Pith. sign in

REVIEW 3 major objections 5 minor 25 references

BR-MPPI: Barrier Rate guided MPPI for Enforcing Multiple Inequality Constraints with Learned Signed Distance Field

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read By treating the CBF class-K decay rate as an augmented state and projecting sampled controls onto the resulting equality constraints, BR-MPPI makes MPPI enforce multiple safety barriers while sampling near the safe-set boundary.

desk verdict Novel state-lifting of the class-K parameter into MPPI, but the central projection in Eq. (19) omits a required term and the paper's safety mechanism does not hold as written. read the letter →

arxiv 2506.07325 v1 pith:ULST2YZJ submitted 2025-06-08 cs.RO math.OC

Hardik Parwana , Taekyung Kim , Kehan Long , Bardh Hoxha , Hideki Okamoto , Georgios Fainekos , Dimitra Panagou This is my paper
classification cs.ROmath.OC
keywords modelpredictivepathintegralcontrolbarrierfunctionssigneddistancefieldssetinvarianceNagumo'stheoremmanifoldprojectionquadrotornavigation
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 tries to give the sampling-based controller MPPI hard safety guarantees rather than soft penalty terms. The idea is to convert each control barrier function inequality into an equality by adding a tunable decay-rate parameter $\tilde{\alpha}$ as an extra state, so MPPI itself decides how fast the robot may approach an obstacle boundary. Random controls are then projected onto the manifold defined by those equalities, and a Nagumo-inspired cost term pushes $\tilde{\alpha}$ negative near the boundary to keep the safe set invariant. The paper reports that the resulting controller needs fewer samples, passes through narrow corridors that vanilla MPPI detours around, and works for point robots, multi-shape robots, and a quadrotor without retuning.

What carries the argument

The central object is the augmented state $z_t=(x_t,\tilde{\alpha}_t)$, where $\tilde{\alpha}_t$ is the vector of per-constraint decay rates in the linear class-K function of the CBF condition. MPPI samples controls $u'_{x,t}, u'_{\tilde{\alpha},t}$ for this lifted system, and the projection operator $P$ in (20) maps them onto the intersection of the equality constraints $h_i(F(x_t,v))-h_i(x_t)=-(\tilde{\alpha}_{i,t}+a_i)h_i(x_t)$. For control-affine dynamics the constraints are linearized with the first-order expansion (17), so the projection becomes an equality-constrained minimum-norm problem with closed-form solution (20). The cost $Q_h$ penalizes $\tilde{\alpha}_{i,t+1}/h_i(x_t)$ inside the buffer $0\le h_i\le d_i$, which is the discrete-time form of Nagumo's condition that keeps trajectories from crossing the boundary.

What would settle it

One concrete check: for a nonlinear constraint such as $h(x)=\|x-c\|^2-r^2$ under single-integrator dynamics, sample $u'$, apply the printed projection (20)-(19), and measure the residual $|h(F(x,u))-h(x)+\tilde{\alpha}h(x)|$; the paper's Remark 3 already concedes first-order error, so a growing residual near a concave obstacle would show that the safety equality is not actually enforced.

Watch

Extended reading notes

Core claim

The paper's claim is that MPPI's unconstrained sampling can be made to respect multiple state- and control-dependent inequality constraints by replacing each CBF inequality with the equality $h_i(F(x_t,u_t))-h_i(x_t)=-\tilde{\alpha}_{i,t+1}h_i(x_t)$, where $\tilde{\alpha}$ is a state whose increments are additional control inputs chosen by MPPI. With this lifting, trajectories stay on an allowed-motion manifold in the augmented space while projecting to multimodal fans of paths in the robot's own position space, so both sides of an obstacle can be explored in one iteration. A learned signed distance field of the robot's full body provides the constraint functions $h_i$, letting the same controller handle different robot shapes. The empirical case studies assert that this design outperforms vanilla MPPI in sample efficiency and in closeness to the safe boundary, demonstrated on single-integrator, double-integrator, extended-unicycle, and quadrotor systems. The authors position this as the first MPPI formulation with multiple state- and control-dependent equality constraints.

Load-bearing premise

Safety rests on the equality constraints (13) being exactly satisfied for every sampled rollout, which the printed projection (20) cannot guarantee because equation (19) omits the $-\tilde{\alpha}_i h_i(x_t)$ term that (16c) requires, and because the underlying first-order Taylor expansion (17) can drift for nonlinear $h_i$.

Editorial extensions

If this is right

  • With the same MPPI hyperparameters, BR-MPPI can take a hexagonal robot and a point robot through environments where vanilla MPPI stalls or detours (Figs. 2-3).
  • Threading a narrow corridor that requires 20,000 samples for vanilla MPPI is achieved with far fewer samples by the proposed method.
  • Sampled trajectories in the lifted $(\tilde{\alpha},h)$ space remain unimodal while projecting to multimodal fans of paths in the robot's position space, allowing both sides of an obstacle to be explored in one iteration.
  • The learned SDF of the robot body means the same algorithm can be deployed on a quadrotor footprint without retuning for robot shape.

Reading between the lines

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

  • Editorial extension: the printed projection equations appear inconsistent--equation (19) omits the $-\tilde{\alpha}_i h_i(x_t)$ term that constraint (16c) requires, so any implementation should first verify which $b$ actually produces the reported results; a residual check of (13) would catch the omission immediately.
  • Editorial extension: the projection discards the off-manifold component of every sampled control, so the effective exploration noise is lower-dimensional; that may explain the higher sample efficiency, but it also means MPPI's importance-weighting convergence analysis would need revisiting for the projected sampler.
  • Editorial extension: because the linearization (17) is re-taken at each step, a drift-repair strategy like recomputing the tangent space when the residual grows, as in [23], could be tested on highly curved SDFs; the paper leaves this to future work.
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

3 major / 5 minor

Summary. The paper proposes BR-MPPI, a Model Predictive Path Integral controller that enforces multiple state- and control-dependent inequality constraints by converting Control Barrier Function conditions into equality constraints with a time-varying linear class-K parameter treated as an augmented state. The additional parameter rate is designed by MPPI through a projection operation that maps randomly sampled controls onto the manifold defined by the equality constraints. The authors train a signed distance field to model the robot geometry, and claim through simulations and a quadrotor experiment that BR-MPPI is more sample-efficient than vanilla MPPI and can operate closer to the safe-set boundary.

Significance. The core idea of adapting the class-K parameter online within MPPI and using manifold projection to satisfy multiple equality constraints is novel and potentially valuable for safety-critical real-time navigation, especially with learned SDFs. If the technical issues are corrected, the approach could offer a practical way to embed CBF-like reasoning into sampling-based control without solving QPs at runtime. However, the current manuscript contains a load-bearing algebraic inconsistency in the projection, the empirical evidence is anecdotal and lacks quantitative comparison, and the safety claims are stronger than what the soft-cost mechanism actually guarantees. The paper also does not provide code, which would help verify the implementation against the stated equations.

major comments (3)
  1. [Section III-C, Eq. (16c) and Eq. (19)] The exact text is: "the b vector in Eq. (19) is printed as -∂h_i/∂x f(x_t), omitting the -\tilde{\alpha}_{i,t} h_i(x_t) term."
  2. [Sections IV and V] This is a major comment because the abstract and introduction explicitly claim 'better sampled efficiency and enhanced capability to operate closer to the safe set boundary'.
  3. [Section III-D and Eq. (22)] The paper itself acknowledges in Section VI that 'derivation of guarantees of inequality constraint maintenance' is future work, but the abstract and Section III-D use stronger language.
minor comments (5)
  1. [Throughout] There are several typos: 'sampled efficiency' should be 'sample efficiency', 'Signed Distance filed' should be 'field', and 'V oxblox' has a broken spacing.
  2. [Section III-C, Eq. (20)] The citation for the weighted minimum-norm solution appears as '[?]' in the manuscript; this reference is missing.
  3. [Section III-D, Eq. (22)] The indicator condition '1(xt ∈D_i ∩ x_t ∈ \bar{S}_i)' is redundant because D_i is already defined as a subset of S_i, and the notation \bar{S}_i is not defined. Please clarify the intended set.
  4. [Section IV and V] The paper does not report the common MPPI hyperparameters (sample count K, horizon H, noise covariance, cost weights λ and γ) for either BR-MPPI or vanilla MPPI, making the experiments difficult to reproduce or compare.
  5. [Section III-C, Eq. (20)] The projection (20) requires the matrix A W^{-1} A^T to be invertible; the paper does not discuss rank conditions or numerical conditioning for the case of redundant or nearly dependent constraints.

Circularity Check

1 steps flagged · score 2.0 of 10

One mild self-definitional step: the safety behavior near the boundary is encoded directly into the MPPI cost, while the central empirical claims remain benchmarked against vanilla MPPI.

  1. self definitional [Section III-B, Eq. (13); Section III-D, Eq. (22)]
    "we impose the following equality constraint h_i(F(x_t, u_{x_t})) - h_i(x_t) = -\alpha_{i,t} h_i(x_t), ∀i ∈ {1, .., N} (13) ... Inside this buffer zone, we impose the condition that α_t < 0 in (13) to promote positive ˙h_i. ... Q_h(z_t, z_{t+1}, u) = Σ_i 1(x_t ∈ D_i ∩ x_t ∈ \bar{S}_i) \tilde{α}_{i,t+1} / h_i(x) (22)"

    By (13), h_i(F)-h_i = -\tilde{\alpha}_{i,t+1} h_i(x_t) when \tilde{\alpha}_{i,t+1} plays the role of α_t. Thus choosing \tilde{\alpha}_{i,t+1}<0 inside the buffer (where h_i>0) is exactly the condition that makes h_i increase, i.e., the discrete Nagumo condition. Equation (22) assigns a negative cost to negative \tilde{\alpha}_{i,t+1} in D_i, so MPPI's minimization is constructed to reward the very safety condition that the paper says it 'reignites.' The safety behavior near the boundary is therefore installed by construction in the objective rather than derived from the projection (20) or from an independent first-principles prediction.

full rationale

The derivation chain from the equality constraint (13) to the projection (16)-(20) is a genuine algorithmic construction, and the empirical comparison against vanilla MPPI is an external benchmark, so the central claim does not reduce to its inputs. The only self-referential element is the cost design in Section III-D: the buffer-zone cost (22) directly rewards negative \tilde{\alpha}, which by (13) is the same as promoting positive ˙h_i near the boundary; this is an explicit design choice, not a hidden circularity. There are no load-bearing self-citations: Remark 2 cites the authors' prior parameter-adaptation works [11], [12] only as related ideas, and the learned SDF is trained independently via (11)-(12). A separate, non-circular correctness concern is that the printed A and b matrices in (19) omit the -\tilde{\alpha}_{i,t} h_i(x_t) term required by (16c), so the analytical projection (20) as printed enforces a different equality than the one that defines BR-MPPI; this is an algebraic inconsistency and a soundness risk, but it is not an input-output equivalence of the kind that constitutes circularity. Overall the paper's main safety and sample-efficiency claims are not forced by definition, and the circularity score is low.

Assumptions & free parameters 6 free parameters · 4 assumptions · 1 invented entities

The algorithm rests on four load-bearing modeling choices: linearization of the safety functions, accuracy of the learned SDF, a discrete-time adaptation of Nagumo's theorem, and an assumption that weighted averaging preserves constraint satisfaction. The projection equations as printed also contain an inconsistency, so the ledger's counts understate the fixability burden.

free parameters (6)
  • Buffer length d_i = not reported
    Defines the buffer zone D_i in (21) where the Nagumo cost Q_h is active; called 'a tuning parameter in practice' in Section III-D. The central safety behavior depends on its value, and no value is given.
  • MPPI noise covariance Σ_w = not reported
    Sampling distribution for control perturbations in (7c); user-chosen and central to MPPI sample efficiency, but not specified.
  • MPPI cost weights λ, γ and convergence cost Q_c = not reported
    Determine trajectory weights (8) and task objective; not specified in the paper.
  • Projection weight matrices Q1, Q2 (W) = not reported
    Trade off projection between u_x and u_alpha in (16b); not specified.
  • Sample count K and horizon H = not reported
    MPPI rollout parameters; the claimed sample efficiency comparison depends on K and H, but no values are given.
  • SDF training hyperparameters (e.g., λ_E) = not reported
    Balance distance and Eikonal loss in (11); no architecture or training details are given.
assumptions (4)
  • domain assumption First-order Taylor expansion of h_i is accurate enough to enforce the equality constraints
    Used in (17) to reduce the projection to a linear system; for highly nonlinear constraints this introduces errors, acknowledged in Remark 3.
  • domain assumption The learned SDF and its gradient accurately represent the robot body and environment
    Sections III-A and V rely on trained SDF for h_i; accuracy is not quantified.
  • domain assumption Nagumo's theorem can be applied in discrete time by requiring alpha_t < 0 in the buffer zone
    Section III-D invokes Nagumo's theorem with a discrete-time equality; this is an adaptation, not the continuous-time theorem, and no formal proof is given.
  • ad hoc to paper The weighted average of projected control samples satisfies the equality constraint
    MPPI returns the weighted average (9) of projected controls; the paper does not show this average lies on the constraint manifold, especially under the exact nonlinear h_i.
invented entities (1)
  • Augmented class-K parameter state \tilde{\alpha}_t
    purpose: Lifts the state space with one coordinate per constraint so the CBF inequality becomes an equality and alpha can be adapted online by MPPI.
    This is a mathematical state-lifting; it has no falsifiable handle outside the algorithm and is entirely internal to the formulation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of BR-MPPI: Barrier Rate guided MPPI for Enforcing Multiple Inequality Constraints with Learned Signed Distance Field." pith.science (2026). https://pith.science/paper/ULST2YZJ

@misc{pith2026250607325,
  author       = {Pith},
  title        = {Pith review of: BR-MPPI: Barrier Rate guided MPPI for Enforcing Multiple Inequality Constraints with Learned Signed Distance Field},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ULST2YZJ}},
  note         = {Machine review of arXiv:2506.07325}
}
read the original abstract

Model Predictive Path Integral (MPPI) controller is used to solve unconstrained optimal control problems and Control Barrier Function (CBF) is a tool to impose strict inequality constraints, a.k.a, barrier constraints. In this work, we propose an integration of these two methods that employ CBF-like conditions to guide the control sampling procedure of MPPI. CBFs provide an inequality constraint restricting the rate of change of barrier functions by a classK function of the barrier itself. We instead impose the CBF condition as an equality constraint by choosing a parametric linear classK function and treating this parameter as a state in an augmented system. The time derivative of this parameter acts as an additional control input that is designed by MPPI. A cost function is further designed to reignite Nagumo's theorem at the boundary of the safe set by promoting specific values of classK parameter to enforce safety. Our problem formulation results in an MPPI subject to multiple state and control-dependent equality constraints which are non-trivial to satisfy with randomly sampled control inputs. We therefore also introduce state transformations and control projection operations, inspired by the literature on path planning for manifolds, to resolve the aforementioned issue. We show empirically through simulations and experiments on quadrotor that our proposed algorithm exhibits better sampled efficiency and enhanced capability to operate closer to the safe set boundary over vanilla MPPI.

Figures

Figures reproduced from arXiv: 2506.07325 by the authors.

Figure 1
Figure 1. The sampled paths (green) and the weighted average path (blue) of our proposed MPPI for a robot navigating among black obstacles. As shown, BR-MPPI plans paths in an augmented state space that produces multimodal paths in the original state space. and CBF have demonstrated their capabilities in generating safe paths for several complex systems. however, they suffer from several shortcomings. The most notable being t… view at source ↗
Figure 2
Figure 2. Extended Unicycle point mass robot navigating in an obstacle environment. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Hexagonal Single Integrator robot navigating in an obstacle environment. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Experiment Setup. The left figure shows the obstacle environment, and the right figure shows our custom-assembled [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: A simulated run of the experiment setup. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

25 extracted references · 16 canonical work pages

  1. [3]

    Control barrier function augmentation in sampling-based control algorithm for sample efficiency,

    C. Tao, H. Kim, H. Yoon, N. Hovakimyan, and P. V oulgaris, “Control barrier function augmentation in sampling-based control algorithm for sample efficiency,” in2022 American Control Conference (ACC). IEEE, 2022, pp. 3488–3493

    work page 2022

  2. [8]

    Shield model predictive path integral: A computationally efficient robust mpc method using control barrier functions,

    J. Yin, C. Dawson, C. Fan, and P. Tsiotras, “Shield model predictive path integral: A computationally efficient robust mpc method using control barrier functions,”IEEE Robotics and Automation Letters, 2023

    work page 2023

  3. [10]

    Safety in Augmented Importance Sampling: Performance Bounds for Robust MPPI

    M. Gandhi, H. Almubarak, Y . Aoyama, and E. Theodorou, “Safety in augmented importance sampling: Performance bounds for robust mppi,”arXiv preprint arXiv:2204.05963, 2022

    work page Pith review arXiv 2022

  4. [1]

    Information-theoretic model predictive control: Theory and applica- tions to autonomous driving,

    G. Williams, P. Drews, B. Goldfain, J. M. Rehg, and E. A. Theodorou, “Information-theoretic model predictive control: Theory and applica- tions to autonomous driving,”IEEE Transactions on Robotics, vol. 34, no. 6, pp. 1603–1622, 2018

    2018

  5. [2]

    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

    2016

  6. [4]

    Robust model predictive path integral control: Analysis and performance guarantees,

    M. S. Gandhi, B. Vlahov, J. Gibson, G. Williams, and E. A. Theodorou, “Robust model predictive path integral control: Analysis and performance guarantees,”IEEE Robotics and Automation Letters, vol. 6, no. 2, pp. 1423–1430, 2021

    2021

  7. [5]

    Trajectory dis- tribution control for model predictive path integral control using covariance steering,

    J. Yin, Z. Zhang, E. Theodorou, and P. Tsiotras, “Trajectory dis- tribution control for model predictive path integral control using covariance steering,” in2022 International Conference on Robotics and Automation (ICRA). IEEE, 2022, pp. 1478–1484

    work page 2022

  8. [6]

    Biased-mppi: Informing sampling- based model predictive control by fusing ancillary controllers,

    E. Trevisan and J. Alonso-Mora, “Biased-mppi: Informing sampling- based model predictive control by fusing ancillary controllers,”IEEE Robotics and Automation Letters, 2024

    work page 2024

Show all 25 references
  1. [7]

    Constrained stochastic optimal control with learned importance sampling: A path integral approach,

    J. Carius, R. Ranftl, F. Farshidian, and M. Hutter, “Constrained stochastic optimal control with learned importance sampling: A path integral approach,”The International Journal of Robotics Research, vol. 41, no. 2, pp. 189–209, 2022

    2022

  2. [9]

    Chance-constrained information- theoretic stochastic model predictive control with safety shielding,

    P. Tsiotras, K. Berntorp,et al., “Chance-constrained information- theoretic stochastic model predictive control with safety shielding,” arXiv preprint arXiv:2408.00494, 2024

    2024 arXiv

  3. [11]

    Recursive feasibility guided optimal parameter adaptation of differential convex optimization policies for safety-critical systems,

    H. Parwana and D. Panagou, “Recursive feasibility guided optimal parameter adaptation of differential convex optimization policies for safety-critical systems,” in2022 International Conference on Robotics and Automation (ICRA). IEEE, 2022, pp. 6807–6813

    2022

  4. [12]

    Trust-based rate-tunable control barrier functions for non-cooperative multi-agent systems,

    H. Parwana, A. Mustafa, and D. Panagou, “Trust-based rate-tunable control barrier functions for non-cooperative multi-agent systems,” in 2022 IEEE 61st Conference on Decision and Control (CDC). IEEE, 2022, pp. 2222–2229

    2022

  5. [13]

    Adaptive control barrier functions,

    W. Xiao, C. Belta, and C. G. Cassandras, “Adaptive control barrier functions,”IEEE Transactions on Automatic Control, vol. 67, no. 5, pp. 2267–2281, 2021

    2021

  6. [14]

    Safety-critical control using optimal-decay control barrier function with guaranteed point- wise feasibility,

    J. Zeng, B. Zhang, Z. Li, and K. Sreenath, “Safety-critical control using optimal-decay control barrier function with guaranteed point- wise feasibility,” in2021 American Control Conference (ACC). IEEE, 2021, pp. 3856–3863

    2021

  7. [15]

    Learning dif- ferentiable safety-critical control using control barrier functions for generalization to novel environments,

    H. Ma, B. Zhang, M. Tomizuka, and K. Sreenath, “Learning dif- ferentiable safety-critical control using control barrier functions for generalization to novel environments,” in2022 European Control Conference (ECC). IEEE, 2022, pp. 1301–1308. Fig. 2: Extended Unicycle point mas...

    2022

  8. [16]

    Fiesta: Fast incremental Euclidean distance fields for online motion planning of aerial robots,

    L. Han, F. Gao, B. Zhou, and S. Shen, “Fiesta: Fast incremental Euclidean distance fields for online motion planning of aerial robots,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2019

    2019

  9. [17]

    V oxblox: Incremental 3d Euclidean signed distance fields for on- board mav planning,

    H. Oleynikova, Z. Taylor, M. Fehr, R. Siegwart, and J. Nieto, “V oxblox: Incremental 3d Euclidean signed distance fields for on- board mav planning,” inIEEE/RSJ International Conference on In- telligent Robots and Systems (IROS), 2017, pp. 1366–1373

    2017

  10. [18]

    Learning barrier functions with memory for robust safe navigation,

    K. Long, C. Qian, J. Cort ´es, and N. Atanasov, “Learning barrier functions with memory for robust safe navigation,”IEEE Robotics and Automation Letters, vol. 6, no. 3, pp. 4931–4938, 2021

    2021

  11. [19]

    Sensor-based distributionally robust control for safe robot navigation in dynamic environments,

    K. Long, Y . Yi, Z. Dai, S. Herbert, J. Cort ´es, and N. Atanasov, “Sensor-based distributionally robust control for safe robot navigation in dynamic environments,”arXiv preprint arXiv:2405.18251, 2024

    2024 arXiv

  12. [20]

    Representing robot ge- ometry as distance fields: Applications to whole-body manipulation,

    Y . Li, Y . Zhang, A. Razmjoo, and S. Calinon, “Representing robot ge- ometry as distance fields: Applications to whole-body manipulation,” inProc. IEEE Intl Conf. on Robotics and Automation (ICRA), 2024, pp. 15 351–15 357

    2024

  13. [21]

    Neural configuration distance function for continuum robot control,

    K. Long, H. Parwana, G. Fainekos, B. Hoxha, H. Okamoto, and N. Atanasov, “Neural configuration distance function for continuum robot control,” 2024. [Online]. Available: https://arxiv.org/abs/2409. 13865

    2024

  14. [22]

    Sampling-based methods for motion planning with constraints,

    Z. Kingston, M. Moll, and L. E. Kavraki, “Sampling-based methods for motion planning with constraints,”Annual review of control, robotics, and autonomous systems, vol. 1, no. 1, pp. 159–185, 2018

    2018

  15. [23]

    A randomized kinodynamic Fig. 3: Hexagonal Single Integrator robot navigating in an obstacle environment. planner for closed-chain robotic systems,

    R. Bordalba, L. Ros, and J. M. Porta, “A randomized kinodynamic Fig. 3: Hexagonal Single Integrator robot navigating in an obstacle environment. planner for closed-chain robotic systems,”IEEE Transactions on Robotics, vol. 37, no. 1, pp. 99–115, 2020

    2020

  16. [24]

    ¨Uber die lage der integralkurven gew ¨ohnlicher differen- tialgleichungen,

    M. Nagumo, “ ¨Uber die lage der integralkurven gew ¨ohnlicher differen- tialgleichungen,”Proceedings of the physico-mathematical society of Japan. 3rd Series, vol. 24, pp. 551–559, 1942

    1942

  17. [25]

    Set invariance in control,

    F. Blanchini, “Set invariance in control,”Automatica, vol. 35, no. 11, pp. 1747–1767, 1999. Fig. 4: Experiment Setup. The left figure shows the obstacle environment, and the right figure shows our custom-assembled quad. Fig. 5: A simulated run of the experiment setup

    1999

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.