Pith. sign in

REVIEW 5 major objections 6 minor 2 cited by

Linear Supervision for Nonlinear, High-Dimensional Neural Control and Differential Games

T0 review · 5 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read Training a neural value function with an extra pull toward the linearized game solution makes high-dimensional nonlinear control and differential games both faster and more accurate to learn.

desk verdict A well-posed idea with real empirical gains, but the theory underpinning the linear supervisor is narrower than the claims, and the experimental reporting needs more rigor. read the letter →

arxiv 2412.02033 v3 pith:KMUPEUDB submitted 2024-12-02 math.OC cs.SYeess.SY

classification math.OCcs.SYeess.SY MSC 49L2549N7068T07
keywords Hamilton-JacobiequationsdifferentialgamesreachabilityvaluefunctionapproximationneuralnetworksHopfformulahigh-dimensionalcontrollinearization
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

The paper proposes that a neural network trained to solve a nonlinear Hamilton-Jacobi PDE can be made faster and more accurate by adding a supervision term that pulls the network toward the solution of the corresponding linearized game, computed cheaply via the Hopf formula. Two training programs are defined: one anneals from the linear solution into the nonlinear PDE loss, and one learns the value of an augmented game whose dynamics interpolate between linear and nonlinear, with the linear value acting as a boundary condition. In a 50-dimensional publisher-subscriber differential game, the augmented program raises intersection-over-union by 2.4 times and lowers mean-squared error by 23.7 times relative to PDE-only training, while the decayed program runs about 20 times faster. In a 10-dimensional quadrotor avoidance problem, the decayed schemes recover about 95% of the conformally expanded safe volume versus 81% for the baseline, and cut false positives from 1.86% to as low as 0.23%.

What carries the argument

The central object is the linear supervision loss $L_{LS}(\theta) = \rho\|V_\theta - V_\ell\| + \rho_g\|\nabla_x V_\theta - \nabla_x V_\ell\|$, which is added to the Hamilton-Jacobi PDE residual loss $L_{PDE}$. The linear value $V_\ell$ is the value of the linearized game computed by the Hopf formula, giving a cheap and globally smooth proxy for the true value. The two programs differ in how the losses are combined: the decayed program multiplies $L_{LS}$ by $(1-\lambda_k)$ and $L_{PDE}$ by $\lambda_k$ with $\lambda_k$ increasing from 0 to 1, while the augmented program learns the value $V_\lambda$ of a game whose dynamics are $(1-\lambda)\ell + \lambda f$, with $V_\lambda = V_\ell$ at $\lambda=0$ and $V_\lambda = V$ at $\lambda=1$. The augmentation makes the linear and PDE losses share a set of global minimizers, so the network's task is refinement rather than search from scratch.

What would settle it

Train both semi-supervision programs on a high-dimensional nonlinear system where the linearization error is known to be large, for example a 50-D publisher-subscriber game with large $\alpha$ and $\beta$ so that $\delta^*$ from Theorem 1 is comparable to the value range, and compare intersection-over-union and mean-squared error against the PDE-only baseline; if the supervised programs do not beat the baseline, the claim that linear supervision reliably improves learned value functions is refuted.

Watch

Extended reading notes

Core claim

The paper's central claim is that the value function of a nonlinear differential game can be learned more reliably by using the value function of a linearly approximated game as a structured training signal rather than training only against the nonlinear PDE residual. The linear value $V_\ell$, obtained with the Hopf formula at orders-of-magnitude lower cost than grid-based dynamic programming, is close enough to the true value $V$ over the region of interest that it can serve as a proxy target: the network first approximates $V_\ell$, then refines toward $V$. The paper proves a bound on $|V - V_\ell|$ in terms of the maximum difference between the nonlinear and linear dynamics along relevant trajectories, and shows the bound vanishes at the linearization operating point. Empirically, both proposed programs beat the PDE-loss baseline on a 50-dimensional benchmark and a 10-dimensional quadrotor problem, with the augmented game giving the largest accuracy gains and the decayed scheme the largest speed gains.

Load-bearing premise

The methods depend on the linearized value $V_\ell$ being a good approximation of the true nonlinear value $V$ over the region used for training; if the nonlinearity is strong enough to make the linearization error large, the supervision can pull the network away from the true solution instead of toward it.

Editorial extensions

If this is right

  • The learning problem shifts from generating a value function from scratch to refining a partially correct one, so the time-curriculum used by prior learned Hamilton-Jacobi solvers can be dropped; in the 50-D benchmark the decayed program completes in about one twentieth of the baseline time.
  • The augmented program's accuracy gains (2.4x intersection-over-union, 23.7x lower mean-squared error) show that adding the $\lambda=0$ linear boundary condition supplies structure that helps the network approximate the $\lambda=1$ nonlinear solution.
  • The drop in false positives from 1.86% to 0.23% in the quadrotor task means linear supervision counteracts the optimistic bias that makes high-dimensional learned value functions mark unsafe states as safe.
  • When the true solution is mildly nonlinear, the decayed scheme performs best because it polishes the linear solution cheaply; when the nonlinearity is strong, the augmented scheme should be more robust.

Reading between the lines

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

  • The same supervision idea should transfer to other cheap approximate value functions, such as reduced-order models or coarser-grid solutions; the decayed schedule would likely retain most of its acceleration for any structured prior.
  • A practical sanity check emerges from the augmented construction: after training, the network's prediction on the $\lambda=0$ slice should match $V_\ell$; a significant mismatch would indicate the linear boundary condition was not internalized.
  • The theoretical bound's assumptions (min-over-time equality and confinement to $\bar S$) are restrictive; if they fail for a given system, the empirical benefit could persist even where the proof does not apply, so practitioners may need to treat the bound as qualitative guidance rather than a certificate.
  • One testable extension is to make the supervision weight $\rho$ depend on a local estimate of the linearization error $\delta^*$, so the network trusts $V_\ell$ only in regions where it is known to be accurate.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes to accelerate and improve neural solutions of high-dimensional, nonlinear Hamilton-Jacobi reachability problems by supervising a neural value network with V_ell, the value of a linearized (or Hopf-formula-solvable) game. Two loss programs are introduced: a decayed linear semi-supervision loss (Definition 2) that anneals from the supervision loss to the PDE residual loss, and a nonlinear-spectrum augmented loss (Definition 4) that trains a network on an augmented state (x, lambda) with supervision on the lambda=0 slice and PDE loss everywhere. The methods are evaluated on a 50-D publisher-subscriber differential game with dynamic-programming ground truth and on a 10-D quadrotor collision-avoidance problem, reporting improved IoU, MSE, recovered safe volume, and reduced false positives. The paper also provides Theorem 1 and Corollary 1 to justify closeness of V and V_ell, and Theorem 2 for the augmented-game construction.

Significance. If substantiated, the central idea is valuable: cheap linear/Hopf solutions are used as inductive bias for learned nonlinear HJ solvers, potentially removing the time-curriculum bottleneck and improving accuracy in high dimensions. The 50-D benchmark scored against a dynamic-programming ground truth is a genuine strength, and the augmented-game construction in Theorem 2 is a clean and useful addition. The paper is also honest about some of its limitations, explicitly noting in Section 3.2 that there is no guarantee of a globally minimizing path and in Section 4.2 that the quadrotor solution is not particularly nonlinear. However, the theoretical support covers only value closeness under assumptions that are not verified, no result bounds the gradient error that the supervision loss explicitly fits, and the empirical claims rest on single runs with per-problem hyperparameter tuning. The idea is promising, but the current evidence does not yet establish the full scope claimed in the abstract and conclusion.

major comments (5)
  1. [Theorem 1 / Appendix 6.1] The bound |V - V_ell| <= epsilon* in Eq. (10) depends on two assumptions stated only in the proof: the min-over-time equality V(x,t) = min_tau sup_d inf_u J_T(x(tau)) for the states considered, and the confinement of all relevant trajectories to the set S_bar(tau). Neither assumption is verified for the 50-D or 10-D experiments, and Corollary 1 gives closeness only in the limit m -> m0. Since the training distributions in Section 4 cover states far from the operating point, Theorem 1 does not justify the use of V_ell on the actual training region.
  2. [Definition 1 / Eq. (13)] The linear supervision loss explicitly fits the gradient, LLS = rho ||V_theta - V_ell|| + rho_g ||grad V_theta - grad V_ell||, but no theorem or experiment bounds ||grad V - grad V_ell||. In the reachability formulation of Section 2 the optimal control is u* = argmin max <grad V, f>, so an incorrect supervised gradient can yield incorrect controls even when the value itself is close. The paper needs either a gradient-error bound under the assumptions of Theorem 1 or an empirical evaluation of control/policy error, ideally in a strongly nonlinear regime.
  3. [Definition 2 / Section 6.5] The decayed program is described as transitioning from the linear solution to the nonlinear solution as lambda_k increases, but Section 6.5 reports lambda_K = 0.6 for the LSS Decay method, so the final loss still contains 40% linear supervision and never reaches the pure PDE objective. In addition, Section 3.2 states that there is no guarantee of a globally minimizing path between LLS and L_PDE for lambda_0 -> lambda_K. Together these points mean that the mechanism claimed for the decayed method is asserted rather than established.
  4. [Section 6.5 / Tables 1 and Fig. 2] The empirical claims rest on point estimates from single runs. The training details state that experiments were chosen because they 'performed best' and that a coarse parameter search was used to select lambda_K, rho, and rho_g, but no seeds, no error bars, and no ablations are reported. The headline numbers (2.4x IoU, 23.7x MSE, 20x speedup) are therefore not yet established as significant or robust to hyperparameter choice.
  5. [Section 4.2 / Conclusion] The paper's own Section 4.2 says the quadrotor solution 'is not particularly nonlinear,' and the 50-D benchmark is a decomposable publisher-subscriber game whose value has the special additive structure of Remark 1. No experiment is run in a regime where V_ell is known to be a poor approximation of V. The conclusion that the augmented method should be preferred 'if the problem is very nonlinear' is therefore an extrapolation beyond the demonstrated evidence.
minor comments (6)
  1. [Eq. (9)] The notation H^+-_{ell+epsilon} with nested plus/minus signs is ambiguous; H+ and H- should be defined explicitly as H_ell + max_epsilon <p,epsilon> and H_ell - max_epsilon <p,epsilon> (or with min, as appropriate).
  2. [Definition 1 / Eq. (13)] The norms in the supervision loss are not specified; please state whether these are L2 norms, weighted norms, or something else.
  3. [Corollary 1 proof] The integral notation 'Z s t' is malformed and should be written as an integral from t to s; there is also an inconsistent use of s as both the integration variable and the final time.
  4. [Section 6.5] The text says 'our fork of the existing DeepReach software may be found here,' but no URL is provided; a link or repository identifier is needed for reproducibility.
  5. [Throughout] There are several typos and formatting inconsistencies, including 'Aknowledgements', 'IOU' versus 'IoU', and 'assm.'; these should be cleaned up in revision.
  6. [Figure 2] The runtime panel should state units and clarify that the reported runtimes include the time to generate the linear supervisor, as described in Section 6.5.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the linear supervisor is computed independently and all reported gains are measured against external ground truth, not against the supervised target.

full rationale

The central claim is that adding LLS(θ)=ρ‖Vθ−Vℓ‖+ρg‖∇Vθ−∇Vℓ‖ to the PDE loss improves learning of the nonlinear value function. The supervised target Vℓ is generated in Sec. 3.4 from the Hopf formula for the linearized dynamics, or by a separately trained network on the linear system; it is not defined in terms of the nonlinear value V or the learned network Vθ. The 50-D evaluation in Sec. 4.1 scores against dynamic-programming ground truth obtained via the exact decomposition in Remark 1, and the 10-D quadrotor evaluation in Sec. 4.2 uses roll-outs and conformal expansion. Neither metric is a function of Vℓ or of the LLS term, so the reported IOU, MSE, and volume improvements are not forced by construction. Theorem 1 is offered as theoretical motivation; its proof cites Sharpless et al. (2024a) for a trajectory-matching lemma. That is a self-citation, but it is not a definitional reduction: the cited lemma asserts existence of a linear-with-error trajectory matching the nonlinear one, not that Vθ equals Vℓ or that the test metrics equal the loss terms. The paper also acknowledges in Sec. 3.2 that there is no guaranteed globally minimizing path between LLS and LPDE, which is an honest limitation rather than a hidden circular assumption. Hyperparameters and loss schedules are tuned on the demonstration problems, which raises a generalization risk but is not circularity. Overall, no prediction in the paper is equivalent by construction to a fitted input.

Assumptions & free parameters 5 free parameters · 5 assumptions · 1 invented entities

The main scientific content is an empirical algorithm. Its success relies on tuned hyperparameters, an unproven optimization landscape assumption, and the closeness of the linearized value to the nonlinear value. The only invented entity is the lambda augmentation, which is a training device.

free parameters (5)
  • lambda_K (final decay weight for LSS-D) = 0.6
    Coarse search for the LSS Decay program in the quadrotor experiment (Sec. 6.5).
  • rho (value supervision weight) = 0.1
    Coarse search for LSS Decay (Sec. 6.5).
  • rho_g (gradient supervision weight) = 0.2
    Coarse search for LSS Decay (Sec. 6.5).
  • I_start, I_end (adaptive weighting bounds) = 10.0, 1.0
    Chosen empirically for the adaptive weighting scheme (Sec. 6.7).
  • learning rates, batch sizes, iteration counts = e.g., lr=5e-6 or 1e-5, batch 60k/65k/10k, iters 300k/100k/10k/60k
    Training hyperparameters selected per program and per problem (Sec. 6.5); no sensitivity analysis is reported.
assumptions (5)
  • standard math The HJ-PDE viscosity solution theory (Evans-Souganidis) holds for the games and augmented game considered.
    Used to justify the PDE loss for training and for Theorem 2 (Sec. 2.1, 3.3).
  • standard math The Hopf formula of Darbon and Osher applies to the linearized game, requiring convex JT and convex Hamiltonian H_ell.
    Basis for computing the linear supervisor V_ell (Sec. 2.3, Eq. 8).
  • domain assumption The sinusoidal (SIREN) network with 3 layers and 512 neurons can represent V and V_lambda accurately.
    The paper adopts DeepReach's architecture (Sec. 6.5) without capacity analysis.
  • ad hoc to paper The decayed loss schedule has a beneficial optimization path.
    Explicitly stated in Sec. 3.2 as unproven: 'there is no guarantee of a globally minimizing path between LLS and LPDE for lambda_0 to lambda_K, but empirically this works well'.
  • domain assumption Trajectories relevant to the value remain in the set S_bar(gamma) so that Theorem 1's error bound delta* is finite and applicable.
    Implicit in the proof of Theorem 1 (Appendix 6.1) and not verified for the benchmark problems.
invented entities (1)
  • lambda (nonlinear spectrum augmentation coordinate)
    purpose: Adds a scalar state with zero dynamics to interpolate continuously between the linear system (lambda=0) and the nonlinear system (lambda=1), producing V_lambda for training.
    A computational state-space augmentation introduced in Def. 3; it has no physical meaning and no falsifiable prediction outside the method.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Linear Supervision for Nonlinear, High-Dimensional Neural Control and Differential Games." pith.science (2026). https://pith.science/paper/KMUPEUDB

@misc{pith2026241202033,
  author       = {Pith},
  title        = {Pith review of: Linear Supervision for Nonlinear, High-Dimensional Neural Control and Differential Games},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KMUPEUDB}},
  note         = {Machine review of arXiv:2412.02033}
}
read the original abstract

As the dimension of a system increases, traditional methods for control and differential games rapidly become intractable, making the design of safe autonomous agents challenging in complex or team settings. Deep-learning approaches avoid discretization and yield numerous successes in robotics and autonomy, but at a higher dimensional limit, accuracy falls as sampling becomes less efficient. We propose using rapidly generated linear solutions to the partial differential equation (PDE) arising in the problem to accelerate and improve learned value functions for guidance in high-dimensional, nonlinear problems. We define two programs that combine supervision of the linear solution with a standard PDE loss. We demonstrate that these programs offer improvements in speed and accuracy in both a 50-D differential game problem and a 10-D quadrotor control problem.

Figures

Figures reproduced from arXiv: 2412.02033 by the authors.

Figure 1
Figure 1. Demonstration of Thm. 2 [Vλ] and Cor. 1 On top, the true value of Vλ at t = 1 for the problem posed in (16) with N = 3 along the range of λ is given. Note the smooth change from λ = 0, where Vλ = Vℓ, to λ = 1, where Vλ = V . In the bottom row, the error between Vλ and Vℓ is plotted as λ increases. Note the gradual increase in error and the large regions of V with low error. Definition 4 (Linear Semi-Supervision Loss… view at source ↗
Figure 2
Figure 2. 50-D Benchmark Result Comparison On the left, a slice of the learned solution for four variations (columns) of (21) where (α, β) ∈ {(20, 0),(−20, 0),(−20, 20),(10, −10)} is shown for each proposed method (rows), and the ground truth zero-level set is overlaid in black. On the right, the IOU, MSEs, and run time are given for each of the variations and methods. with ui ∈ {|ui | ≤ 1}, di ∈ {|di | ≤ 1} and a, b, c, α, β… view at source ↗
Figure 3
Figure 3. 10-D Quadrotor Result Comparison In the upper-left, the problem in which the drone is flying toward an obstacle is depicted with two trajectories demonstrating success and failure. On the right, slices of the sub-zero level set of the learned value that approximate the unsafe set are shown (gold), along with the 99.9%-confidence conformal expansion of the learned set (teal) and a sample of the roll-outs (blue if saf… view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Bridging Model Predictive Control and Deep Learning for Scalable Reachability Analysis

    cs.RO 2025-05 conditional novelty 7.0 of 10

    MPC-generated approximate value labels guide a DeepReach-style network to learn Hamilton-Jacobi reachability solutions, yielding larger verified safe sets in 2D, 7D, 13D, and 40D systems.

  2. Reachability Barrier Networks: Learning Hamilton-Jacobi Solutions for Smooth and Flexible Control Barrier Functions

    cs.RO 2025-05 conditional novelty 5.0 of 10

    RBN, a physics-informed neural network, approximates control barrier value functions with smooth gradients, adjustable conservativeness, and conformal prediction based safety coverage.

Reference graph

Works this paper leans on

43 extracted references · 33 canonical work pages · cited by 2 Pith papers

  1. [1]

    o der, J. P. and Findeisen, R. and Nagy, Z. and Allg \

    11em plus .33em minus .07em @technote 4000 4000 100 4000 4000 500 `\.=1000 = #1 #1 #1 0pt [0pt][0pt] #1 * \| ** #1 \@IEEEauthorblockNstyle \@IEEEauthorblockAstyle \@IEEEauthordefaulttextstyle \@IEEEauthorblockconfadjspace -0.25em \@IEEEauthorblockNtopspace 0.0ex \@IEEEauthorblockAtopspace 0.0ex \@IEEEauthorblockNinterlinespace 2.6ex \@IEEEauthorblockAinte...

  2. [2]

    A general hamilton-jacobi framework for non-linear state-constrained control problems

    Albert Altarovici, Olivier Bokanowski, and Hasnaa Zidani. A general hamilton-jacobi framework for non-linear state-constrained control problems. ESAIM: Control, Optimisation and Calculus of Variations, 19 0 (2): 0 337--357, 2013

  3. [3]

    Reachability analysis of nonlinear systems with uncertain parameters using conservative linearization

    Matthias Althoff, Olaf Stursberg, and Martin Buss. Reachability analysis of nonlinear systems with uncertain parameters using conservative linearization. In 2008 47th IEEE Conference on Decision and Control, pages 4042--4048. IEEE, 2008

  4. [4]

    Bansal, M

    S. Bansal, M. Chen, J. F. Fisac, and C. J. Tomlin. Safe Sequential Path Planning of Multi-Vehicle Systems Under Presence of Disturbances and Imperfect Information . Proc. American Control Conference, 2017

  5. [5]

    DeepReach : A deep learning approach to high-dimensional reachability

    Somil Bansal and Claire Tomlin. DeepReach : A deep learning approach to high-dimensional reachability. In IEEE International Conference on Robotics and Automation (ICRA), 2021

  6. [6]

    Elementary differential equations and boundary value problems

    William E Boyce, Richard C DiPrima, and Douglas B Meade. Elementary differential equations and boundary value problems. John Wiley & Sons, 2021

  7. [7]

    Exact and Efficient Hamilton-Jacobi Reachability for Decoupled Systems

    Mo Chen and Claire Tomlin. Exact and Efficient Hamilton-Jacobi Reachability for Decoupled Systems . In Conference on Decision and Control, 12 2015

  8. [8]

    Safe Platooning of Unmanned Aerial Vehicles via Reachability

    Mo Chen, Qie Hu, Casey Mackin, Jaime Fisac, and Claire J Tomlin. Safe Platooning of Unmanned Aerial Vehicles via Reachability . In Conference on Decision and Control, 2015

Show all 43 references
  1. [9]

    Decomposition of reachable sets and tubes for a class of nonlinear systems

    Mo Chen, Sylvia L Herbert, Mahesh S Vashishtha, Somil Bansal, and Claire J Tomlin. Decomposition of reachable sets and tubes for a class of nonlinear systems. IEEE Transactions on Automatic Control, 63 0 (11): 0 3675--3688, 2018

  2. [10]

    Algorithm for overcoming the curse of dimensionality for time-dependent non-convex Hamilton--Jacobi equations arising from optimal control and differential games problems

    Yat Tin Chow, J \'e r \^o me Darbon, Stanley Osher, and Wotao Yin. Algorithm for overcoming the curse of dimensionality for time-dependent non-convex Hamilton--Jacobi equations arising from optimal control and differential games problems. Journal of Scientific Computing, 73: 0...

  3. [11]

    Algorithms for overcoming the curse of dimensionality for certain Hamilton--Jacobi equations arising in control theory and elsewhere

    J \'e r \^o me Darbon and Stanley Osher. Algorithms for overcoming the curse of dimensionality for certain Hamilton--Jacobi equations arising in control theory and elsewhere. Research in the Mathematical Sciences, 3 0 (1): 0 19, 2016

  4. [12]

    Overcoming the curse of dimensionality for some hamilton--jacobi partial differential equations via neural network architectures

    J \'e r \^o me Darbon, Gabriel P Langlois, and Tingwei Meng. Overcoming the curse of dimensionality for some hamilton--jacobi partial differential equations via neural network architectures. Research in the Mathematical Sciences, 7 0 (3): 0 20, 2020

  5. [13]

    Neural approximation of pde solutions: An application to reachability computations

    Badis Djeridane and John Lygeros. Neural approximation of pde solutions: An application to reachability computations. In Proceedings of the 45th IEEE Conference on Decision and Control, pages 3034--3039. IEEE, 2006

  6. [14]

    Finite-difference least square methods for solving hamilton-jacobi equations using neural networks, 2024

    Carlos Esteve-Yag \"u e, Richard Tsai, and Alex Massucco. Finite-difference least square methods for solving hamilton-jacobi equations using neural networks, 2024

  7. [15]

    L. C. Evans and P. E. Souganidis. Differential games and representation formulas for solutions of Hamilton-Jacobi-Isaacs equations. Indiana Univ. Math. J., 33 0 (5): 0 773--797, 1984 a

  8. [16]

    Differential games and representation formulas for solutions of Hamilton-Jacobi -isaacs equations

    Lawrence C Evans and Panagiotis E Souganidis. Differential games and representation formulas for solutions of Hamilton-Jacobi -isaacs equations. Indiana University mathematics journal, 33 0 (5): 0 773--797, 1984 b

  9. [17]

    Evans and Panagiotis E

    L.C. Evans and Panagiotis E. Souganidis. Differential games and representation formulas for solutions of Hamilton-Jacobi-Isaacs equations , 1984 c . ISSN 0022-2518

  10. [18]

    Bridging hamilton-jacobi safety analysis and reinforcement learning

    Jaime F Fisac, Neil F Lugovoy, Vicen c Rubies-Royo, Shromona Ghosh, and Claire J Tomlin. Bridging hamilton-jacobi safety analysis and reinforcement learning. In 2019 International Conference on Robotics and Automation (ICRA), pages 8550--8556. IEEE, 2019

  11. [19]

    Hamilton-jacobi reachability in reinforcement learning: A survey

    Milan Ganai, Sicun Gao, and Sylvia Herbert. Hamilton-jacobi reachability in reinforcement learning: A survey. IEEE Open Journal of Control Systems, 2024

  12. [20]

    Robust control lyapunov-value functions for nonlinear disturbed systems, 2024

    Zheng Gong and Sylvia Herbert. Robust control lyapunov-value functions for nonlinear disturbed systems, 2024. URL https://arxiv.org/abs/2403.03455

  13. [21]

    Solving high-dimensional partial differential equations using deep learning

    Jiequn Han, Arnulf Jentzen, and Weinan E. Solving high-dimensional partial differential equations using deep learning. Proceedings of the National Academy of Sciences, 115 0 (34): 0 8505--8510, 2018

  14. [22]

    Efficient and guaranteed hamilton-jacobi reachability via self-contained subsystem decomposition and admissible control sets

    Chong He, Zheng Gong, Mo Chen, and Sylvia Herbert. Efficient and guaranteed hamilton-jacobi reachability via self-contained subsystem decomposition and admissible control sets. IEEE Control Systems Letters, 2023

  15. [23]

    A differential game approach to planning in adversarial scenarios: A case study on capture-the-flag

    Haomiao Huang, J Ding, Wei Zhang, and C J Tomlin. A differential game approach to planning in adversarial scenarios: A case study on capture-the-flag . In International Conference on Robotics and Automation (ICRA), pages 1451--1456, 2011

  16. [24]

    Robots that suggest safe alternatives

    Hyun Joe Jeong and Andrea Bajcsy. Robots that suggest safe alternatives. arXiv preprint arXiv:2409.09883, 2024

  17. [25]

    Kirchner, Robert Mar, Gary Hewer, Jerome Darbon, Stanley Osher, and Y

    Matthew R. Kirchner, Robert Mar, Gary Hewer, Jerome Darbon, Stanley Osher, and Y. T. Chow. Time-optimal collaborative guidance using the generalized Hopf formula. IEEE Control Systems Letters , 2 0 (2): 0 201--206, apr 2018. doi:10.1109/lcsys.2017.2785357. URL https://doi.org/...

  18. [26]

    Constrained polynomial zonotopes

    Niklas Kochdumper and Matthias Althoff. Constrained polynomial zonotopes. Acta Informatica, 60 0 (3): 0 279--316, 2023

  19. [27]

    Bridging the Gap Between Safety and Real-Time Performance in Receding-Horizon Trajectory Design for Mobile Robots

    Shreyas Kousik, Sean Vaskov, Fan Bu, Matthew Johnson-Roberson, and Ram Vasudevan. Bridging the Gap Between Safety and Real-Time Performance in Receding-Horizon Trajectory Design for Mobile Robots . arXiv, 2018. URL http://arxiv.org/abs/1809.06746

  20. [28]

    Dynamics and control of trajectory tubes

    Alexander B Kurzhanski. Dynamics and control of trajectory tubes. theory and computation. In 2014 20th International Workshop on Beam Dynamics and Optimization (BDO), pages 1--1. IEEE, 2014

  21. [29]

    Iterative method using the generalized Hopf formula: Avoiding spatial discretization for computing solutions of Hamilton-Jacobi equations for nonlinear systems

    Donggun Lee and Claire J Tomlin. Iterative method using the generalized Hopf formula: Avoiding spatial discretization for computing solutions of Hamilton-Jacobi equations for nonlinear systems. In 2019 IEEE 58th Conference on Decision and Control (CDC), pages 1486--1493. IEEE, 2019

  22. [30]

    Verification of neural reachable tubes via scenario optimization and conformal prediction

    Albert Lin and Somil Bansal. Verification of neural reachable tubes via scenario optimization and conformal prediction. arXiv preprint arXiv:2312.08604, 2023

  23. [31]

    Loss landscapes and optimization in over-parameterized non-linear systems and neural networks

    Chaoyue Liu, Libin Zhu, and Mikhail Belkin. Loss landscapes and optimization in over-parameterized non-linear systems and neural networks. Applied and Computational Harmonic Analysis, 59: 0 85--116, 2022

  24. [32]

    A hamilton-jacobi-bellman approach to ellipsoidal approximations of reachable sets for linear time-varying systems

    Vincent Liu, Chris Manzie, and Peter M Dower. A hamilton-jacobi-bellman approach to ellipsoidal approximations of reachable sets for linear time-varying systems. arXiv preprint arXiv:2401.06352, 2024

  25. [33]

    Majumdar and R

    A. Majumdar and R. Tedrake. Funnel libraries for real-time robust feedback motion planning . Int. J. Robotics Research, pages 947--982, Jun. 2017. doi:10.1177/0278364917712421

  26. [34]

    A time-dependent Hamilton-Jacobi formulation of reachable sets for continuous dynamic games

    Ian M Mitchell, Alexandre M Bayen, and Claire J Tomlin. A time-dependent Hamilton-Jacobi formulation of reachable sets for continuous dynamic games. IEEE Transactions on automatic control, 50 0 (7): 0 947--957, 2005 a

  27. [35]

    Mitchell, A.M

    I.M. Mitchell, A.M. Bayen, and C.J. Tomlin. A time-dependent Hamilton-Jacobi formulation of reachable sets for continuous dynamic games. IEEE Trans. Automatic Control, 50 0 (7): 0 947--957, 2005 b . doi:10.1109/TAC.2005.851439

  28. [36]

    Physics informed deep learning (part i): Data-driven solutions of nonlinear partial differential equations

    Maziar Raissi, Paris Perdikaris, and George Em Karniadakis. Physics informed deep learning (part i): Data-driven solutions of nonlinear partial differential equations. arXiv preprint arXiv:1711.10561, 2017

  29. [37]

    Generalized Hopf formulas for the nonautonomous Hamilton-Jacobi equation

    IV Rublev. Generalized Hopf formulas for the nonautonomous Hamilton-Jacobi equation. Computational Mathematics and Modeling, 11 0 (4): 0 391--400, 2000

  30. [38]

    Conservative linear envelopes for high-dimensional, Hamilton-Jacobi reachability for nonlinear systems via the Hopf formula, 2024 a

    Will Sharpless, Yat Tin Chow, and Sylvia Herbert. Conservative linear envelopes for high-dimensional, Hamilton-Jacobi reachability for nonlinear systems via the Hopf formula, 2024 a

  31. [39]

    State-augmented linear games with antagonistic error for high-dimensional, nonlinear hamilton-jacobi reachability

    Will Sharpless, Yat Tin Chow, and Sylvia Herbert. State-augmented linear games with antagonistic error for high-dimensional, nonlinear hamilton-jacobi reachability. arXiv preprint arXiv:2403.16982, 2024 b

  32. [40]

    Koopman-Hopf Hamilton-Jacobi reachability and control

    William Sharpless, Nikhil Shinde, Matthew Kim, Yat Tin Chow, and Sylvia Herbert. Koopman-Hopf Hamilton-Jacobi reachability and control. arXiv preprint arXiv:2303.11590, 2023

  33. [41]

    Implicit neural representations with periodic activation functions

    Vincent Sitzmann, Julien Martel, Alexander Bergman, David Lindell, and Gordon Wetzstein. Implicit neural representations with periodic activation functions. Advances in neural information processing systems, 33: 0 7462--7473, 2020

  34. [42]

    Global convergence of admm in nonconvex nonsmooth optimization

    Yu Wang, Wotao Yin, and Jinshan Zeng. Global convergence of admm in nonconvex nonsmooth optimization. Journal of Scientific Computing, 78: 0 29--63, 2019

  35. [43]

    Efficient backward reachability using the minkowski difference of constrained zonotopes

    Liren Yang, Hang Zhang, Jean-Baptiste Jeannin, and Necmiye Ozay. Efficient backward reachability using the minkowski difference of constrained zonotopes. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 41 0 (11): 0 3969--3980, 2022

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.