EL-AGHF: Extended Lagrangian Affine Geometric Heat Flow

Hae-Won Park; Sangmin Kim

REVIEW 3 major objections 5 minor 17 references

EL-AGHF: Extended Lagrangian Affine Geometric Heat Flow

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 adding a dual trajectory for inadmissible control directions, EL-AGHF generates dynamically feasible trajectories with finite penalty weights.

desk verdict EL-AGHF puts a dual trajectory inside AGHF and empirically dodges the infinite-penalty barrier, but the 'ensuring admissibility' claim outruns the unproven convergence of the coupled PDE. read the letter →

arxiv 2505.24751 v1 pith:CWLLO526 submitted 2025-05-30 cs.RO

Sangmin Kim , Hae-Won Park This is my paper

classification cs.RO

keywords affinegeometricheatflowmotionplanningaugmentedLagrangianconstrainedvariationalproblemparabolicpartialdifferentialequationdynamicfeasibilityunderactuatedsystemsdualtrajectory

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 an extension of the Affine Geometric Heat Flow (AGHF) motion-planning method that removes the need to drive a penalty weight to infinity in order to enforce dynamic feasibility. AGHF encodes dynamics into a Riemannian metric and deforms an arbitrary trajectory into a control-effort-minimizing one by a parabolic PDE, but admissible trajectories only emerge in the λ → ∞ limit, which is computationally expensive and numerically unstable. EL-AGHF instead treats the dynamics gaps in inadmissible control directions as hard constraints and introduces a dual trajectory µ(t) for them, turning the problem into a min-max optimization over an extended Lagrangian. The resulting coupled parabolic PDE evolves the state trajectory and the dual trajectory together in an artificial time s, and at a stationary limit the pair satisfies both the Euler–Lagrange condition and zero dynamics gap, so the trajectory is dynamically feasible. Simulations on unicycle and mid-air diver models show that the method keeps dynamics gaps small even at low penalty weights, where standard AGHF would need very large penalties.

What carries the argument

The engine is the extended Lagrangian $$\bar L(x,\dot x,\mu) = (\dot x - F_d + F_c\mu)^\top G(\dot x - F_d + F_c\mu) - \$\lambda$ \mu^\top \mu,$$ together with the coupled gradient-like PDE system (19)–(20): the state trajectory $x(t,s)$ evolves by the AGHF descent direction $G^{-1}(\frac{d}{dt}\frac{\partial \bar L}{\partial \dot x} - \frac{\partial \bar L}{\partial x})$, and the dual trajectory $\mu(t,s)$ evolves by the ascent direction $(F_c^\top G F_c)^{-1}\frac{\partial \bar L}{\partial \mu}$. The $x$-flow inherits the property that the action is non-increasing along $s$ whenever $\mu$ is held fixed, and the $\mu$-flow drives the dynamics gap $F_c^\dagger(\dot x - F_d)$ toward zero. Kinematic inequality constraints are handled by the same construction, adding dual variables $\mu^c_j$ with the penalty term $\lambda_c((h_j(x)+\mu^c_j)^2 - (\mu^c_j)^2)S(h_j(x))$.

What would settle it

Run EL-AGHF on a simple nonholonomic system, for example the dynamic unicycle with $\lambda = 1$, and monitor the residual $F_c^\dagger(\dot x - F_d)$ and the dual update $\partial \mu/\partial s$ as $s$ grows: if either stalls above zero, grows without bound, or forces the solver to hit its time limit while still changing, then the claimed limit (22)–(23) is not attained and the feasibility guarantee fails.

Watch

Extended reading notes

Core claim

The central claim is that the limiting pair $(x^*, \mu^*)$ of the extended parabolic flow satisfies the stationarity conditions (22)–(23) of the constrained variational problem (13)–(15), and that $x^*(t)$ is therefore a local minimum that ensures dynamic feasibility. The load-bearing mechanism is the augmented-Lagrangian identity $$\bar L(x,\dot x,\mu) = (\dot x - F_d + F_c\mu)^\top G(\dot x - F_d + F_c\mu) - \$\lambda$ \mu^\top \mu,$$ whose saddle point in $x$ and $\mu$ encodes both control-effort minimization and the vanishing of the dynamics gap in inadmissible directions. Because the constraint is carried by the dual trajectory rather than by unbounded metric scaling, the method aims to achieve admissibility with finite penalty weights. The same construction is applied to kinematic constraints by adding dual variables $\mu^c_j$ for each constraint, with the penalty term $(h_j(x)+\mu^c_j)^2 - (\mu^c_j)^2$ imitating augmented-Lagrangian smoothing.

Load-bearing premise

The dynamic-feasibility guarantee rests on the assumption that the coupled state-dual PDE (19)–(20) converges to a stationary pair $(x^*, \mu^*)$ as $s$ grows; the paper explicitly states that establishing this convergence is future work, so the paper does not yet prove that its flow reaches the stationarity conditions (22)–(23).

Editorial extensions

If this is right

Motion planners can obtain dynamically feasible trajectories for underactuated systems with finite, moderate penalty weights instead of the $\lambda \to \infty$ limit that makes AGHF numerically unstable.
The same dual-trajectory construction applies to kinematic inequality constraints, so dynamic feasibility and constraint satisfaction are handled in one PDE evolution.
EL-AGHF preserves the geometric structure of AGHF: when the dual is fixed, the action functional is non-increasing along the evolution direction, so the method remains a variational descent rather than a generic optimizer.
In the reported unicycle and mid-air diver experiments, EL-AGHF keeps dynamics gaps small and satisfies kinematic constraints even at $\lambda = 1$, where standard AGHF exhibits large gaps.

Reading between the lines

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

If the planned convergence proof succeeds, EL-AGHF would become an infinite-dimensional augmented Lagrangian method for trajectory optimization, making penalty coefficients a systematic object of study rather than a hand-tuned parameter.
The paper applies free boundary conditions to the dual trajectories in the numerical solver, a choice that is not derived from the variational principle; testing other boundary conditions would reveal whether the reported feasibility depends on that solver detail.
The steady-state dual trajectory could be read as a sensitivity measure: larger $|\mu^*(t)|$ would indicate control directions and time instants where the dynamics constraint is most expensive to satisfy, which could help identify inherently hard maneuvers before planning.
A natural next experiment is obstacle avoidance with inequality state constraints, since the barrier-free augmented-Lagrangian form $(h_j + \mu^c_j)^2 - (\mu^c_j)^2$ is designed to keep constraints active without the numerical stiffness of large penalties.

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.

Desk Editor's Note

read the letter

Dear colleague,

EL-AGHF is worth a look. The paper extends Affine Geometric Heat Flow by adding a dual trajectory that adaptively cancels dynamics gaps in inadmissible control directions. The coupled PDE system (19)–(20) is genuinely new relative to the prior AGHF work, and the simulations back it up: on the unicycle and mid-air benchmarks, the method keeps terminal dynamics error small even at metric penalty λ = 1, where standard AGHF needs λ = 10⁴ to get comparable numbers. The kinematic-constraint extension is less novel, but it follows the same pattern.

What's done well: the augmented Lagrangian reformulation is clean, the completion-square identity (18) is a neat trick to keep the metric positive definite, and the authors are upfront in Section V that convergence is future work. The experiments use the original AGHF code and same initializations, so the comparison is fair. The reported times also show the method is not a magic bullet—it hits the same 600-second cutoff at high λ on the mid-air model—but the low-λ regime is where the pain was.

The soft spot is exactly the one the authors flag, and it is load-bearing. The abstract says the method 'ensures admissibility,' but that guarantee rests on the coupled state-dual PDE converging to a stationary point, and no convergence proof is given. The only descent statement (21) holds for a frozen dual; along the actual flow the action derivative contains a positive term from the dual ascent, so there's no Lyapunov function. The stress-test note is right on this. Stationarity (22)–(23) alone is necessary, not sufficient, for a local minimum. And the numerical implementation applies free boundary conditions to µ even though (20) has no t-derivative in µ—that choice is unexplained and could affect the results. None of this kills the empirical contribution, but it means 'ensuring admissibility' overstates what is established.

Who this is for: people working on AGHF-style variational motion planning, or on differential multiplier methods for trajectory optimization. It deserves a serious referee. I'd send it to review with a request to either prove descent along the coupled flow (or a nontrivial subclass), or soften the abstract and theorem claims to describe a heuristic that works well in practice. The idea is sensible, the experiments are informative, and the proof gap is a well-posed problem rather than a hand-wave.

Hope that helps.

Referee Report

3 major / 5 minor

Summary. The paper proposes EL-AGHF, an augmented-Lagrangian extension of the Affine Geometric Heat Flow method for motion planning of control-affine systems. The key idea is to introduce dual trajectories µ(t) associated with the inadmissible control directions of the dynamics, reformulate the trajectory optimization problem as a min-max problem over an extended action functional, and solve the resulting coupled parabolic PDE system in an artificial evolution variable s. The same construction is extended to kinematic inequality constraints via additional dual variables µ_c. The authors claim that, at convergence, the limiting state-dual pair satisfies the stationarity conditions of the constrained variational problem, so that the resulting trajectory is dynamically feasible, and they support this with simulations on two unicycle models and a mid-air diver model, comparing against AGHF.

Significance. If the convergence claim were proven, EL-AGHF would be a practically useful extension of AGHF: it promises dynamically feasible trajectories with finite penalty weights, thereby avoiding the large-λ numerical stiffness that limits AGHF. The paper is also commendable for benchmarking against the original AGHF code, for being transparent about initialization and parameter choices, and for not fitting any result to data. The numerical comparisons consistently show smaller dynamics gaps for EL-AGHF at low λ, which is encouraging. However, the central theoretical guarantee advertised in the abstract and conclusions is not established: the convergence of the coupled state-dual PDE is explicitly deferred to future work, and the only monotonicity result is for a frozen dual variable. The local-minimum claim is also stronger than what the first-order stationarity conditions justify. These are load-bearing gaps, but they are fixable in a revision by either supplying a convergence proof or by carefully limiting the paper's claims to empirical observations.

major comments (3)

[§III-A, Eqs. (19)–(20); §V] The central claim that EL-AGHF 'ensures the admissibility of the resulting trajectory' rests on convergence of the coupled state-dual PDE (19)–(20) to a stationary pair (x*, µ*). Section V states that convergence is future work, and the only monotonicity statement, Eq. (21), is for µ fixed at µ*, not for the actual coupled flow. Along the actual flow, d/ds of the extended action contains a negative contribution from the x-descent step and a positive contribution from the µ-ascent step in (20), so no Lyapunov inequality is available; no global-in-s existence or convergence argument is given either. Therefore Eqs. (22)–(23) are conditional statements, and the abstract's guarantee is unsupported. The authors should either prove convergence (e.g., via an energy argument in the spirit of the BDMM reference [17]) or explicitly restrict the paper's claims to an empirical demonstration.
[§III-A, Eqs. (22)–(23)] Even if the limit (x*, µ*) satisfying (22)–(23) exists, those equations are first-order necessary conditions. The sentence following them, 'The trajectory x*(t) is a local minimum of (13)–(15) that ensures dynamic feasibility,' is therefore stronger than what is proved: the strengthened Legendre condition (16) plus stationarity does not by itself establish a local minimum in the presence of the constraint (14), and no second-order sufficient condition or constraint-qualification argument is given. Please either prove the local-minimum claim or replace 'local minimum' with 'stationary point' and separately state that dynamic feasibility follows from (23) whenever the limit is attained.
[§IV-A, boundary conditions for (20), (29), (30)] The numerical implementation applies 'free boundary conditions' to the dual variables µ and µ_c, but this choice is neither derived nor justified. Equation (20) contains no t-derivative of µ, so pdepe's boundary conditions for these variables are not determined by the variational formulation; they may select a particular solution of the infinite-dimensional saddle-point problem. The authors should derive the appropriate boundary conditions from the first variation of the extended action (for example, natural boundary conditions from integration by parts) or explain why the free-boundary choice is inconsequential for the reported results.

minor comments (5)

[Abstract and Introduction] The word 'trajectorys' is misspelled; it should be 'trajectories' throughout.
[§II-A, Eq. (7) and surrounding text] The equivalence in the completion-square identity (18) relies on Fc having orthonormal columns and on the relation Fc^T G = λ Fc†. The paper mentions Gram–Schmidt orthogonalization, but it would be helpful to state the orthonormality assumption explicitly before using (18), since the identity is central to the extended Lagrangian construction.
[§IV-C, model definition] In the mid-air motion model, the text says 'Here, ¯F(x) = [Fc(x) | F(x)] ∈ R6×6 is chosen as follows: F(x) = [I3 03; 03 D^-1(q)]'. The matrix displayed is 6×6 and appears to be ¯F(x), not F(x); please correct the notation to avoid confusing the 6×6 full-rank matrix with the 6×2 control-input matrix.
[§IV-B-2, dynamic unicycle] The sentence describing the initial trajectory is confusing: 'Both EL-AGHF and AGHF use a linearly interpolated trajectory between the boundary conditions (except for x(t,0)) as the initial trajectory. The x-component follows a sinusoidal trajectory.' Please clarify which components are linearly interpolated and which use the sinusoidal initialization.
[§IV-C, convergence tolerances] The convergence threshold ε is set to 10^-2 in Section IV-C-1 and to 10^-1 in Section IV-C-2, with no explanation. Since these thresholds directly affect the reported terminal errors and the decision to terminate at the 600 s time limit, a brief justification or a sensitivity check would strengthen the experimental section.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central guarantee rests on an explicitly deferred convergence assumption, not on fitting, renaming, or self-citation.

full rationale

The EL-AGHF derivation is self-contained in the relevant sense. No parameter is fitted to the quantities later reported as results: lambda, lambda_c, epsilon, and k_s are user-selected and varied across experiments, and e(T) and e_viol are measured by independent forward integration of the system dynamics against the generated control inputs. The imported monotonicity results are Lemma 1 and Theorem 1 of [1], authored by Liu, Fan, and Belabbas, not by the present authors, so the paper does not rely on a load-bearing self-citation chain. The conditional statement in Section III-A, 'If the trajectories x(t,s) and mu(t,s) converge to x*(t) and mu*(t), then ...', is exactly that: a conditional. Section V explicitly defers the convergence proof to future work, and the numerical use of free boundary conditions for mu is an implementation choice. These are proof gaps and correctness risks, not reductions of the output to the input. There is no equation whose definition makes a target result true by construction, no fitted parameter renamed as a prediction, and no ansatz imported through a same-author citation to force the conclusion. Hence the paper is not circular, though its admissibility guarantee is conditional on an unproved convergence statement.

Assumptions & free parameters 4 free parameters · 4 assumptions · 2 invented entities

The central claim rests on four assumptions: external Lemma 1 for monotonicity, the full-rank and orthogonal-complement structure of Fbar, unproven convergence of the coupled PDE, and the validity of free boundary conditions in pdepe. Free parameters (lambda, lambda_c, epsilon, k_s) are user-chosen, not fit to data, but the reported results depend on their values.

free parameters (4)

lambda (metric / penalty scaling) = 1, 10, 100, 1000, 10000 in experiments
Controls the relative weight of inadmissible directions in G and the multiplier term; chosen by hand, not fit to data.
lambda_c (kinematic penalty coefficient) = same as lambda in experiments (1 to 1000)
Penalty for inequality constraints h_j in Eq. (27); user-selected.
convergence threshold epsilon = 1e-4 for unicycle, 1e-2 and 1e-1 for mid-air
Stops s-integration when derivatives of x, mu, and mu_c are small; affects reported s_max and e(T).
k_s (Heaviside smoothness) = 100
Sharpness of smooth approximation in Eq. (26) and Eqs. (37)-(38); chosen by hand.

assumptions (4)

standard math The action functional A is bounded below and Lemma 1 of [1] holds (non-increase along AGHF flow).
Used in Section III-A to assert (21) for fixed dual; imported from prior work without proof.
domain assumption The control-affine system is controllable and Fbar=[Fc|F] is full rank with Fc spanning an orthonormal orthogonal complement.
Stated in Preliminaries; needed for G to be nonsingular and for the completion-square identity (18).
ad hoc to paper The coupled PDE (19)-(20) has a solution for all s and converges to a stationary point.
No proof given; the paper says 'If the trajectories converge...' and lists convergence as future work.
ad hoc to paper Free boundary conditions on mu and pdepe integration produce the intended saddle.
Section IV-A states free boundary conditions were applied, without theoretical backing.

invented entities (2)

dual trajectory mu(t) in R^(n-m)
purpose: Enforces dynamics constraints Fc^dagger (x_dot - Fd) = 0 in inadmissible control directions.
Mathematical multiplier variable; no direct physical observable.
kinematic dual trajectory mu_c_j(t)
purpose: Enforces inequality kinematic constraints via augmented Lagrangian in Eq. (27).
Optimization variable; no independent measurement.

how reviews work

0 comments

Cite this review

Pith. "Pith review of EL-AGHF: Extended Lagrangian Affine Geometric Heat Flow." pith.science (2026). https://pith.science/paper/CWLLO526

@misc{pith2026250524751,
  author       = {Pith},
  title        = {Pith review of: EL-AGHF: Extended Lagrangian Affine Geometric Heat Flow},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CWLLO526}},
  note         = {Machine review of arXiv:2505.24751}
}

read the original abstract

We propose a constrained Affine Geometric Heat Flow (AGHF) method that evolves so as to suppress the dynamics gaps associated with inadmissible control directions. AGHF provides a unified framework applicable to a wide range of motion planning problems, including both holonomic and non-holonomic systems. However, to generate admissible trajectories, it requires assigning infinite penalties to inadmissible control directions. This design choice, while theoretically valid, often leads to high computational cost or numerical instability when the penalty becomes excessively large. To overcome this limitation, we extend AGHF in an Augmented Lagrangian method approach by introducing a dual trajectory related to dynamics gaps in inadmissible control directions. This method solves the constrained variational problem as an extended parabolic partial differential equation defined over both the state and dual trajectorys, ensuring the admissibility of the resulting trajectory. We demonstrate the effectiveness of our algorithm through simulation examples.

Figures

Figures reproduced from arXiv: 2505.24751 by the authors.

**Figure 3.** Snapshots at T = 0, 0.25, 0.5, 0.75, 1.0 of trajectories generated by EL-AGHF and AGHF for the unconstrained mid-air motion model. The dashed line represents the snapshot of the solution trajectory, while the dotted line shows the trajectory snapshot obtained by integrating the system dynamics under the corresponding control input. The red curves indicate the trajectories of the center of mass (COM), and the red dot… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

17 extracted references · 13 canonical work pages

[17]

Constrained differential optimization,

J. Platt and A. Barr, “Constrained differential optimization,” in Neural Information Processing Systems , 1987

work page 1987
[1]

Affine geometric heat flow and motion planning for dynamic systems,

S. Liu, Y . Fan, and M.-A. Belabbas, “Affine geometric heat flow and motion planning for dynamic systems,” IFAC-PapersOnLine, vol. 52, no. 16, pp. 168–173, 2019

work page 2019
[2]

New method for motion planning for non-holonomic systems using partial differential equations,

M. A. Belabbas and S. Liu, “New method for motion planning for non-holonomic systems using partial differential equations,” in 2017 American Control Conference (ACC) , IEEE, 2017, pp. 4189–4194

work page 2017
[3]

Geometric heat flow method for legged locomotion planning,

Y . Fan, S. Liu, and M.-A. Belabbas, “Geometric heat flow method for legged locomotion planning,” IEEE Control Systems Letters , vol. 5, no. 3, pp. 941–946, 2020

work page 2020
[4]

Bring the heat: Rapid trajectory optimization with pseudospectral techniques and the affine geometric heat flow equation,

C. E. Adu, C. E. R. Chuquiure, B. Zhang, and R. Vasudevan, “Bring the heat: Rapid trajectory optimization with pseudospectral techniques and the affine geometric heat flow equation,” IEEE Robotics and Automation Letters , 2025

work page 2025
[5]

Phasing Through the Flames: Rapid Motion Planning with the AGHF PDE for Arbitrary Objective Functions and Constraints

C. E. Adu, C. E. R. Chuquiure, Y . Zhou, et al., “Phasing through the flames: Rapid motion planning with the aghf pde for arbitrary ob- jective functions and constraints,” arXiv preprint arXiv:2505.01589, 2025

work page Pith review arXiv 2025
[6]

Geometric fabrics: Generalizing classical mechanics to capture the physics of behavior,

K. Van Wyk, M. Xie, A. Li, et al., “Geometric fabrics: Generalizing classical mechanics to capture the physics of behavior,” IEEE Robotics and Automation Letters , vol. 7, no. 2, pp. 3202–3209, 2022

work page 2022
[7]

Geometric fabrics: A safe guiding medium for policy learning,

K. Van Wyk, A. Handa, V . Makoviychuk, Y . Guo, A. Allshire, and N. D. Ratliff, “Geometric fabrics: A safe guiding medium for policy learning,” in 2024 IEEE International Conference on Robotics and Automation (ICRA), IEEE, 2024, pp. 6537–6543

work page 2024

Show all 17 references

[8]

Geometric fabrics for the acceleration-based design of robotic motion,

M. Xie, K. Van Wyk, A. Li, et al. , “Geometric fabrics for the acceleration-based design of robotic motion,” arXiv preprint arXiv:2010.14750, 2020

2010 arXiv
[9]

Riemannian motion policies,

N. D. Ratliff, J. Issac, D. Kappler, S. Birchfield, and D. Fox, “Riemannian motion policies,” arXiv preprint arXiv:1801.02854 , 2018

2018 arXiv
[10]

Mid-air motion planning of robot using heat flow method with state constraints,

Y . Fan, S. Liu, and M.-A. Belabbas, “Mid-air motion planning of robot using heat flow method with state constraints,” Mechatronics, vol. 66, p. 102 323, 2020

2020
[11]

Geometric motion planning for affine control systems with indefinite boundary conditions and free terminal time,

S. Liu, Y . Fan, and M.-A. Belabbas, “Geometric motion planning for affine control systems with indefinite boundary conditions and free terminal time,” arXiv preprint arXiv:2001.04540 , 2020

2001 arXiv
[12]

Motion planning for control-affine systems by geometric heat flow with fine-tuning,

J. Chen and S. Liu, “Motion planning for control-affine systems by geometric heat flow with fine-tuning,” in 2024 IEEE International Conference on Unmanned Systems (ICUS) , IEEE, 2024, pp. 113– 118

2024
[13]

The second variation,

B. van Brunt, “The second variation,” The Calculus of Variations , pp. 221–260, 2004

2004
[14]

Holonomic and nonholonomic constraints,

B. van Brunt, “Holonomic and nonholonomic constraints,” The calculus of variations , pp. 119–133, 2004

2004
[15]

Liberzon, Calculus of variations and optimal control theory: a concise introduction

D. Liberzon, Calculus of variations and optimal control theory: a concise introduction. Princeton university press, 2011

2011
[16]

A homotopy method for motion planning,

S. Liu and M. A. Belabbas, “A homotopy method for motion planning,” arXiv preprint arXiv:1901.10094 , 2019

1901 arXiv

Pith tools

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

[1] [17]

Constrained differential optimization,

J. Platt and A. Barr, “Constrained differential optimization,” in Neural Information Processing Systems , 1987

work page 1987

[2] [1]

Affine geometric heat flow and motion planning for dynamic systems,

S. Liu, Y . Fan, and M.-A. Belabbas, “Affine geometric heat flow and motion planning for dynamic systems,” IFAC-PapersOnLine, vol. 52, no. 16, pp. 168–173, 2019

work page 2019

[3] [2]

New method for motion planning for non-holonomic systems using partial differential equations,

M. A. Belabbas and S. Liu, “New method for motion planning for non-holonomic systems using partial differential equations,” in 2017 American Control Conference (ACC) , IEEE, 2017, pp. 4189–4194

work page 2017

[4] [3]

Geometric heat flow method for legged locomotion planning,

Y . Fan, S. Liu, and M.-A. Belabbas, “Geometric heat flow method for legged locomotion planning,” IEEE Control Systems Letters , vol. 5, no. 3, pp. 941–946, 2020

work page 2020

[5] [4]

Bring the heat: Rapid trajectory optimization with pseudospectral techniques and the affine geometric heat flow equation,

C. E. Adu, C. E. R. Chuquiure, B. Zhang, and R. Vasudevan, “Bring the heat: Rapid trajectory optimization with pseudospectral techniques and the affine geometric heat flow equation,” IEEE Robotics and Automation Letters , 2025

work page 2025

[6] [5]

Phasing Through the Flames: Rapid Motion Planning with the AGHF PDE for Arbitrary Objective Functions and Constraints

C. E. Adu, C. E. R. Chuquiure, Y . Zhou, et al., “Phasing through the flames: Rapid motion planning with the aghf pde for arbitrary ob- jective functions and constraints,” arXiv preprint arXiv:2505.01589, 2025

work page Pith review arXiv 2025

[7] [6]

Geometric fabrics: Generalizing classical mechanics to capture the physics of behavior,

K. Van Wyk, M. Xie, A. Li, et al., “Geometric fabrics: Generalizing classical mechanics to capture the physics of behavior,” IEEE Robotics and Automation Letters , vol. 7, no. 2, pp. 3202–3209, 2022

work page 2022

[8] [7]

Geometric fabrics: A safe guiding medium for policy learning,

K. Van Wyk, A. Handa, V . Makoviychuk, Y . Guo, A. Allshire, and N. D. Ratliff, “Geometric fabrics: A safe guiding medium for policy learning,” in 2024 IEEE International Conference on Robotics and Automation (ICRA), IEEE, 2024, pp. 6537–6543

work page 2024