Pith. sign in

REVIEW 3 major objections 5 minor 50 references

Discriminative Barrier Functions for Safe Adversarial Imitation Learning from Observation

T0 review · 3 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read Restricting imitation-learning rewards to discrete-time control barrier functions yields barriers that generalize to unsafe states absent from expert data, making safe learning from observation possible.

desk verdict Worth reading for empirical collision reduction, but the paper overclaims by calling the learned h a barrier function when Eq. (2) is never enforced; treat it as a safety-shaped discriminator with promising experiments. read the letter →

arxiv 2607.13938 v1 pith:ZKKTG6WD submitted 2026-07-15 cs.RO

Anubhav Vishwakarma , Bhaumik Mehta , Caleb Hsu , Byron Boots , Karen Leung , Tyler Han This is my paper
classification cs.RO
keywords controlbarrierfunctionsinversereinforcementlearningfromobservationadversarialimitationsafeexplorationunlabeleddemonstrationsobstacleavoidancediscrete-timeCBFmodelpredictiveplanning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to show that a robot can learn a safety barrier purely from watching unlabeled expert trajectories, and that the barrier will reject unsafe states the expert never encountered. The idea is to force the reward search in imitation learning from observation to stay inside the class of discrete-time control barrier functions. The resulting object, called a Discriminative Barrier Function, is trained adversarially: expert states are labeled safe and the learner's own rollouts are labeled unsafe. The authors report that the recovered barrier transfers zero-shot to novel obstacles in simulation and on hardware, with zero collisions in the tested settings. If correct, this removes a major obstacle to safe real-world imitation learning: no action labels, no unsafe labels, and no hand-designed safety constraints are needed.

What carries the argument

The central object is the Discriminative Barrier Function (DBF), a scalar h(s) trained to satisfy h>0 on expert states and h<0 on learner states, while its transition cost q_h(s,s') = h(s') - h(s) + alpha(h(s)) serves as the adversary's score. The discrete-time CBF inequality requires, for each safe state, existence of an action keeping h from decreasing too fast; the training loss uses an adversarial objective with a gradient penalty for Lipschitz continuity, plus hinge sign constraints. This formulation carries the argument because it converts a safety certificate (a barrier) into a learned reward that a planner or policy can optimize, and it is what lets the same function both imitate and

What would settle it

Take the trained barrier h and, on a dense grid of states in the claimed safe set, enumerate every available action; if any state has h(s)>0 but h(f(s,a)) - h(s) < -alpha(h(s)) for every action a, forward invariance is violated and the barrier guarantee fails. A collision at an unseen obstacle under the same dynamics would also disprove the generalization claim.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the discriminator in adversarial imitation learning from observation can itself be a discrete-time control barrier function. Writing the barrier condition as a transition cost q_h(s,s') = h(s') - h(s) + alpha(h(s)) and restricting the IRL cost class to functions of this form turns the usual expert-vs-learner game into a search for a barrier that separates safe from unsafe state space. Because the barrier's zero contour is learned from expert data by construction, it can classify states the expert never visited—obstacles placed at test time in a maze and real-world obstacle layouts unseen during training—as unsafe. The paper reports that

Load-bearing premise

The load-bearing premise is that marking all expert states safe and all learner rollouts unsafe produces an h that actually satisfies the discrete-time CBF condition under the true dynamics, even though that condition is never checked during training.

Editorial extensions

If this is right

  • A robot can acquire a safety barrier from state-only demonstration data, eliminating the need for action labels, unsafe labels, or hand-designed constraints.
  • The learned barrier transfers zero-shot to obstacle configurations absent from training, so deployed robots can avoid hazards they never saw.
  • Online exploration during imitation becomes safer: the constraint-violation rate falls as learning proceeds instead of accumulating collisions.
  • Because the barrier is interpretable, the same object that drives control also provides a human-readable map of the safe region.
  • Both policy-based and planning-based learners benefit, and planning-based learners can use the barrier directly as a stage cost.

Reading between the lines

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

  • Editorial: the forward-invariance guarantee stands or falls on whether h satisfies the discrete-time CBF inequality at unvisited states; the training loss only enforces sign separation and a dynamic-constraint shape, so the practical claim is best read as a barrier-shaped reward unless a verification step is added.
  • Editorial: the expert-safe / learner-unsafe labeling is a strong prior; extending DBF to label some expert states unsafe or to add uncertainty-aware labels could reduce conservatism when demonstrations are narrow, a concern the paper's own limitations raise.
  • Editorial: because the K-class function modulates the safety/performance trade-off, learning it from data could let the barrier adapt its conservatism per region; this is a direct, testable extension.
  • Editorial: one could certify the learned h offline with a reachability or sampling-based verification over the action set; if violations appear, the learned h can be used as an initialization for a certified CBF synthesis rather than a terminal output.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes Discriminative Barrier Functions (DBF), an approach to Inverse Reinforcement Learning from Observation (IRLfO) in which the learned cost/discriminator is restricted to have the form q_h(s,s') = h(s') - h(s) + α(h(s)) for a scalar function h, with sign constraints h>0 on expert/safe states and h<0 on learner/unsafe states. The authors argue that this restriction makes the recovered h a discrete-time Control Barrier Function, and they combine this with a Wasserstein GAN objective, gradient penalty, and hinge loss to train the discriminator jointly with a policy or planner. Experiments in a simulated navigation task and on a real MUSHR robot report reduced collision rates and generalization to obstacle configurations unseen during training, especially for the planning-based MPAIL variant.

Significance. If the formal claim were established, this would be a significant step: learning a CBF with safety semantics directly from unlabeled state-only demonstrations, without action labels or explicit safe/unsafe labels, would address a real gap between IRLfO and safety-critical robotics. The paper also contains real-hardware validation, an interesting integration with model-predictive adversarial imitation learning, and an empirical demonstration of collision reduction on unseen obstacles. However, the central formal claim is not currently supported: the learned h is not shown to satisfy the defining CBF inequality, and the derivation connecting IRLfO to the implemented loss has nontrivial gaps. The contribution is therefore at present a safety-shaped adversarial imitation heuristic with promising empirical evidence rather than a method that provably recovers barrier functions.

major comments (3)
  1. [Section 3.3-3.4, Eq. (4)-(7), Algorithm 1] The central claim that DBF 'recovers barrier functions' is not supported by the objective. The discrete-time CBF condition Eq. (2), sup_a h(f(s,a))-h(s) ≥ -α(h(s)), never appears as a constraint, a loss term, or a post-training verification. The class C_CBF in Eq. (4) is defined only by sign constraints h(s)>0 on S and h(s)<0 on U; the objectives Eq. (5) and Eq. (7) and Algorithm 1 do not enforce Eq. (2). Maximizing E_πE[q_h]-E_π[q_h] encourages expert transitions to have larger q_h than learner transitions, but this does not imply that for every safe state there exists an action keeping h≥0. Gradient penalty and sign constraints only produce a Lipschitz classifier. To support the term 'barrier function' and the safety claims in the Abstract and Section 4.2, the authors must either add Eq. (2) as a constraint or penalty (using the known or learned dynamics) and verify it on evaluation st
  2. [Appendix A, Eq. (11) and Eq. (7)] The theoretical derivation in Appendix A is not rigorous enough to carry the paper's claims. Eq. (11) asserts an entropy-regularized state-transition-occupancy version of Proposition 3.2 of Ho and Ermon [34], but the original proposition is for state-action occupancy, and the GAIfO analogue is only cited for the non-entropy-regularized case. This is load-bearing because Eq. (16)-(17) rely on Eq. (11) to justify the minimax DBF objective. In addition, the implemented loss Eq. (7) includes a hinge loss with margin δ and weight λ_sign that is absent from the derived objective Eq. (5), which has hard sign constraints Eq. (6). The paper therefore does not actually derive the objective it optimizes. Please provide a proof of Eq. (11) or state it as an explicit assumption, and reconcile Eq. (5) with Eq. (7).
  3. [Section 3.5 and Section 4.3] The assumption that all learner rollouts are unsafe and all expert states are safe yields a classification boundary, not a control-invariant safe set. Section 4.3 states that 'DBF-based methods constrain the reward function search to satisfy the CBF dynamic constraint (Equation (5))'; this is inaccurate because Eq. (5) does not contain the CBF dynamic constraint Eq. (2). The reported collision rates are an indirect proxy and no experiment reports whether Eq. (2) holds on any state of the learned safe set. A verification experiment (e.g., the fraction of states in {h≥0} where Eq. (2) is satisfied under the simulator dynamics) is needed before calling h a CBF. Without it, the conclusion that DBFs 'recover coherent safety boundaries' is only an empirical claim about a trained classifier.
minor comments (5)
  1. [Section 3.3] The symbol S is used both for the state space and for the safe set (S := {s ∈ S | h(s) ≥ 0}). This ambiguity makes Eq. (4) and the surrounding text confusing; please use distinct symbols, e.g., X for the state space and C for the safe set.
  2. [Eq. (1) vs. Eq. (7)] Eq. (1) has entropy regularization weight λ in -λH(π), while Eq. (7) writes -H(π) with no λ. Please clarify whether λ is absorbed into the policy optimization or set to 1.
  3. [Algorithm 1 and Table 1] Algorithm 1 line 8 says r_θ(s,s') ← AIL(D_hθ(s,s')) but does not specify how each AIL variant maps the DBF discriminator to a reward. Table 1 lists GAIL, AIRL, and MPAIL reward structures, but the DBF variants' reward construction should be given explicitly for reproducibility.
  4. [Tables 4, 6, and 7] The loss column entry 'ReLU spec' is not defined; it presumably refers to a spectral-normalization variant or a ReLU-based Lipschitz penalty. Please define the term and clarify whether spectral normalization is applied in the DBF variants.
  5. [Table 2 and Figure 9] AIRL already achieves 0.0 collisions in Table 2, and Figure 9 shows that AIRL has lower cost rate and higher reward than DBF-AIRL. The claim of improved safety over 'standard IRL baselines' is therefore only supported for GAIL and MPAIL, and the AIRL comparison should be discussed honestly in the main text rather than relegated to a footnote.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the DBF objective is not a fit-equals-prediction scheme; the unenforced CBF condition is a correctness/support gap, not a circular step.

full rationale

The paper's derivation restricts the IRLfO cost class to q_h(s,s') = h(s') - h(s) + alpha(h(s)) with sign constraints h(s)>0 on expert/safe states and h(s)<0 on learner/unsafe states, then optimizes a WGAN objective with gradient penalty and hinge losses. The reported safety results are evaluated on obstacle configurations not used to fit the discriminator (e.g., the vertical wall in Figure 4d is unseen during training), so the collision numbers are not re-disclosures of fitted targets. The discrete-time CBF condition in Eq. (2), sup_a h(f(s,a)) - h(s) >= -alpha(h(s)), is stated but never enforced as a constraint or verified post hoc; this means the claim that the learned h is certified barrier function is under-supported, but that is a correctness/rigor concern about a missing condition, not a case where an output equals an input by construction. The paper's self-citation to MPAIL [6] supplies a planning component and comparison baseline; the DBF contribution is independent of MPAIL's correctness and is benchmarked against MPAIL as an external baseline. Appendix A even concedes that the exact parametrized network class is not guaranteed to be convex or closed, weakening the formal derivation but not creating circularity. No uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled in via self-citation. The main limitation flagged by the skeptics is that sign constraints alone define a classifier, not a forward-invariant set; this belongs under correctness risk, not circularity. Score 1 reflects only the presence of a minor, non-load-bearing self-citation to MPAIL.

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

The central proof relies on an unproved extension of the GAIL/IRL equivalence to state-transition occupancy measures, plus a series of domain assumptions about the safety of expert data and the convexity of the barrier-function class. The method also introduces several hand-tuned hyperparameters (λs, K-class function, hinge margin) that materially affect the reported safety outcomes.

free parameters (5)
  • λ_WGAN (WGAN loss weight) = 1.0
    Set in Appendix D; balances adversarial loss and safety regularization.
  • λ_GP (gradient penalty weight) = 10.0
    Set in Appendix D; enforces approximate 1-Lipschitz continuity.
  • λ_sign (hinge loss weight) = 5.0
    Set in Appendix D; enforces sign constraints h>0 on safe states and h<0 on unsafe states.
  • K-class function α = linear (κ, β unspecified)
    Hand-specified in Section 3.4 and Appendix D; Section 4.3 shows it affects cost rate and collisions.
  • hinge margin δ = not reported
    Used in Eq (7); δ+ and δ- are not defined or specified.
assumptions (5)
  • standard math Ho and Ermon's Proposition 3.2 (equivalence of IRL and occupancy matching for state-action MDPs with convex regularizer)
    Invoked in Appendix A to derive the DBF objective; underlies the whole formulation.
  • ad hoc to paper Entropy-regularized state-transition occupancy version of Proposition 3.2 (Eq 11)
    Stated without proof as the 'natural entropy-regularized extension of the GAIfO objective'; load-bearing for the derivation of Eq (17).
  • domain assumption Assumption (i) learner data is unsafe and (ii) expert data is safe (Section 3.5)
    Provides the positive/negative labels for the discriminator; if expert trajectories are not safe or learner rollouts contain safe behavior, the learned barrier will be mislabeled.
  • domain assumption The class CCBF is nonempty, closed, convex, and well-approximated by the neural parameterization (Appendix A Remark)
    Needed for the IRL regularizer to be proper/closed/convex; the actual neural network class is not convex or closed, so the proof only applies to an idealized ϵ-net approximation.
  • domain assumption Expert states are surrounded by ϵ-balls of safe states (min-distance to unsafe boundary; Section 3.5, after [23])
    Justifies the sign constraints and Lipschitz assumptions; if the expert grazes close to obstacles, the barrier may be ill-posed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Discriminative Barrier Functions for Safe Adversarial Imitation Learning from Observation." pith.science (2026). https://pith.science/paper/ZKKTG6WD

@misc{pith2026260713938,
  author       = {Pith},
  title        = {Pith review of: Discriminative Barrier Functions for Safe Adversarial Imitation Learning from Observation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZKKTG6WD}},
  note         = {Machine review of arXiv:2607.13938}
}
read the original abstract

Inverse Reinforcement Learning (IRL) algorithms are powerful tools for learning from and generalizing expert demonstrations, but they often rely on unconstrained exploration, rendering them unsafe for real-world deployment. Meanwhile, Control Barrier Functions (CBFs) can guarantee the safety of control systems, but the analytical design of CBFs can be time-consuming and esoteric. In this work, we address these limitations jointly by constraining reward function candidacy during IRL to the space of CBFs, yielding a formulation that exhibits safe online control with continuous experiential improvement. Crucially, this framework enables the data-driven recovery of barrier functions directly from unlabeled expert observations. We demonstrate that the recovered barrier function is robust to unsafe states entirely absent from the expert data. Furthermore, we benchmark our method against standard IRL baselines in a simulated navigation environment, demonstrating improved safety performance. Finally, we investigate the trade-offs of planning-based versus policy-based IRL methods across both simulation and a real world obstacle avoidance task.

Figures

Figures reproduced from arXiv: 2607.13938 by the authors.

Figure 1
Figure 1. Hardware deployment across two obstacle configurations unseen during training (a,b), demonstrating generalization to unseen unsafe observations. Robots are shown mid-inference nav￾igating toward the goal. (c,e) Closed-loop trajectories for DBF based planning method with expert reference (dotted). (d,f) bird’s-eye view map of the robot representing the barrier function range for the map region. See [PITH_FULL_IMAGE:… view at source ↗
Figure 2
Figure 2. Overview of DBF: a barrier func￾tion and imitation policy are recovered jointly from unlabeled expert observations within an AIL framework. The discriminator enforces CBF constraints, classifying expert states as safe and learner rollouts as unsafe, while the generator (planning or policy) optimizes against the learned barrier. These binary classification objectives appear practically adjacent: can a barrier functio… view at source ↗
Figure 3
Figure 3. Qualitative comparison of the learned barrier function h(s) for DBF-based methods and the reward function r(s) for non-DBF variants of GAIL, AIRL, and MPAIL. For DBF methods, the learned barrier function h(s) is plotted; for non-DBF baselines, the learned reward function r(s) is plotted. The black curve denotes the zero contour, and the red curve denotes the expert trajectory. Here s ∈ R 2 represents the agent’s pos… view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Qualitative analysis of barrier function generalization synthesized via DBF-MPAIL. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Normalized reward and cost rate dur￾ing online learning, comparing DBF and non￾DBF variants across all methods. Shaded regions represent standard deviation across training seeds. DBF-based methods consistently reduce cost rate while maintaining competitive reward [PIT…
Figure 6
Figure 6. Figure 6: Effect of K-class functions on cost rate and total collisions, evaluated on DBF-MPAIL. Linear K-class functions achieve the fastest reduc￾tion in cost rate, consistent with their steeper slope in the discriminator logit range. In our experiments, discriminator logit ou…
Figure 7
Figure 7. Figure 7: Hardware deployment across additional two obstacle configurations than [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Expert trajectories col￾lected in the maze environment by training PPO with varied wall con￾figurations and reward heuristics. To benchmark the AIL algorithms, we designed a maze en￾vironment and measured cumulative regret over the course of learning. Expert trajectori…
Figure 9
Figure 9. Figure 9: Performance comparison between AIRL and DBF-AIRL during online learning. AIRL achieves both higher reward and lower cost rate than DBF-AIRL, suggesting that AIRL’s in￾herently conservative reward structure (σ −1 ◦ D) already provides implicit safety regularization in t…
Figure 10
Figure 10. Figure 10: Extension of [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

50 extracted references · 15 linked inside Pith

  1. [34]

    Ho and S

    J. Ho and S. Ermon. Generative Adversarial Imitation Learning. InAd- vances in Neural Information Processing Systems, volume 29. Curran Associates, Inc.,

  2. [1]

    A. Y . Ng and S. J. Russell. Algorithms for inverse reinforcement learning. InProceedings of the Seventeenth International Conference on Machine Learning, ICML ’00, page 663–670, San Francisco, CA, USA, 2000. Morgan Kaufmann Publishers Inc. ISBN 1558607072

    2000

  3. [2]

    Abbeel and A

    P. Abbeel and A. Y . Ng. Apprenticeship learning via inverse reinforcement learning. In Proceedings of the Twenty-First International Conference on Machine Learning, ICML ’04, page 1, New York, NY , USA, 2004. Association for Computing Machinery. ISBN 1581138385. doi:10.1145/1015330.1015430. URLhttps://doi.org/10.1145/1015330.1015430

    arXiv 2004

  4. [3]

    B. D. Ziebart, A. Maas, J. A. Bagnell, and A. K. Dey. Maximum entropy inverse reinforcement learning. InProceedings of the 23rd National Conference on Artificial Intelligence - Volume 3, AAAI’08, page 1433–1438. AAAI Press, 2008. ISBN 9781577353683

    2008

  5. [4]

    Abbeel, A

    P. Abbeel, A. Coates, and A. Ng. Autonomous helicopter aerobatics through apprenticeship learning.I. J. Robotic Res., 29:1608–1639, 11 2010. doi:10.1177/0278364910371999

    doi:10.1177/0278364910371999 2010

  6. [5]

    Torabi, G

    F. Torabi, G. Warnell, and P. Stone. Generative Adversarial Imitation from Observation, June

  7. [6]

    T. Han, Y . Bao, B. Mehta, G. Guo, A. Vishwakarma, E. Kang, S. Jung, R. Scalise, J. Zhou, B. Xu, and B. Boots. Model Predictive Adversarial Imitation Learning for Planning from Observation, July 2025. URLhttp://arxiv.org/abs/2507.21533. arXiv:2507.21533 [cs]

    arXiv 2025

  8. [7]

    Garc ˜Aa and F

    J. Garc ˜Aa and F. Fern˜A¡ndez. A comprehensive survey on safe reinforcement learning.Journal of Machine Learning Research, 16(42):1437–1480, 2015. URLhttp://jmlr.org/papers/ v16/garcia15a.html

    2015

Show all 50 references
  1. [8]

    J. T. Nigg. On Inhibition/Disinhibition in Developmental Psychopathology: Views from Cog- nitive and Personality Psychology and a Working Inhibition Taxonomy.Psychological Bul- letin, 126(2):220–246, 2000

    2000

  2. [9]

    J. L. Ulloa, D. E. N ´u˜nez, P. A. Gaspar, and M. Brass. Imitative inhibitory control is associated with psychotic experiences in a sample from the general population.Frontiers in psychiatry, 15, 1470030, 126, 2024

    2024

  3. [10]

    Lederer, E

    A. Lederer, E. Noorani, J. S. Baras, and S. Hirche. Risk-sensitive inhibitory control for safe reinforcement learning. In2023 62nd IEEE Conference on Decision and Control (CDC), pages 1040–1045, 2023. doi:10.1109/CDC49753.2023.10383524

    2023

  4. [11]

    Torabi, G

    F. Torabi, G. Warnell, and P. Stone. Behavioral cloning from observation, 2018. URLhttps: //arxiv.org/abs/1805.01954

    2018 arXiv

  5. [12]

    A. D. Ames, S. Coogan, M. Egerstedt, G. Notomista, K. Sreenath, and P. Tabuada. Con- trol Barrier Functions: Theory and Applications. In2019 18th European Control Confer- ence (ECC), pages 3420–3431, June 2019. doi:10.23919/ECC.2019.8796030. URLhttps: //ieeexplore.ieee.org/abs...

    2019

  6. [13]

    A. D. Ames, J. W. Grizzle, and P. Tabuada. Control barrier function based quadratic programs with application to adaptive cruise control. In53rd IEEE Conference on Decision and Control, pages 6271–6278, 2014. doi:10.1109/CDC.2014.7040372. 11

    2014

  7. [14]

    Harms, M

    M. Harms, M. Kulkarni, N. Khedekar, M. Jacquet, and K. Alexis. Neural control barrier functions for safe navigation, 2024. URLhttps://arxiv.org/abs/2407.19907

    2024 arXiv

  8. [15]

    Q. Liao, Z. Li, A. Thirugnanam, J. Zeng, and K. Sreenath. Walking in narrow spaces: Safety- critical locomotion control for quadrupedal robots with duality-based optimization, 2023. URL https://arxiv.org/abs/2212.14199

    2023 arXiv

  9. [16]

    Akella, A

    P. Akella, A. Badithela, R. M. Murray, and A. D. Ames. Lipschitz continuity of signal temporal logic robustness measures: Synthesizing control barrier functions from one expert demonstra- tion, 2023. URLhttps://arxiv.org/abs/2304.03849

    2023 arXiv

  10. [17]

    K. Garg, J. Usevitch, J. Breeden, M. Black, D. Agrawal, H. Parwana, and D. Panagou. Ad- vances in the theory of control barrier functions: Addressing practical challenges in safe con- trol synthesis for autonomous and robotic systems, 2023. URLhttps://arxiv.org/abs/ 2312.16719

    2023 arXiv

  11. [18]

    Zhang, O

    S. Zhang, O. So, K. Garg, and C. Fan. Gcbf+: A neural graph control barrier function frame- work for distributed safe multiagent control.IEEE Transactions on Robotics, 41:1533–1552,

  12. [19]

    Agrawal and K

    A. Agrawal and K. Sreenath. Discrete control barrier functions for safety-critical control of dis- crete systems with application to bipedal robot navigation. InRobotics: Science and Systems, volume 13, pages 1–10. Cambridge, MA, USA, 2017

    2017

  13. [20]

    Guerrier, H

    M. Guerrier, H. Fouad, and G. Beltrame. Learning Control Barrier Functions and their ap- plication in Reinforcement Learning: A Survey, Apr. 2024. URLhttp://arxiv.org/abs/ 2404.16879. arXiv:2404.16879 [cs]

    2024 arXiv

  14. [21]

    O. So, Z. Serlin, M. Mann, J. Gonzales, K. Rutledge, N. Roy, and C. Fan. How to train your neural control barrier function: Learning safety filters for complex input-constrained systems,

  15. [22]

    Srinivasan, A

    M. Srinivasan, A. Dabholkar, S. Coogan, and P. A. Vela. Synthesis of Control Barrier Func- tions Using a Supervised Machine Learning Approach. In2020 IEEE/RSJ International Con- ference on Intelligent Robots and Systems (IROS), pages 7139–7145, Oct. 2020. doi:10.1109/ IROS4574...

    2020

  16. [23]

    Robey, H

    A. Robey, H. Hu, L. Lindemann, H. Zhang, D. V . Dimarogonas, S. Tu, and N. Matni. Learning Control Barrier Functions from Expert Demonstrations. In2020 59th IEEE Conference on Decision and Control (CDC), pages 3717–3724, Dec. 2020. doi:10.1109/CDC42340.2020. 9303785. URLhttps:...

    2020

  17. [24]

    Y . Yang, L. Chen, and M. Gombolay. Safe Inverse Reinforcement Learning via Control Barrier Function, Mar. 2023. URLhttp://arxiv.org/abs/2212.02753. arXiv:2212.02753 [cs]

    2023 arXiv

  18. [25]

    Y . Yang, L. Chen, Z. Zaidi, S. van Waveren, A. Krishna, and M. Gombolay. Enhancing Safety in Learning from Demonstration Algorithms via Control Barrier Function Shielding. In2024 19th ACM/IEEE International Conference on Human-Robot Interaction (HRI), pages 820– 829, Mar. 202...

    2024

  19. [26]

    Nakamura, A

    K. Nakamura, A. L. Bishop, S. Man, A. M. Johnson, Z. Manchester, and A. Bajcsy. How to train your latent control barrier function: Smooth safety filtering under hard-to-model con- straints. InLearning for Dynamics and Control, 2026. 12

    2026

  20. [27]

    Casta ˜neda, H

    F. Casta ˜neda, H. Nishimura, R. McAllister, K. Sreenath, and A. Gaidon. In-distribution barrier functions: Self-supervised policy filters that avoid out-of-distribution states, Jan. 2023. URL https://arxiv.org/abs/2301.12012. arXiv:2301.12012 [cs.RO]

    2023 arXiv

  21. [28]

    Tabbara and H

    I. Tabbara and H. Sibai. Learning neural control barrier functions from offline data with con- servatism, Sept. 2025. URLhttps://arxiv.org/abs/2505.00908v1. arXiv:2505.00908 [cs.LG]

    2025

  22. [29]

    Tayal, M

    M. Tayal, M. Tayal, A. Singh, S. Kolathaya, and R. Prakash. V-ocbf: Learning safety filters from offline data via value-guided offline control barrier functions, Apr. 2026. URLhttps: //arxiv.org/abs/2512.10822. arXiv:2512.10822 [cs.AI]

    2026

  23. [30]

    S. Gu, L. Yang, Y . Du, G. Chen, F. Walter, J. Wang, and A. Knoll. A Review of Safe Reinforce- ment Learning: Methods, Theories, and Applications.IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(12):11216–11235, Dec. 2024. ISSN 1939-3539. doi:10.1109/ TPAMI.2...

    2024

  24. [31]

    Jang, G.-H

    Y . Jang, G.-H. Kim, J. Lee, S. Sohn, B. Kim, H. Lee, and M. Lee. Safedice: Of- fline safe imitation learning with non-preferred demonstrations. In A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, editors,Advances in Neural In- formation Processing Systems,...

  25. [32]

    L. Yang, B. Werner, M. de Sa, and A. D. Ames. Cbf-rl: Safety filtering reinforcement learn- ing in training with control barrier functions, 2026. URLhttps://arxiv.org/abs/2510. 14959

    2026

  26. [33]

    Triest, M

    S. Triest, M. G. Castro, P. Maheshwari, M. Sivaprakasam, W. Wang, and S. Scherer. Learning risk-aware costmaps via inverse reinforcement learning for off-road navigation, 2023. URL https://arxiv.org/abs/2302.00134

    2023 arXiv

  27. [35]

    URLhttps://proceedings.neurips.cc/paper_files/paper/2023/file/ ed2fb79f2664c3d9ba878be7e575b2af-Paper-Conference.pdf

    2023

  28. [36]

    Gulrajani, F

    I. Gulrajani, F. Ahmed, M. Arjovsky, V . Dumoulin, and A. Courville. Improved training of wasserstein gans, 2017. URLhttps://arxiv.org/abs/1704.00028

    2017 arXiv

  29. [37]

    Orsini, A

    M. Orsini, A. Raichuk, L. Hussenot, D. Vincent, R. Dadashi, S. Girgin, M. Geist, O. Bachem, O. Pietquin, and M. Andrychowicz. What matters for adversarial imitation learning? In M. Ranzato, A. Beygelzimer, Y . Dauphin, P. Liang, and J. W. Vaughan, editors,Advances in Neural In...

    2021

  30. [38]

    J. Fu, K. Luo, and S. Levine. Learning Robust Rewards with Adverserial Inverse Reinforce- ment Learning. InInternational Conference on Learning Representations (ICLR), Feb. 2018. URLhttps://openreview.net/forum?id=rkHywl-A-

    2018

  31. [39]

    S. S. Srinivasa, P. Lancaster, J. Michalove, M. Schmittle, C. Summers, M. Rockett, J. R. Smith, S. Chouhury, C. Mavrogiannis, and F. Sadeghi. MuSHR: A low-cost, open-source robotic racecar for education and research.CoRR, abs/1908.08031, 2019. 13

    1908 arXiv

  32. [40]

    Arjovsky, S

    M. Arjovsky, S. Chintala, and L. Bottou. Wasserstein gan.ArXiv, abs/1701.07875, 2017. URL https://api.semanticscholar.org/CorpusID:13943041

    2017 arXiv

  33. [41]

    A. Ray, J. Achiam, and D. Amodei. Benchmarking safe exploration in deep reinforcement learning.arXiv preprint, 2019. URLhttps://cdn.openai.com/safexp-short.pdf

    2019

  34. [42]

    J. Sun, L. Yu, P. Dong, B. Lu, and B. Zhou. Adversarial Inverse Reinforcement Learning With Self-Attention Dynamics Model.IEEE Robotics and Automation Letters, 6(2):1880–1886, Apr. 2021. ISSN 2377-3766. doi:10.1109/LRA.2021.3061397. URLhttps://ieeexplore. ieee.org/document/9361118

    2021

  35. [43]

    J. F. Fisac, N. F. Lugovoy, V . Rubies-Royo, S. Ghosh, and C. J. Tomlin. Bridging hamilton- jacobi safety analysis and reinforcement learning. In2019 International Conference on Robotics and Automation (ICRA), pages 8550–8556, 2019. doi:10.1109/ICRA.2019.8794107

    2019

  36. [44]

    Mitchell, A

    I. Mitchell, A. Bayen, and C. Tomlin. A time-dependent hamilton-jacobi formulation of reach- able sets for continuous dynamic games.IEEE Transactions on Automatic Control, 50(7): 947–957, 2005. doi:10.1109/TAC.2005.851439

    2005

  37. [45]

    Mittal, P

    NVIDIA, :, M. Mittal, P. Roth, J. Tigue, A. Richard, O. Zhang, P. Du, A. Serrano-Mu ˜noz, X. Yao, R. Zurbr ¨ugg, N. Rudin, L. Wawrzyniak, M. Rakhsha, A. Denzler, E. Heiden, A. Borovicka, O. Ahmed, I. Akinola, A. Anwar, M. T. Carlson, J. Y . Feng, A. Garg, R. Gasoto, L. Gulich,...

    2025 arXiv

  38. [50]

    Sharma, N

    V . Sharma, N. Mehr, and N. Hovakimyan. Learning differentiable and safe multi-robot control for generalization to novel environments using control barrier functions, 2024. 14 A Appendix DBF as Cost-Regularized Observation-Only IRL.We now show how the DBF objective in Equation...

    2024

  39. [2016]

    URLhttps://proceedings.neurips.cc/paper_files/paper/2016/hash/ cc7e2b878868cbae992d1fb743995d8f-Abstract.html

    2016

  40. [2019]

    arXiv:1807.06158 [cs]

    URLhttp://arxiv.org/abs/1807.06158. arXiv:1807.06158 [cs]

    arXiv

  41. [2023]

    URLhttps://arxiv.org/abs/2310.15478

    arXiv

  42. [2025]

    doi:10.1109/tro.2025.3530348

    ISSN 1941-0468. doi:10.1109/tro.2025.3530348. URLhttp://dx.doi.org/10. 1109/TRO.2025.3530348

    1941

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.