Pith. sign in

REVIEW 5 major objections 5 minor 3 cited by

Bridging Model Predictive Control and Deep Learning for Scalable Reachability Analysis

T0 review · 5 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read MPC-generated value labels steer neural reachability training to accurate, verifiable safe sets.

desk verdict Solid engineering contribution showing MPC-generated value labels improve neural HJB reachability learning, but the iterative refinement loop deserves deeper scrutiny. read the letter →

arxiv 2505.03830 v1 pith:ZI5TC64N submitted 2025-05-04 cs.RO cs.SYeess.SY

Zeyuan Feng , Le Qiu , Somil Bansal This is my paper · ORCID
classification cs.ROcs.SYeess.SY
keywords Hamilton-Jacobireachabilitybackwardreachabletubemodelpredictivecontrolphysics-informedneuralnetworksHJBvariationalinequalityvaluefunctionconformalpredictionsafetyverification
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

Hamilton–Jacobi (HJ) reachability gives a safety value function whose sub-zero level sets are the states from which failure is inevitable, but grid solvers scale exponentially with dimension and residual-only neural training often locks onto non-physical solutions. This paper tries to establish that approximate value labels generated cheaply by sampling-based model predictive control (MPC) — solving the same optimal control problem that underlies HJ reachability — can guide a neural network to the correct HJB variational inequality solution. The authors add these labels as a supervised loss alongside the PDE residual, re-fit the MPC dataset during training using the current value function as terminal cost, and verify the result with conformal prediction. On a 2D drone, a 13D quadrotor, a 7D F1Tenth car, and a 40D publisher-subscriber system, the claimed outcome is a substantially larger verified safe set than residual-only or distillation-only training. The payoff would be practical, verified reachability analysis for higher-dimensional nonlinear systems.

What carries the argument

The machinery is a MPC dataset generator (Algorithm 1) plus a three-phase training loop (Algorithm 2). Algorithm 1 samples control sequences around a nominal policy, rolls out discretized dynamics, and records the best worst-case safety cost $\hat V(t_i,x_i)=\max_u \min_{h} l(\xi_h)$, bootstrapping labels along the best trajectory; it is GPU-parallel and treats the learned $V_\theta$ as terminal cost when the horizon extends. Algorithm 2 then pretrains on these labels, performs curriculum training with combined loss $L = L_{\mathrm{PDE}} + \lambda L_{\mathrm{data}}$ and periodic dataset refinement every $H_R$ seconds, and fine-tunes with an asymmetric false-positive penalty. This machinery is what carries the paper's claim that approximate optimal-control data can steer a residual-trained network away from bad local solutions.

What would settle it

Take a low-dimensional system with a known ground-truth value function and a long-horizon avoidance problem where random-shooting MPC systematically misses the optimal trajectory (for example, a U-shaped obstacle requiring a detour longer than the sampling horizon). Train the proposed method with a deliberately biased initial value function (e.g., pre-trained to label all states safe) and start refinement immediately. If the recovered verified volume stays at or below the residual-only DeepReach baseline, the central claim that MPC labels guide the network to the true HJB-VI solution is falsified.

Watch

Extended reading notes

Core claim

Starting from the observation that the backward reachable tube is the solution of the optimal control problem $V(x,t) = \sup_{u(\cdot)} \min_{\tau\in[t,T]} l(\xi(\tau))$, the paper claims that a sampling-based MPC solver computing $\hat V$ at collocation points $(t_i,x_i)$ produces valuable semi-supervised labels for learning the HJB variational inequality $\min\{D_t V + H, l - V\}=0$ with terminal condition $V(x,T)=l(x)$. Training proceeds in three phases: supervised warm-start on the MPC dataset, curriculum training that jointly minimizes the PDE residual and a data-matching loss while periodically refining the MPC dataset with the current learned value function as terminal cost, and a fine-tuning phase that up-weights false-positive errors to keep the safe set conservative. The paper claims this hybrid training is more accurate and stable than residual-only DeepReach and more accurate than pure MPC distillation, with verified safe-set volumes of 24.62%, 93.69%, 76.08%, and 97.14% on the four case studies.

Load-bearing premise

The refinement loop assumes the current learned value function is already accurate enough to serve as the terminal cost and nominal policy when the MPC dataset is regenerated; if early curriculum labels are biased, the new MPC labels inherit that bias and the data loss can lock the network into the wrong reachable set.

Editorial extensions

If this is right

  • If the central claim is right, learned safe sets can be verified: the recovered verified volumes (24.62% on the vertical drone, 93.69% on the 13D quadrotor, 76.08% on F1Tenth, 97.14% on the 40D system) exceed every baseline tested.
  • Pure MPC distillation is insufficient: the same MPC labels without HJB residuals yield zero verified volume on both the drone and F1Tenth, so the PDE residual remains necessary.
  • The time curriculum is the load-bearing training stage: removing it collapses recovered volume to zero on the quadrotor and F1Tenth, with negligible cost savings.
  • Iterative dataset refinement is directly beneficial: on the vertical drone, keeping the MPC dataset fixed drops recovered volume from 24.62% to 20.04% and raises MSE to 0.2285.
  • The learned safety policy is usable for downstream filtering: on F1Tenth it turns a nominal policy with 100% collision rate and 36.17 m average travel into a collision-free filtered policy averaging 187.01 m.

Reading between the lines

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

  • A natural reading is that the approach inherits the failure profile of the MPC guide: wherever random-shooting MPC cannot approximate the optimal avoidance (long detours, narrow passages), the supervised loss could pull the network toward a suboptimal value function, which the paper's ablations do not isolate.
  • Because the method's cost scales with problem complexity rather than nominal dimension (7D F1Tenth costs more than 13D quadrotor), a promising extension is adaptive sampling of collocation points informed by PDE residual or label disagreement to cut MPC calls further.
  • The refinement loop suggests a self-consistency criterion: labels and residuals should agree as training converges, so monitoring the gap between $L_{\mathrm{data}}$ and $L_{\mathrm{PDE}}$ could detect when the MPC guides have gone stale or biased.
  • The same MPC-guidance scheme could in principle be ported to reach-avoid problems or disturbance-robust games by replacing the single-player maximization with a saddle-point robust MPC, though the paper explicitly leaves disturbances and long-horizon problems out.
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 / 5 minor

Summary. The paper proposes a hybrid approach to learning Hamilton-Jacobi reachability value functions in which a sampling-based MPC routine (Algorithm 1) generates approximate value-function labels at collocation points, and a neural network is trained with a weighted combination of the HJB-VI PDE residual (Eq. 10) and a data-supervision term (Eq. 9). Training proceeds through pretraining, curriculum training, and fine-tuning (Algorithm 2), with the MPC dataset periodically regenerated using the current learned value function as terminal cost. The learned value function is then processed through a conformal-prediction verification step to produce a high-confidence safe set. Experiments on four systems (2D vertical drone, 13D quadrotor, 7D F1Tenth, 40D publisher-subscriber) are reported, comparing against vanilla DeepReach, distillation, and neural CBF baselines, with an additional ablation of dataset size and training stages.

Significance. If validated, the core idea is valuable: it injects a strong optimal-control-based learning signal into physics-informed reachability, potentially mitigating the weak-supervision problem that causes residual-only training to converge to nonphysical solutions. The paper includes an open-source implementation and demonstrates large reported gains on high-dimensional examples (e.g., 93.69% vs 71.26% recovered volume on the 13D quadrotor). The conformal verification step is a useful practical addition, as it connects the learned value function to a probabilistic safety guarantee. However, the current validation has important gaps: single-seed main results, a missing benchmark against the closest label-based prior, an unexplained 167.14% volume for F1Tenth Neural CBF, and no analysis of the dataset-refinement feedback loop. The time-indexing inconsistency in Algorithm 1 is a serious technical issue that must be resolved before the method can be reproduced.

major comments (5)
  1. [Section III-A, Algorithm 1, Remark 2] The sampled time t_i is not used in the rollout, so the stored label V̂(t_i,x_i) is computed from a fixed-horizon H rollout starting at x_i at time 0, not from horizon T−t_i. Since the value function depends on the remaining horizon, the label is inconsistent with the stated (x,t) unless HΔ=T and t_i=0. The same issue occurs in Remark 2, where bootstrapped labels are assigned time Δ·h instead of t_i+Δ·h. Please clarify how t_i enters the label generation (e.g., set H per sample as (T−t_i)/Δ) or correct the time indexing; as written, Eq. (9) trains against labels that may not be evaluations of the target value function at the reported times.
  2. [Algorithm 2, Section IV-D] The dataset-refinement loop uses the current learned V_theta as terminal cost for regenerating D_MPC, with no mechanism or bound ensuring that the new labels are closer to the true value function than the labels they replace. If V_theta is over-optimistic, the refined labels inherit that bias and the supervised loss in Eq. (9) can reinforce it. The only ablation is a single vertical-drone run (MSE 0.2285 vs 0.009, volume 20.04% vs 24.62%) with no seeds or measurement of label quality. To support the claim that refinement 'mitigates convergence to local optima,' please provide multi-seed ablations on more than one system and, where ground truth exists, report the error of the regenerated labels versus the ground truth across refinement iterations.
  3. [Table II, Sections IV-B and IV-F] The F1Tenth entry for Neural CBF reports a recovered volume of 167.14%, which exceeds 100% of the state-space volume and is not a valid volume fraction. The table also reports single-run values for all methods, so the claim that the proposed approach 'consistently' achieves higher volumes is not backed by variance estimates. Please correct the metric (or explain what is being measured) and report mean and standard deviation over at least five seeds for the main comparisons, as is already done for the vertical-drone dataset-size ablation.
  4. [Sections I and IV-A] The introduction identifies the Hopf-formula semi-supervised method [32] as the closest label-based alternative to the proposed approach, but [32] is not included in the benchmark. Without a comparison to this method, the claim that the MPC-guided approach outperforms 'existing methods' is incompletely supported. Please add this baseline or justify its omission.
  5. [Section III-B, Eq. (12)] The false-positive fine-tuning loss is written as λ_FP∥V̂−Vθ(x,t)∥Vθ(x,t) when Vθ(x,t)≥0 and V̂<0, while the text states that the loss is amplified by the scalar λ_FP. The extra multiplicative Vθ(x,t) makes the amplification vanish at Vθ=0 and grow with the value magnitude, which is not what the text describes. Clarify whether this factor is intended, and if not, correct the equation, since the exact loss is essential for reproducibility.
minor comments (5)
  1. [Section IV-F] The word 'exasperated' should be 'exacerbated'.
  2. [Table II caption] The parenthetical entries for the vertical drone (31.28 and 56.51) are not explained in the table caption; add a note defining that these correspond to K=12.
  3. [References] References [32] and [33] point to the same arXiv paper; merge to avoid duplicate citation.
  4. [Section III-C] The confidence value β=10^{-16} is extremely small; please report the number of calibration samples used in the conformal procedure so the reader can judge the statistical validity.
  5. [Algorithm 2, line 7] The PDE collocation times are drawn uniformly in [t,T], but the boundary at t=T is explicitly enforced by the ansatz; it may be helpful to state this in the text.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: MPC guidance is generated from dynamics rollouts, residual loss is the independent HJB-VI equation, and final metrics are checked against ground-truth or conformal verification.

full rationale

The central derivation is self-contained rather than circular. The initial MPC dataset (Algorithm 1) is produced by simulating the discretized dynamics and evaluating the minimum of the safety function l along sampled control sequences; it does not depend on the learned value function V_theta. The PDE loss (10) is the standard HJB-VI residual, an independent equation that the learned function must satisfy, and the boundary-condition parameterization V_theta(x,t) = l(x) + (T-t)O_theta(x,t) is a standard ansatz, not a disguised form of the MPC label. Reported successes are measured either against numerically computed ground-truth value functions (2D vertical drone, 40D publisher-subscriber) or through conformal-prediction-verified safe sets computed from rollouts, both external to the training labels. The iterative refinement in Algorithm 2 does make the MPC labels a function of the current V_theta, since V_theta supplies the terminal cost and nominal policy for a short receding horizon. This creates a potential bias-inheritance or self-confirmation risk, but the label is not equal to V_theta by construction: it is obtained by a nontrivial min-cost roll-out over the discretized dynamics, and the combined loss still includes the independent PDE residual. The paper does not prove convergence of this fixed-point-style refinement, which is a correctness risk rather than a definitional circularity. Self-citations to DeepReach and conformal verification are used as baselines or external tools, and they do not force the outcome by themselves. No equation in the paper reduces a predicted quantity to its own input by definition.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a standard HJB-VI characterization, on the quality of the MPC-generated labels, and on the validity of the conformal verification procedure. The free parameters are mostly training hyperparameters chosen by hand; the most consequential are the refinement horizon H_R, the false-positive weight lambda_FP, and the MPC sampling settings N, R, Delta. No new physical entities are introduced.

free parameters (4)
  • lambda_FP (false-positive loss weight) = 100
    Set in Section IV-C for all experiments; amplifies loss on states the learned value function labels safe while MPC labels unsafe. It directly shapes the conservatism and reported safe-set volumes.
  • H_R (dataset refinement horizon) = 0.2 s
    Set in Section IV-C; determines when the MPC dataset is regenerated using the learned value function. It controls how often the circular self-supervision loop is applied.
  • MPC sampling hyperparameters N, R, Delta = N=100, R=10, Delta=0.02 s
    Defined in Section IV-C. These control the accuracy and cost of the approximate value labels; the perturbation variance sigma^2 is not specified.
  • Curriculum iterations N_c and lambda balancing decay = N_c per case; lambda <- 0.9 lambda + 0.1 ratio
    The curriculum schedule in Algorithm 2 and Table I is hand-chosen; the adaptive weight is a heuristic balancing scheme rather than a derived quantity.
assumptions (5)
  • standard math The HJB-VI in Eq. (5) correctly characterizes the BRT for the continuous-time system.
    Invoked in Section II-B via standard dynamic programming; the entire method inherits this characterization.
  • ad hoc to paper The approximate MPC labels are close enough to the true value function to provide a useful learning signal.
    Section III-A uses sampling-based MPC with N=100 and R=10; no bound on the suboptimality of the resulting V-hat is provided, and the paper acknowledges the labels are noisy.
  • ad hoc to paper The learned value function used in refinement is a valid terminal cost for longer-horizon MPC label generation.
    Algorithm 2 regenerates DMPC using V_theta; this assumes the current approximation is locally trustworthy, which is exactly what the method is trying to establish.
  • domain assumption The conformal-prediction procedure from [22] provides valid high-confidence safe sets for the learned neural policy.
    Section III-C applies the published verification method; its validity conditions are not restated or checked.
  • domain assumption First-order Euler discretization with Delta=0.02 s is an adequate surrogate for the continuous dynamics in label generation.
    Algorithm 1 propagates states with Euler steps; no discretization-error analysis is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bridging Model Predictive Control and Deep Learning for Scalable Reachability Analysis." pith.science (2026). https://pith.science/paper/ZI5TC64N

@misc{pith2026250503830,
  author       = {Pith},
  title        = {Pith review of: Bridging Model Predictive Control and Deep Learning for Scalable Reachability Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZI5TC64N}},
  note         = {Machine review of arXiv:2505.03830}
}
read the original abstract

Hamilton-Jacobi (HJ) reachability analysis is a widely used method for ensuring the safety of robotic systems. Traditional approaches compute reachable sets by numerically solving an HJ Partial Differential Equation (PDE) over a grid, which is computationally prohibitive due to the curse of dimensionality. Recent learning-based methods have sought to address this challenge by approximating reachability solutions using neural networks trained with PDE residual error. However, these approaches often suffer from unstable training dynamics and suboptimal solutions due to the weak learning signal provided by the residual loss. In this work, we propose a novel approach that leverages model predictive control (MPC) techniques to guide and accelerate the reachability learning process. Observing that HJ reachability is inherently rooted in optimal control, we utilize MPC to generate approximate reachability solutions at key collocation points, which are then used to tactically guide the neural network training by ensuring compliance with these approximations. Moreover, we iteratively refine the MPC generated solutions using the learned reachability solution, mitigating convergence to local optima. Case studies on a 2D vertical drone, a 13D quadrotor, a 7D F1Tenth car, and a 40D publisher-subscriber system demonstrate that bridging MPC with deep learning yields significant improvements in the robustness and accuracy of reachable sets, as well as corresponding safety assurances, compared to existing methods.

Figures

Figures reproduced from arXiv: 2505.03830 by the authors.

Figure 1
Figure 1. We propose a framework that efficiently generates approximate safety value function datasets using a sampling-based MPC approach and integrates these data labels to guide the learning of reachability solutions for high-dimensional autonomous systems. The learned value function is then verified using conformal prediction, providing probabilistic safety assurances for the system under the induced safe policy. ity, we … view at source ↗
Figure 2
Figure 2. Parameterized Vertical Drone: Value function slices at K = 12. The brown lines represent the ground and the ceiling – the failure set in this case. In (a), the solid black lines are the contours of ground-truth safe sets. In (b), all MPC data samples with K ∈ (11, 12) used for the proposed approach are shown. (c) illustrates the learned safe set using the neural CBF approach. For (d), (e), and (f), the dashed contou… view at source ↗
Figure 4
Figure 4. The x-axis shows the size of non-bootstrapped dataset. The blue line represents the mean recovered volumes along with their standard deviation across five random seeds, whereas the red line shows the mean false positive rates [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figures from the paper (4 more)
Figure 3
Figure 3. Figure 3: 13D Quadrotor: slices of learned value functions on the X-Y plane at (pz, qω, qx, qy, qz, vx, vy, vz, ωx, ωy, ωz) = (0.54, 0.44, −0.45, 0.27, −0.73, 5.00, −1.07, −3.34, 3.19, −2.80, 3.43). The brown circles represent the cylinder obstacle (not the failure set contour d…
Figure 5
Figure 5. Figure 5: F1Tenth: X-Y Plane Slices of the Learned Value Functions. The remaining state variables are (ϕ, v, θyaw, ωyaw, βslip) = (0, 8, 0, 0, 0). The track boundaries are shown as the brown contours. The black solid contours are the verified safe sets while the dashed contours …
Figure 6
Figure 6. Figure 6: Trajectories under the nominal controller and the safety filter with γ = 1. The green dot marks the starting points of the trajectories, while the red “X” denotes collisions. The filtered policy successfully completes an entire lap without collision [PITH_FULL_IMAGE:f…
Figure 7
Figure 7. Figure 7: Value function slices for the 40D system. The brown [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

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

  1. Forward Trajectory Steering for Hamilton-Jacobi Reachability Analysis

    eess.SY 2026-08 conditional novelty 6.0 of 10

    A simple forward-trajectory steering scheme for collocation sampling lets PINNs solve high-dimensional Hamilton-Jacobi reachability problems with accuracy comparable to or better than MPC-supervised solvers.

  2. Manifold-constrained Hamilton-Jacobi Reachability Learning for Decentralized Multi-Agent Motion Planning

    cs.RO 2025-11 conditional novelty 6.0 of 10

    HaMMAR learns manifold-constrained Hamilton-Jacobi reachability value functions and uses them for decentralized collision-free multi-robot motion planning under task constraints.

  3. Safe and Performant Deployment of Autonomous Systems via Model Predictive Control and Hamilton-Jacobi Reachability Analysis

    cs.RO 2025-06 conditional novelty 3.0 of 10

    Adding a Hamilton-Jacobi reachability safety value as a terminal constraint in model predictive control makes the controller recursively feasible and reduces safety violations in car and robot arm simulations.

Reference graph

Works this paper leans on

38 extracted references · 25 canonical work pages · cited by 3 Pith papers

  1. [1]

    Bansal, M

    S. Bansal, M. Chen, S. Herbert, and C. J. Tomlin. Hamilton-Jacobi reachability: A brief overview and re- cent advances. In IEEE Conference on Decision and Control, 2017

    work page 2017

  2. [2]

    Somil Bansal and Claire J. Tomlin. Deepreach: A deep learning approach to high-dimensional reachability. In 2021 IEEE International Conference on Robotics and Automation (ICRA) , pages 1817–1824. doi: 10.1109/ ICRA48506.2021.9561949

    arXiv 2021

  3. [3]

    On safety and liveness filtering using hamilton-jacobi reachability analysis

    Javier Borquez, Kaustav Chakraborty, Hao Wang, and Somil Bansal. On safety and liveness filtering using hamilton-jacobi reachability analysis. IEEE Transactions on Robotics, 2024

    2024

  4. [4]

    Optimizeddp: An efficient, user-friendly library for optimal control and dynamic programming

    Minh Bui, George Giovanis, Mo Chen, and Arrvindh Shriraman. Optimizeddp: An efficient, user-friendly library for optimal control and dynamic programming. arXiv preprint arXiv:2204.05520 , 2022

    arXiv 2022

  5. [5]

    Physics-informed neural networks (pinns) for fluid mechanics: A review

    Shengze Cai, Zhiping Mao, Zhicheng Wang, Minglang Yin, and George Em Karniadakis. Physics-informed neural networks (pinns) for fluid mechanics: A review. Acta Mechanica Sinica , 37(12):1727–1738, 2021

    2021

  6. [6]

    Reachability analysis for black-box dynamical systems

    Vamsi Krishna Chilakamarri, Zeyuan Feng, and Somil Bansal. Reachability analysis for black-box dynamical systems. arXiv preprint arXiv:2410.07796 , 2024

    arXiv 2024

  7. [7]

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

    Yat Tin Chow, J´erˆome Darbon, Stanley Osher, and Wotao Yin. Algorithm for overcoming the curse of dimension- ality for time-dependent non-convex Hamilton–Jacobi equations arising from optimal control and differential games problems. Journal of Scientific Computing , 73 (2-3):617–643, 2017. ISSN 0885-7474

    work page 2017

  8. [8]

    Algorithms for over- coming the curse of dimensionality for certain Hamil- ton–Jacobi equations arising in control theory and else- where

    J ´erˆome Darbon and Stanley Osher. Algorithms for over- coming the curse of dimensionality for certain Hamil- ton–Jacobi equations arising in control theory and else- where. Research in the Mathematical Sciences , 3(1):19,

    work page

Show all 38 references
  1. [9]

    Learning safe, generalizable perception- based hybrid control with certificates

    Charles Dawson, Bethany Lowenkamp, Dylan Goff, and Chuchu Fan. Learning safe, generalizable perception- based hybrid control with certificates. IEEE Robotics and Automation Letters , 7(2):1904–1911, 2022

    1904

  2. [10]

    Safe nonlinear control using robust neural lyapunov- barrier functions

    Charles Dawson, Zengyi Qin, Sicun Gao, and Chuchu Fan. Safe nonlinear control using robust neural lyapunov- barrier functions. In Aleksandra Faust, David Hsu, and Gerhard Neumann, editors, Proceedings of the 5th Conference on Robot Learning , volume 164 of Proceed- ings of Mac...

    2022

  3. [11]

    Paral- lelotope bundles for polynomial reachability

    Tommaso Dreossi, Thao Dang, and Carla Piazza. Paral- lelotope bundles for polynomial reachability. In Interna- tional Conference on Hybrid Systems: Computation and Control (HSCC), pages 297–306, 2016

    2016

  4. [12]

    Reach-avoid problems with time-varying dynam- ics, targets and constraints

    Jaime F Fisac, Mo Chen, Claire J Tomlin, and S Shankar Sastry. Reach-avoid problems with time-varying dynam- ics, targets and constraints. In Proceedings of the 18th international conference on hybrid systems: computation and control, pages 11–20, 2015

    2015

  5. [13]

    Bridging hamilton-jacobi safety analysis and reinforcement learn- ing

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

    2019

  6. [14]

    SpaceEx: Scalable verification of hybrid systems

    Goran Frehse, Colas Le Guernic, Alexandre Donz ´e, Scott Cotton, Rajarshi Ray, Olivier Lebeltel, Rodolfo Ripado, Antoine Girard, Thao Dang, and Oded Maler. SpaceEx: Scalable verification of hybrid systems. In International Conference on Computer Aided Verification , pages 379–

  7. [15]

    Convex computation of the region of attraction of polynomial control systems

    Didier Henrion and Milan Korda. Convex computation of the region of attraction of polynomial control systems. Transactions on Automatic Control (TAC) , 59(2):297– 312, 2013. ISSN 0018-9286

    2013

  8. [16]

    Convergence guarantees for neu- ral network-based hamilton-jacobi reachability

    William Hofgard. Convergence guarantees for neu- ral network-based hamilton-jacobi reachability. arXiv preprint arXiv:2410.02904, 2024

    2024 arXiv

  9. [17]

    Safety and liveness guaran- tees through reach-avoid reinforcement learning

    Kai-Chieh Hsu, Vicenc ¸ Rubies-Royo, Claire J Tom- lin, and Jaime F Fisac. Safety and liveness guaran- tees through reach-avoid reinforcement learning. arXiv preprint arXiv:2112.12288, 2021

    2021 arXiv

  10. [18]

    Isaacs: Iterative soft adversarial actor- critic for safety

    Kai-Chieh Hsu, Duy Phuong Nguyen, and Jaime Fer- nandez Fisac. Isaacs: Iterative soft adversarial actor- critic for safety. In Learning for Dynamics and Control Conference, pages 90–103. PMLR, 2023

    2023

  11. [19]

    On ellip- soidal techniques for reachability analysis

    Alexander B Kurzhanski and Pravin Varaiya. On ellip- soidal techniques for reachability analysis. part ii: Inter- nal approximations box-valued constraints. Optimization methods and software, 17(2):207–237, 2002. ISSN 1055- 6788

    2002

  12. [20]

    Certifiable deep learning for reachability using a new lipschitz continuous value function, 2025

    Jingqi Li, Donggun Lee, Jaewon Lee, Kris Shengjun Dong, Somayeh Sojoudi, and Claire Tomlin. Certifiable deep learning for reachability using a new lipschitz continuous value function, 2025. URL https://arxiv.org/ abs/2408.07866

    2025 arXiv

  13. [21]

    Physics-informed neural operator for learning partial differential equations, 2022

    Zongyi Li, Hongkai Zheng, Nikola Borislavov Kovachki, David Jin, Haoxuan Chen, Burigede Liu, Andrew Stu- art, Kamyar Azizzadenesheli, and Anima Anandkumar. Physics-informed neural operator for learning partial differential equations, 2022. URL https://openreview.net/ forum?id=...

    2022

  14. [22]

    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. In 6th Annual Learning for Dynamics & Control Conference, pages 719–731. PMLR, 2024

    2024

  15. [23]

    On reachability and minimum cost optimal control

    John Lygeros. On reachability and minimum cost optimal control. Automatica, 40(6):917–927, 2004

    2004

  16. [24]

    Lagrangian methods for approximating the viability kernel in high- dimensional systems

    John N Maidens, Shahab Kaynama, Ian M Mitchell, Meeko M K Oishi, and Guy A Dumont. Lagrangian methods for approximating the viability kernel in high- dimensional systems. Automatica, 49(7):2017–2029,

    2017

  17. [25]

    Control design along trajectories with sums of squares programming

    Anirudha Majumdar, Amir Ali Ahmadi, and Russ Tedrake. Control design along trajectories with sums of squares programming. In International Conference on Robotics and Automation (ICRA) , pages 4054–4061. IEEE, 2013. ISBN 1467356433

    2013

  18. [26]

    Margellos and J

    K. Margellos and J. Lygeros. Hamilton-Jacobi For- mulation for Reach–Avoid Differential Games. IEEE Transactions on Automatic Control , 56(8):1849–1861, 2011

    2011

  19. [27]

    A novel sequential method to train physics informed neural networks for allen cahn and cahn hilliard equations

    Revanth Mattey and Susanta Ghosh. A novel sequential method to train physics informed neural networks for allen cahn and cahn hilliard equations. Computer Meth- ods in Applied Mechanics and Engineering , 390:114474, 2022

    2022

  20. [28]

    Mitchell

    I. Mitchell. A toolbox of level set methods. http://www. cs. ubc. ca/mitchell/ToolboxLS/toolboxLS.pdf, 2004

    2004

  21. [29]

    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(7):947–957, 2005

    2005

  22. [30]

    Adaptive deep learning for high-dimensional hamilton– jacobi–bellman equations

    Tenavi Nakamura-Zimmerer, Qi Gong, and Wei Kang. Adaptive deep learning for high-dimensional hamilton– jacobi–bellman equations. SIAM Journal on Scientific Computing, 43(2):A1221–A1247, 2021

    2021

  23. [31]

    Raissi, P

    M. Raissi, P. Perdikaris, and G.E. Karniadakis. Physics- informed neural networks: A deep learning frame- work for solving forward and inverse problems involv- ing nonlinear partial differential equations. Journal of Computational Physics , 378:686–707, 2019. ISSN 0021-9991. d...

    2019 doi

  24. [33]

    Linear supervision for nonlinear, high- dimensional neural control and differential games, 2024

    William Sharpless, Zeyuan Feng, Somil Bansal, and Sylvia Herbert. Linear supervision for nonlinear, high- dimensional neural control and differential games, 2024. URL https://arxiv.org/abs/2412.02033

    2024 arXiv

  25. [34]

    Imposing exact safety specifications in neural reachable tubes

    Aditya Singh, Zeyuan Feng, and Somil Bansal. Imposing exact safety specifications in neural reachable tubes. arXiv preprint arXiv:2404.00814 , 2024. VII. A PPENDIX A. Quadrotor

    2024 arXiv

  26. [36]

    (px,py,pz) denotes the position and (vx,vy,vz) denotes the linear velocities

    Dynamics: The quadrotor dynamics is given as follows: ˙px =νx, ˙py =vy, ˙pz =vz, ˙qω =−(ωx·qx)/2− (ωy·qy)/2− (ωz·qz)/2, ˙qx = (ωx·qw)/2 + (ωz·qy)/2− (ωy·qz)/2, ˙qy = (ωy·qw)/2− (ωz·qx)/2 + (ωx·qz)/2, ˙qz = (ωz·qw)/2 + (ωy·qx)/2− (ωx·qy)/2, ˙vx =CT· (2·qw·qy + 2·qx·qz)F/m, ˙vy ...

  27. [37]

    Boundary condition: The safety function l(x) for the quadrotor case study is computed as: vn = q· e3· ¯q, dx = r2 ap2 xν2 z p2xν2x +p2xν2z + 2pxpyνxνy +p2yν2y +p2yν2z , dy = r2 ap2 yν2 z p2xν2x +p2xν2z + 2pxpyνxνy +p2yν2y +p2yν2z , l(x) =max( p x2 +y2− p dx +dy, 0)−ro, where q...

  28. [38]

    The kinematic-mode dynamics are: f =   v cos(θyaw) v sin(θyaw) ˙ϕ a v lr+lf tan(ϕ) a lr+lf tan(ϕ) + v (lr+lf ) cos2(ϕ) ˙ϕ 0  

    Kinematic Model Dynamics ( |v|< 0.5): For low veloc- ities (|v|< 0.5), the kinematic model is used. The kinematic-mode dynamics are: f =   v cos(θyaw) v sin(θyaw) ˙ϕ a v lr+lf tan(ϕ) a lr+lf tan(ϕ) + v (lr+lf ) cos2(ϕ) ˙ϕ 0  

  29. [39]

    Dynamic Model Dynamics ( |v| ≥ 0.5): For higher velocities (|v|≥ 0.5), the dynamic model is used. The dynamic-mode dynamics are: f =   v cos(θyaw +βslip) v sin(θyaw +βslip) ˙ϕ a ωyaw − µm vI (lr+lf ) l2 fCSf (glr−ah) +l2 rCSr(glf +ah) ωyaw + µm I(lr+lf ) (...

  30. [45]

    URL https://www.sciencedirect.com/science/article/ pii/S0021999118307125

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.