tracking-MPC arguments [23], [25]. None of these steps fits a parameter to the collision outcomes or predicts a quantity that was used to define the model. The collision-avoidance part uses B(c) = {q : ||q-c|| <= r(c)} subset Cf in (20), which is definitionally true for the exact SCDF r(q) in (19). The implementation replaces r(q) with a deep-network approximation from the authors' prior work [11] and then imposes (21); the paper does not prove that the learned r is a conservative lower bound on true clearance. This is a soundness/certification gap (and the paper's 'guarantee' wording overstates it), but it is not a circular reduction: the learned SCDF is an external, previously published input, and the simulated trajectories are independently checked with a conventional collision checker (Sec. VII-A4). The cited tracking-MPC lemmas are standard, reproducible results rather than self-citation used to forbid alternatives. Overall, the robust tube and convergence results are self-contained; only the SCDF-based safety certificate depends on an unverified external ingredient, which warrants a low but nonzero circularity/self-citation score rather than a zero.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

No new physical entities are introduced. The tube, corridor, and virtual goal are mathematical/algorithmic constructs. The SCDF is a learned function, not a new physical object.

free parameters (5)
  • model error constants a, b, c = sampled maxima, values not reported
    In (7)-(9), a, b, c are worst-case maxima over the uncertainty set and state space. The paper says they 'can be well approximated by sampling', so the exact bounds are not certified; Proposition 1 and the tube guarantee depend on them.
  • contraction rate rho = selected from 20 values in [0.8, 0.99]
    Chosen offline in Appendix B3; it determines the gain K and Lyapunov matrix P and must satisfy the contraction condition after uncertainty.
  • MPC weights Q, Qe, R and horizon H = Q=diag([10,0.01]), Qe=diag(1e4), R=diag(1e-3), H=20
    User-specified cost weights and horizon; standard tuning parameters.
  • auxiliary steps n_a = 4
    Number of inner-loop control steps between MPC solves; chosen for real-time feasibility.
  • SCDF network weights = trained on data
    The SCDF is a deep network trained in [11]; no formal conservatism guarantee is provided, and the collision-avoidance guarantee rests on it.
assumptions (6)
  • domain assumption Manipulator dynamics (1) with known nominal terms and bounded parametric uncertainty in M, C, g
    The feedback linearization (2) requires exact nominal model knowledge; the error bound (6) requires the uncertainty set Theta to be known and bounded.
  • ad hoc to paper The maxima (7)-(9) are finite and correctly computed
    The paper computes a, b, c by sampling; the theoretical guarantee requires true maxima over continuous sets.
  • domain assumption Contraction condition rho + d L_beta < 1 holds
    Assumed after Proposition 2. If violated, the tube does not contract and recursive feasibility may fail.
  • ad hoc to paper The learned SCDF is a conservative approximation of the signed distance to obstacles
    The paper relies on the deep-learned SCDF from [11]; no guarantee is given that it under-estimates distances.
  • ad hoc to paper The convex acceleration set A obtained by sampling satisfies (17) for all states
    The set is found by sampling states and shrinking; only sampled conditions are verified.
  • domain assumption The discretized double-integrator model (5) with 10 ms Euler step represents the continuous-time system
    All simulations use discrete time; the robustness guarantees are derived for the discrete model.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Robust Convex Model Predictive Control with collision avoidance guarantees for robot manipulators." pith.science (2026). https://pith.science/paper/XLANOHFM

@misc{pith2026250821677,
  author       = {Pith},
  title        = {Pith review of: Robust Convex Model Predictive Control with collision avoidance guarantees for robot manipulators},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XLANOHFM}},
  note         = {Machine review of arXiv:2508.21677}
}
read the original abstract

Industrial manipulators typically operate in cluttered environments, where safe motion planning is critical. However, model uncertainties further complicate this task, which leads to conservative speed limits to reduce the influence of disturbances. Hence, there is a need for control methods that can guarantee safe motions which are executed fast. We address this by suggesting a novel model predictive control (MPC) solution for manipulators, where our two main components are a robust tube MPC and a corridor planning algorithm to obtain collision-free motion. Our solution results in a convex MPC formulation, which we can solve fast, making our method practically useful. We demonstrate the efficacy of our method in a simulated environment with a 6 DOF industrial robot operating in cluttered environments with uncertain model parameters. We outperform benchmark methods by tolerating higher levels of model uncertainty while achieving faster motion.

Figures

Figures reproduced from arXiv: 2508.21677 by the authors.

Figure 1
Figure 1. Illustration of our convex robust MPC. We produce [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. All plots are illustrated in the configuration space. The yellow and purple symbolize the free and the obstacle regions, [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Left: Robot collision-geometry. The wrist axes are over approximated with a sphere (blue mesh). Middle: Illustrates the problem setup and a solved problem instance. The obstacles are represented by the gray spheres. The robot’s configurations at the start, goal, and an intermediate point along the solved trajectory are depicted in red, green, and blue, respectively. Right: The horizontal axis presents scaled uncerta… view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Model Predictive Planner for UAV Navigation in Non-Convex Air Corridors

    cs.RO 2026-07 conditional novelty 5.0 of 10

    Mixed-integer tracking MPC with a shortest-path offset cost plans dynamically feasible UAV trajectories in non-convex zonotopic air corridors and reaches the target without an external global planner.

Reference graph

Works this paper leans on

32 extracted references · 25 canonical work pages · cited by 1 Pith paper

  1. [11]

    Probabilistic Bubble Roadmap

    B. Wullt, M. Norrl ¨of, P. Mattsson, and Thomas B. Sch ¨on, “Probabilistic Bubble Roadmap.” https://arxiv.org/abs/2502.16205, 2025

    arXiv 2025

  2. [1]

    S. M. LaValle, Planning algorithms. Cambridge university press, 2006

    2006

  3. [2]

    Rapidly-exploring random trees: A new tool for path planning,

    S. LaValle, “Rapidly-exploring random trees: A new tool for path planning,” Research Report 9811 , 1998

    work page 1998

  4. [3]

    Probabilistic roadmaps for path planning in high-dimensional configuration spaces,

    L. Kavraki, P. Svestka, J.-C. Latombe, and M. Overmars, “Probabilistic roadmaps for path planning in high-dimensional configuration spaces,” IEEE Transactions on Robotics and Automation, vol. 12, no. 4, pp. 566– 580, 1996

    1996

  5. [4]

    Fast smoothing of manipulator tra- jectories using optimal bounded-acceleration shortcuts,

    K. Hauser and V . Ng-Thow-Hing, “Fast smoothing of manipulator tra- jectories using optimal bounded-acceleration shortcuts,” in International Conference on Robotics and Automation (ICRA) , 2010

    work page 2010

  6. [5]

    Time-optimal path tracking for robots: A convex optimization approach,

    D. Verscheure, B. Demeulenaere, J. Swevers, J. De Schutter, and M. Diehl, “Time-optimal path tracking for robots: A convex optimization approach,” IEEE Transactions on Automatic Control , vol. 54, no. 10, pp. 2318–2327, 2009

    work page 2009

  7. [6]

    Siciliano, L

    B. Siciliano, L. Sciavicco, L. Villani, and G. Oriolo, Robotics: Mod- elling, Planning and Control . Springer Publishing Company, 2008

    work page 2008

  8. [7]

    Lynch and F

    K. Lynch and F. Park, Modern Robotics: Mechanics, Planning, and Control. Cambridge Univeristy Press, 2017

    work page 2017

Show all 32 references
  1. [8]

    Finding and optimizing certified, collision-free regions in configuration space for robot manipulators,

    A. Amice, H. Dai, P. Werner, A. Zhang, and R. Tedrake, “Finding and optimizing certified, collision-free regions in configuration space for robot manipulators,” in International Workshop on the Algorithmic Foundations of Robotics, pp. 328–348, Springer, 2022

    2022

  2. [9]

    Growing convex collision-free regions in configuration space using nonlinear programming,

    M. Petersen and R. Tedrake, “Growing convex collision-free regions in configuration space using nonlinear programming,” arXiv preprint arXiv:2303.14737, 2023

    2023 arXiv

  3. [10]

    Configuration space distance fields for manipulation planning,

    Y . Li, X. Chi, A. Razmjoo, and S. Calinon, “Configuration space distance fields for manipulation planning,” arXiv preprint arXiv:2406.01137 , 2024

    2024 arXiv

  4. [12]

    The corridor map method: Real-time high-quality path planning,

    R. Geraerts and M. H. Overmars, “The corridor map method: Real-time high-quality path planning,” in IEEE International Conference on Robotics and Automation (ICRA) , IEEE, 2007

    2007

  5. [13]

    Flying on point clouds: Online trajectory generation and autonomous navigation for quadrotors in cluttered environments,

    F. Gao, W. Wu, W. Gao, and S. Shen, “Flying on point clouds: Online trajectory generation and autonomous navigation for quadrotors in cluttered environments,” Journal of Field Robotics , vol. 36, no. 4, pp. 710–733, 2019

    2019

  6. [14]

    Bubble planner: Planning high-speed smooth quadrotor trajectories using receding corridors,

    Y . Ren, F. Zhu, W. Liu, Z. Wang, Y . Lin, F. Gao, and F. Zhang, “Bubble planner: Planning high-speed smooth quadrotor trajectories using receding corridors,” in 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pp. 6332–6339, IEEE, 2022

    2022

  7. [15]

    KDF: Kino- dynamic motion planning via geometric sampling-based algorithms and funnel control,

    C. K. Verginis, D. V . Dimarogonas, and L. E. Kavraki, “KDF: Kino- dynamic motion planning via geometric sampling-based algorithms and funnel control,” IEEE Transactions on Robotics, vol. 39, no. 2, pp. 978– 997, 2023

    2023

  8. [16]

    Borrelli, A

    F. Borrelli, A. Bemporad, and M. Morari, Predictive control for linear and hybrid systems . Cambridge University Press, 2017

    2017

  9. [17]

    Safe and fast tracking on a robot manipulator: Robust MPC and neural network control,

    J. Nubert, J. K ¨ohler, V . Berenz, F. Allg¨ower, and S. Trimpe, “Safe and fast tracking on a robot manipulator: Robust MPC and neural network control,” IEEE Robotics and Automation Letters, vol. 5, no. 2, pp. 3050– 3057, 2020

    2020

  10. [18]

    Data-driven model predictive control for trajectory tracking with a robotic arm,

    A. Carron, E. Arcari, M. Wermelinger, L. Hewing, M. Hutter, and M. N. Zeilinger, “Data-driven model predictive control for trajectory tracking with a robotic arm,” IEEE Robotics and Automation Letters , vol. 4, no. 4, pp. 3758–3765, 2019

    2019

  11. [19]

    MPC for robot manipulators with integral sliding modes generation,

    G. P. Incremona, A. Ferrara, and L. Magni, “MPC for robot manipulators with integral sliding modes generation,” IEEE/ASME Transactions on Mechatronics, vol. 22, no. 3, pp. 1299–1307, 2017

    2017

  12. [20]

    Diffusion-based approximate MPC: Fast and consistent imitation of multi-modal action distributions,

    P. M. Julbe, J. Nubert, H. Hose, S. Trimpe, and K. J. Kuchenbecker, “Diffusion-based approximate MPC: Fast and consistent imitation of multi-modal action distributions,” arXiv preprint arXiv:2504.04603 , 2025

    2025

  13. [21]

    MPC-MPNet: Model- predictive motion planning networks for fast, near-optimal planning under kinodynamic constraints,

    L. Li, Y . Miao, A. H. Qureshi, and M. C. Yip, “MPC-MPNet: Model- predictive motion planning networks for fast, near-optimal planning under kinodynamic constraints,” IEEE Robotics and Automation Letters, vol. 6, no. 3, pp. 4496–4503, 2021

    2021

  14. [22]

    R. A. Horn and C. R. Johnson, Matrix analysis. Cambridge university press, 2012

    2012

  15. [23]

    Model predictive control for tracking using artificial references: Fundamentals, recent results and practical imple- mentation,

    P. Krupa, J. K ¨ohler, A. Ferramosca, I. Alvarado, M. N. Zeilinger, T. Alamo, and D. Limon, “Model predictive control for tracking using artificial references: Fundamentals, recent results and practical imple- mentation,” in Proc. Conf. Decision and Control (CDC) , 2024

    2024

  16. [24]

    Robust adaptive MPC using control contraction metrics,

    A. Sasfi, M. N. Zeilinger, and J. K ¨ohler, “Robust adaptive MPC using control contraction metrics,” Automatica, vol. 155, p. 111169, 2023

    2023

  17. [25]

    Analysis and design of model predictive control frameworks for dynamic operation—an overview,

    J. K ¨ohler, M. A. M¨uller, and F. Allg¨ower, “Analysis and design of model predictive control frameworks for dynamic operation—an overview,” Annual Reviews in Control , vol. 57, p. 100929, 2024

    2024

  18. [26]

    RRT-Connect: An efficient approach to single-query path planning,

    J. J. Kuffner and S. M. LaValle, “RRT-Connect: An efficient approach to single-query path planning,” in IEEE International Conference on Robotics and Automation (ICRA) , 2000

    2000

  19. [27]

    Clarabel: An interior-point solver for conic programs with quadratic objectives,

    P. J. Goulart and Y . Chen, “Clarabel: An interior-point solver for conic programs with quadratic objectives,” arXiv preprint arXiv:2405.12762 , 2024

    2024 arXiv

  20. [28]

    trimesh,

    Dawson-Haggerty, “trimesh,” 2019

    2019

  21. [29]

    S. Boyd, L. El Ghaoui, E. Feron, and V . Balakrishnan, Linear matrix inequalities in system and control theory . SIAM, 1994. APPENDIX A. Model error derivation In the following, to increase readability, we drop the input arguments to the functions, e.g. M = M(q). From the mani...

    1994

  22. [30]

    , c2 x,m, c2 u,1,

    Convex optimization program: We solve the following semidefinite program to produce the gain K and Lyapunov matrix P: min E,Y,c2 x,i,c2 u,j , ¯w2 L( ¯w2, c2 x,1, . . . , c2 x,m, c2 u,1, . . . , c2 u,n) (29a) s.t. ρ2E, (AE + BY)⊤ (AE + BY)⊤, E ≽ 0 (29b) c2 x,i, [Ax]iE ([Ax]iE)⊤...

  23. [31]

    In order to achieve this, we first have to derive an expression for the tightened state and input constraints

    Derivation of optimization problem: Our goal is to derive an optimization problem where the amount of tightening is included in the cost, thereby allowing us to reduce the conservatism in the tightening. In order to achieve this, we first have to derive an expression for the t...

  24. [32]

    For the configuration tightening, we used 0.1 [rad], which was the padded clearance added in the path planning

    Candidate selection: We normalized the objective function by dividing the constraint tightenings of the configuration, velocity and control with a representative normalizing factor. For the configuration tightening, we used 0.1 [rad], which was the padded clearance added in th...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.