Learning Ensembles of Vision-based Safety Control Filters

Hussein Sibai; Ihab Tabbara

REVIEW 4 major objections 5 minor 46 references

Learning Ensembles of Vision-based Safety Control Filters

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Diverse safety-filter ensembles beat larger single models

desk verdict Competent empirical study of ensembles for vision-based safety filters, with a load-bearing proxy-label caveat that keeps it from being a safety claim. read the letter →

arxiv 2412.02029 v1 pith:ZEDPC4DD submitted 2024-12-02 cs.LG cs.AIcs.CVcs.ROcs.SYeess.SY

Ihab Tabbara , Hussein Sibai This is my paper · ORCID

classification cs.LGcs.AIcs.CVcs.ROcs.SYeess.SY

keywords safetyfiltersensemblescontrolbarrierfunctionspre-trainedvisionmodelsout-of-distributiongeneralizationautonomousdrivingDeepAccidentmajorityvoting

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 asks whether combining several learned safety filters into an ensemble makes them more reliable than any single filter. It reports that diverse ensembles, built from different pre-trained vision backbones and training methods, classify safe versus unsafe driving states and controls more accurately than their members and than larger individual models with similar parameter budgets. The gains also persist, at a reduced level, on held-out driving towns. The authors present these results as an empirical step toward more reliable vision-based safety filters while formal verification remains out of reach.

What carries the argument

The central object is the diverse ensemble of learned safety filters. Each member filter maps image representations from a frozen pre-trained vision backbone to either a control barrier function value or a discriminating hyperplane that separates safe from unsafe controls; members differ in backbone (CLIP vs VC1), training method (iDBF, SABLAS, DH), and weight initialization. The ensemble combines member outputs by majority voting, weighted averaging, or a consensus rule; an ensemble improvement rate (EIR) measures loss reduction relative to the average member. The diversity across backbones and training methods is what carries the performance gain.

What would settle it

Take the best majority-voting ensemble and deploy it as an online filter in a closed-loop driving simulator benchmark, counting collisions when the filter replaces unsafe controls versus when the nominal controller runs unfiltered. If filtered runs do not reduce collisions on the out-of-distribution towns, the reported accuracy gains do not translate into safety.

Watch

Extended reading notes

Core claim

The authors claim that for vision-based safety filters in autonomous driving, a diverse ensemble—mixing frozen pre-trained vision backbones (CLIP and VC1), multiple training methods (iDBF, SABLAS, and DH), and an input-dependent aggregation rule such as majority voting—achieves better state and control classification accuracy than individual member models and outperforms single large models with comparable total parameters. The best reported configuration, a majority-voting ensemble over all training methods and both backbones, reaches 76.99% safe-action and 80.98% unsafe-action accuracy on DeepAccident, improving on member averages. The ensemble improvement rate stays positive for out-of-distribution towns, indicating a generalization benefit.

Load-bearing premise

The load-bearing premise is that safe and unsafe classification accuracy on the dataset labels predicts real safety-filter effectiveness; the labels mark the five frames before a collision as safe and their controls as unsafe, and the paper does not run closed-loop trajectories to confirm that filtering these controls would prevent collisions.

Editorial extensions

If this is right

Ensembles of vision-based safety filters can serve as a practical reliability upgrade when formal verification of the filters is intractable.
Combining complementary perception backbones gives the largest accuracy gain, so filter design should prioritize feature diversity over model width or depth.
Majority voting, an input-dependent aggregation, outperforms uniform averaging; weighted averaging offers a consistent but slightly weaker alternative.
The safety advantage transfers to out-of-distribution towns that share accident patterns, though the transfer shrinks relative to in-distribution data.
Consensus-based ensembles can cut the number of calls to a strong aggregator by roughly half with little accuracy loss, reducing compute cost.

Reading between the lines

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

The paper measures safety by classification accuracy on heuristic labels; its result does not establish that filtering with these ensembles prevents collisions in a closed loop. A natural extension is to test whether the accuracy gains translate into lower collision rates when the filter actually replaces unsafe controls.
The diversity advantage suggests that the mechanism is complementary representations; an ablation that controls for total parameter count while varying feature similarity could isolate that mechanism.
The positive OOD results are limited by the fact that the withheld towns share accident patterns with training towns; true out-of-distribution generalization remains untested.
Weighted averaging freezes weights at deployment; an input-dependent weighting scheme, such as weighting by each member's confidence, could plausibly beat majority voting and is a direct follow-up.

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.

Desk Editor's Note

read the letter

Quick take: this is a competent empirical study of ensembles for vision-based safety filters, and the main result—diverse ensembles beat their members and larger single models on classification accuracy—is credible for what it measures. The caveat sits on the word 'safety': the evaluation is classification accuracy on proxy labels, not closed-loop collision avoidance.

What's actually new: this is the first systematic look at ensembles of safety filters, combining two PVR backbones (CLIP, VC1), three training methods (iDBF, SABLAS, DH), and four aggregation schemes (uniform/weighted averaging, majority voting, consensus). The paper includes a fair control: large single models with comparable total parameter counts, and shows ensembles mostly win without relying on capacity. The consensus-based method is a nice practical touch—it calls the expensive ensemble only when two cheap members disagree, with small accuracy loss. That's a useful design pattern for real-time deployment.

Soft spots, in order: (1) The label proxy is load-bearing. Section 5.1 labels the five frames before a collision as safe states, and only the controls during that interval as unsafe. So the 'safe state' accuracy mostly rewards doing nothing, and there is no closed-loop evaluation showing that the ensemble prevents collisions. The authors are honest that OOD towns share similar accident patterns, so generalization claims are limited. (2) Ensemble rows in Table 1 are point estimates with no error bars or significance tests, while the member models show large variance. I'd want to know whether the ensemble improvements are stable across seeds. (3) The consensus 'expert' members are described as highly accurate, but the paper doesn't say whether their selection was based on held-out data or on the test set. That needs clarifying.

The math and citations look fine. The CBF background is standard, and the EIR metric is properly adapted from Theisen et al. The self-citation to Yang and Sibai (2024) is legitimate since the labels and backbone setup come from there.

Who this is for: people building vision-based safety filters, especially with learned CBFs, will get a practical recipe. It is not a formal guarantee paper, and the authors don't claim it is.

My recommendation: send it to review. The empirical claim is plausible, the study is well-scoped, and the flaws are fixable—add error bars, clarify the expert-selection protocol, and explicitly discuss the proxy-label limitation. I'd accept it conditionally after those changes.

Referee Report

4 major / 5 minor

Summary. The paper proposes using ensembles of vision-based safety control filters trained with pre-trained vision backbones (CLIP and VC1) and three filter-learning methods (iDBF, SABLAS, DH). It compares four aggregation schemes (majority voting, uniform averaging, weighted averaging, and consensus-based selection) against individual member models and larger single models, using safe/unsafe state and action classification accuracy and the ensemble improvement rate (EIR) on DeepAccident. The main empirical claim is that diverse ensembles, particularly those combining multiple backbones and training methods, improve classification accuracy and out-of-distribution EIR relative to individual models, while no formal safety guarantees are claimed.

Significance. If established, the paper would offer practical guidance for building more reliable learned safety filters by combining diverse perception backbones and training methods. The work has clear strengths: a broad and systematic configuration sweep, a comparison against larger single models with comparable parameter counts, use of the EIR metric, and an explicit (if mild) out-of-distribution split. The contribution is entirely empirical and makes no formal guarantees, which is appropriate given the stated difficulty of verifying vision-based filters. However, the significance of the result depends on whether classification accuracy on the proxy labels actually translates into collision-avoidance behavior, and that connection is not demonstrated.

major comments (4)

[Section 5.1, Table 1] No uncertainty estimates or significance tests are reported for any ensemble row in Table 1; every ensemble entry is a point estimate, whereas member-model rows report mean and standard deviation over five models. For example, the iDBF VC1 member models achieve 75.72±7.07 safe and 71.14±6.63 unsafe action accuracy, while the majority-voting ensemble achieves 77.06 and 73.10; these differences are well within one standard deviation of the member distribution. The paper should provide error bars on ensemble results, for example by bootstrap resampling or by training multiple ensembles from different member subsets, before concluding that ensembles are better than individual models.
[Section 5.1, Section 5.2] The safety-filter claim rests entirely on proxy labels. Section 5.1 defines the five frames before a collision as safe states and the controls during that interval as unsafe, so a filter that never intervenes would score perfectly on safe states and safe actions. No closed-loop evaluation, filtered-control trajectories, collision rates, or comparisons of the filtered closed-loop behavior are reported. Consequently, the conclusion that ensembles improve the reliability of safety filters is not established; the evidence supports only an improvement in classification accuracy on the specific labels defined in Section 5.1. The authors should either add a closed-loop safety evaluation or explicitly re-scope the conclusion to classification accuracy rather than safety-filter effectiveness.
[Section 5.2, consensus-based ensembles] The expert-selection protocol for consensus-based ensembles appears to rely on test-set performance. M1 and M2 are chosen as "highly accurate" members from the same pool used to build M3, and M3 is the best majority-voting ensemble from Table 1, with accuracy (76.99, 80.90). Moreover, M3 is an ensemble that includes M1 and M2 themselves, so the consensus method's tie-breaker is not independent of the two models it arbitrates. If M1, M2, and M3 were selected after inspecting held-out test performance, the reported consensus accuracies are optimistically biased. The paper should specify a validation-based selection procedure or explicitly present the result as post-hoc analysis.
[Section 5.2, Table 2] The out-of-distribution claim is weaker than the conclusion suggests. As the authors state, the held-out towns "share similar accident patterns and trajectories," making this only a mild distribution shift. The positive OOD EIR values in Table 2 therefore provide limited evidence for improved out-of-distribution generalization. The paper should either test on more distinct domains, such as different accident types or environmental conditions, or soften the generalization claim in the abstract and conclusion to match the limited OOD scope.

minor comments (5)

[Throughout] There are numerous typos and spacing artifacts, e.g., "Lispchitz" in Definition 3.2 and "Y ang" in the references; a careful proofreading pass is needed.
[Section 4.2] The notation in the loss definition, such as "· /BD (x′ ∈ Xsafe)", is not defined; it appears to be an indicator function, but this should be stated explicitly to make the loss readable.
[Table 1 and Section 5.2] The caption describes large single models as having "comparable total parameters" to the ensembles, but the text says the deeper models have about eight times and the wider models about ten times the parameters of a member model. Since ensemble size varies between five and fifteen members, the parameter-count comparison should be made explicit for each ensemble size.
[Section 5.1] Training and evaluation details are incomplete: the paper should report the number of epochs, learning rate, optimizer, batch size, train/validation/test split proportions, and the procedure for selecting hyperparameters, since these are needed to reproduce the experiments.
[Table 2] Only EIR values are reported for the OOD split, not absolute accuracies. Reporting the actual safe/unsafe accuracies for IND and OOD would help readers assess whether the positive EIR corresponds to useful performance or merely to a smaller loss when all models perform poorly.

Circularity Check

1 steps flagged · score 4.0 of 10

Partial by-construction EIR result for averaging ensembles; central accuracy claims remain independently measured.

self definitional [Section 4.2 (Weighted averaging-based ensembles) and Section 5.1 (Evaluation metrics); used in Table 2 and Section 5.2 OOD analysis.]
"EIR := (1/N Σ_{i∈[N]} L(f_i) − L(f̄)) / (1/N Σ_{i∈[N]} L(f_i)) ... We define the loss as: L = Lsafe + λ Lunsafe, where Lsafe = σ (−Ḃ_ens(x,u) − γ(B_ens(x))) · ... and Lunsafe = σ (Ḃ_ens(x,u) + γ(B_ens(x))) · ..., σ is the ReLU function ... For this reason, we only consider linear γ in our experiments."

For averaging-based ensembles with linear γ, the ensemble action/state classifier is a convex combination of member classifiers: f̄ = Σ w_i f_i. The per-sample loss is a nonnegative sum of ReLU terms, hence convex in f. Jensen's inequality therefore forces L(f̄) ≤ Σ w_i L(f_i), so EIR ≥ 0 for any dataset, including OOD data, regardless of model quality. The paper's statement that 'EIR remains consistently positive for OOD test data' is thus true by construction for the averaging rows of Table 2; the OOD 'advantage over member models' for averaging ensembles reduces to a convexity identity rather than an empirical finding. Accuracy results and majority-voting EIR are not forced, so the circularity is partial.

full rationale

The central accuracy comparisons are empirical and non-circular: Table 1 accuracies are measured on held-out test data, and the ensemble-versus-individual and ensemble-versus-large-model claims are not defined by the fitted weights or the loss weight λ. The self-citations to Yang and Sibai (2024) supply the member-model construction and safety labels, but they do not force the ensemble result; the ensemble comparison is a new experiment. The paper's own admission that the OOD towns 'share similar accident patterns and trajectories' and the absence of closed-loop collision metrics are validity concerns about the proxy safety labels, not circularity. The one by-construction element is EIR for averaging-based ensembles: with linear γ, the ensemble classifier is a convex combination of member classifiers and the loss is convex in the classifier output, so Jensen's inequality guarantees positive EIR on any test set. Thus the positive OOD EIR for averaging ensembles is a mathematical identity rather than empirical evidence, partially weakening the OOD-generalization claim built on EIR while leaving the accuracy claims intact. Overall, this is partial, non-central circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on dataset-label fidelity and on the proxy that classification accuracy reflects safety-filter reliability. The linear-gamma restriction is a modeling choice that narrows the class of CBFs but does not by itself break the empirical comparison. No new physical or algorithmic entities are introduced.

free parameters (3)

Ensemble weights w_i = Optimized on training data (values not reported)
Weighted averaging ensembles learn input-independent weights to minimize a safety loss (Section 4.2). Reported accuracies depend on these fitted weights.
Loss weight lambda = 18
Chosen to penalize unsafe misclassifications and handle dataset imbalance (Section 5.1). No sensitivity analysis is provided.
Filter head architecture sizes = Member: 2 layers, 64 neurons; deep: 5 layers, 95 neurons; wide: 2 layers, 220 neurons
Capacity choices affect overfitting; the paper notes larger models overfit (Section 5.2).

assumptions (4)

domain assumption The dynamics are control-affine with locally Lipschitz f and g
Section 3 relies on this to formulate the safety-filter QP. Actual vehicle dynamics and perception are unknown and not verified.
ad hoc to paper The class K-infinity function gamma is linear
Section 4.2 restricts gamma to be linear so that DH and CBF-based members can be combined in the same ensemble. This limits the class of CBFs considered.
domain assumption DeepAccident safety labels from Yang and Sibai (2024) are correct
Labels define the five frames before a collision as safe and controls during that interval as unsafe (Section 5.1). No re-validation is performed.
domain assumption High classification accuracy is a meaningful proxy for safety-filter performance
The paper measures accuracy on a static dataset, not closed-loop safety. The conclusion that ensembles are 'more reliable' rests on this proxy.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Learning Ensembles of Vision-based Safety Control Filters." pith.science (2026). https://pith.science/paper/ZEDPC4DD

@misc{pith2026241202029,
  author       = {Pith},
  title        = {Pith review of: Learning Ensembles of Vision-based Safety Control Filters},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZEDPC4DD}},
  note         = {Machine review of arXiv:2412.02029}
}

read the original abstract

Safety filters in control systems correct nominal controls that violate safety constraints. Designing such filters as functions of visual observations in uncertain and complex environments is challenging. Several deep learning-based approaches to tackle this challenge have been proposed recently. However, formally verifying that the learned filters satisfy critical properties that enable them to guarantee the safety of the system is currently beyond reach. Instead, in this work, motivated by the success of ensemble methods in reinforcement learning, we empirically investigate the efficacy of ensembles in enhancing the accuracy and the out-of-distribution generalization of such filters, as a step towards more reliable ones. We experiment with diverse pre-trained vision representation models as filter backbones, training approaches, and output aggregation techniques. We compare the performance of ensembles with different configurations against each other, their individual member models, and large single-model baselines in distinguishing between safe and unsafe states and controls in the DeepAccident dataset. Our results show that diverse ensembles have better state and control classification accuracies compared to individual models.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 27 canonical work pages

[1]

Safe control using vision-based control barrier function (v-cbf)

Hossein Abdi, Golnaz Raja, and Reza Ghabcheloo. Safe control using vision-based control barrier function (v-cbf). In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 782--788. IEEE, 2023

work page 2023
[2]

Introduction to neural network verification

Aws Albarghouthi. Introduction to neural network verification. CoRR, abs/2109.10317, 2021. URL https://arxiv.org/abs/2109.10317

arXiv 2021
[3]

Control barrier function based quadratic programs for safety critical systems

Aaron D Ames, Xiangru Xu, Jessy W Grizzle, and Paulo Tabuada. Control barrier function based quadratic programs for safety critical systems. IEEE Transactions on Automatic Control, 62 0 (8): 0 3861--3876, 2016

2016
[4]

Ames, Samuel Coogan, Magnus Egerstedt, Gennaro Notomista, Koushil Sreenath, and Paulo Tabuada

Aaron D. Ames, Samuel Coogan, Magnus Egerstedt, Gennaro Notomista, Koushil Sreenath, and Paulo Tabuada. Control barrier functions: Theory and applications. In 2019 18th European Control Conference (ECC), pages 3420--3431, 2019. doi:10.23919/ECC.2019.8796030

arXiv 2019
[5]

Formally verified neural network control barrier certificates for unknown systems

Mahathi Anand and Majid Zamani. Formally verified neural network control barrier certificates for unknown systems. IFAC-PapersOnLine, 56 0 (2): 0 2431--2436, 2023. ISSN 2405-8963. doi:https://doi.org/10.1016/j.ifacol.2023.10.1219. URL https://www.sciencedirect.com/science/article/pii/S2405896323016233. 22nd IFAC World Congress

work page doi:10.1016/j.ifacol.2023.10.1219 2023
[6]

Averaged-dqn: Variance reduction and stabilization for deep reinforcement learning

Oron Anschel, Nir Baram, and Nahum Shimkin. Averaged-dqn: Variance reduction and stabilization for deep reinforcement learning. In International conference on machine learning, pages 176--185. PMLR, 2017

2017
[7]

Autonomous driving—a crash explained in detail

Johannes Betz, Alexander Heilmeier, Alexander Wischnewski, Tim Stahl, and Markus Lienkamp. Autonomous driving—a crash explained in detail. Applied Sciences, 9 0 (23), 2019. ISSN 2076-3417. doi:10.3390/app9235126. URL https://www.mdpi.com/2076-3417/9/23/5126

work page doi:10.3390/app9235126 2019
[8]

Guaranteed safe spacecraft docking with control barrier functions

Joseph Breeden and Dimitra Panagou. Guaranteed safe spacecraft docking with control barrier functions. IEEE Control Systems Letters, 6: 0 2000--2005, 2022. doi:10.1109/LCSYS.2021.3136813

arXiv

Show all 46 references

[9]

Scalable surrogate verification of image-based neural network control systems using composition and unrolling

Feiyang Cai, Chuchu Fan, and Stanley Bak. Scalable surrogate verification of image-based neural network control systems using composition and unrolling. arXiv preprint arXiv:2405.18554, 2024

2024 arXiv
[10]

In-distribution barrier functions: Self-supervised policy filters that avoid out-of-distribution states

Fernando Castaneda, Haruki Nishimura, Rowan Thomas McAllister, Koushil Sreenath, and Adrien Gaidon. In-distribution barrier functions: Self-supervised policy filters that avoid out-of-distribution states. In Learning for Dynamics and Control Conference, pages 286--299. PMLR, 2023

2023
[11]

Randomized ensembled double q-learning: Learning fast without a model

Xinyue Chen, Che Wang, Zijian Zhou, and Keith Ross. Randomized ensembled double q-learning: Learning fast without a model. arXiv preprint arXiv:2101.05982, 2021

2021 arXiv
[12]

Verification and synthesis of control barrier functions

Andrew Clark. Verification and synthesis of control barrier functions. In 2021 60th IEEE Conference on Decision and Control (CDC), pages 6105--6112. IEEE, 2021

2021
[13]

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 0 (2): 0 1904--1911, 2022 a

1904
[14]

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 Conference on Robot Learning, pages 1724--1735. PMLR, 2022 b

2022
[15]

Safe control with learned certificates: A survey of neural lyapunov, barrier, and contraction methods for robotics and control

Charles Dawson, Sicun Gao, and Chuchu Fan. Safe control with learned certificates: A survey of neural lyapunov, barrier, and contraction methods for robotics and control. IEEE Transactions on Robotics, 39 0 (3): 0 1749--1767, 2023. doi:10.1109/TRO.2022.3232542

2023
[16]

Carla: An open urban driving simulator

Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. Carla: An open urban driving simulator. In Conference on robot learning, pages 1--16. PMLR, 2017

2017
[17]

Sentinel: taming uncertainty with ensemble based distributional reinforcement learning

Hannes Eriksson, Debabrota Basu, Mina Alibeigi, and Christos Dimitrakakis. Sentinel: taming uncertainty with ensemble based distributional reinforcement learning. In Uncertainty in Artificial Intelligence, pages 631--640. PMLR, 2022

2022
[18]

Neural network ensembles in reinforcement learning

Stefan Fau er and Friedhelm Schwenker. Neural network ensembles in reinforcement learning. Neural Processing Letters, 41: 0 55--69, 2015

2015
[19]

Why so pessimistic? estimating uncertainties for offline rl through ensembles, and why their independence matters

Kamyar Ghasemipour, Shixiang Shane Gu, and Ofir Nachum. Why so pessimistic? estimating uncertainties for offline rl through ensembles, and why their independence matters. Advances in Neural Information Processing Systems, 35: 0 18267--18281, 2022

2022
[20]

Autonomy for surgical robots: Concepts and paradigms

Tamás Haidegger. Autonomy for surgical robots: Concepts and paradigms. IEEE Transactions on Medical Robotics and Bionics, 1 0 (2): 0 65--76, 2019. doi:10.1109/TMRB.2019.2913282

2019
[21]

Ensemble quantile networks: Uncertainty-aware reinforcement learning with applications in autonomous driving

Carl-Johan Hoel, Krister Wolff, and Leo Laine. Ensemble quantile networks: Uncertainty-aware reinforcement learning with applications in autonomous driving. IEEE Transactions on Intelligent Transportation Systems, 24 0 (6): 0 6030--6041, 2023

2023
[22]

Verifying controllers with vision-based perception using safe approximate abstractions

Chiao Hsieh, Yangge Li, Dawei Sun, Keyur Joshi, Sasa Misailovic, and Sayan Mitra. Verifying controllers with vision-based perception using safe approximate abstractions. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 41 0 (11): 0 4205--4216, 2022

2022
[23]

Verification of neural control barrier functions with symbolic derivative bounds propagation

Hanjiang Hu, Yujie Yang, Tianhao Wei, and Changliu Liu. Verification of neural control barrier functions with symbolic derivative bounds propagation. In 8th Annual Conference on Robot Learning, 2024. URL https://openreview.net/forum?id=jnubz7wB2w

2024
[24]

An expert ensemble for detecting anomalous scenes, interactions, and behaviors in autonomous driving

Tianchen Ji, Neeloy Chakraborty, Andre Schreiber, and Katherine Driggs-Campbell. An expert ensemble for detecting anomalous scenes, interactions, and behaviors in autonomous driving. The International Journal of Robotics Research, page 02783649241297998, 2024

2024
[25]

Verification of image-based neural network controllers using generative models

Sydney M Katz, Anthony L Corso, Christopher A Strong, and Mykel J Kochenderfer. Verification of image-based neural network controllers using generative models. Journal of Aerospace Information Systems, 19 0 (9): 0 574--584, 2022

2022
[26]

Safety filters for black-box dynamical systems by learning discriminating hyperplanes

Will Lavanakul, Jason Choi, Koushil Sreenath, and Claire Tomlin. Safety filters for black-box dynamical systems by learning discriminating hyperplanes. In 6th Annual Learning for Dynamics & Control Conference, pages 1278--1291. PMLR, 2024

2024
[27]

An ensemble method for inverse reinforcement learning

Jin-Ling Lin, Kao-Shing Hwang, Haobin Shi, and Wei Pan. An ensemble method for inverse reinforcement learning. Information Sciences, 512: 0 518--532, 2020

2020
[28]

Where are we in the search for an artificial visual cortex for embodied intelligence? Advances in Neural Information Processing Systems, 36: 0 655--677, 2023

Arjun Majumdar, Karmesh Yadav, Sergio Arnaud, Jason Ma, Claire Chen, Sneha Silwal, Aryan Jain, Vincent-Pierre Berges, Tingfan Wu, Jay Vakil, et al. Where are we in the search for an artificial visual cortex for embodied intelligence? Advances in Neural Information Processing S...

2023
[29]

Castnet: A context-aware, spatio-temporal dynamic motion prediction ensemble for autonomous driving

Trier Mortlock, Arnav Malawade, Kohei Tsujio, and Mohammad Al Faruque. Castnet: A context-aware, spatio-temporal dynamic motion prediction ensemble for autonomous driving. ACM Transactions on Cyber-Physical Systems, 8 0 (2): 0 1--20, 2024

2024
[30]

Towards tractable optimism in model-based reinforcement learning

Aldo Pacchiano, Philip Ball, Jack Parker-Holder, Krzysztof Choromanski, and Stephen Roberts. Towards tractable optimism in model-based reinforcement learning. In Uncertainty in Artificial Intelligence, pages 1413--1423. PMLR, 2021

2021
[31]

Dey, and Marco Molinaro

Alberto Del Pia, Santanu S. Dey, and Marco Molinaro. Mixed-integer quadratic programming is in NP . Mathematical Programming, 162 0 (1): 0 225--240, March 2017. ISSN 1436-4646. doi:10.1007/s10107-016-1036-0. URL https://doi.org/10.1007/s10107-016-1036-0

2017 doi
[32]

Sablas: Learning safe control barrier functions for systems with unknown dynamics

Yuxiao Qin, Nikolai Mote, Haruki Nishimura, and Aaron D Ames. Sablas: Learning safe control barrier functions for systems with unknown dynamics. IEEE Robotics and Automation Letters, 7 0 (3): 0 7357--7364, 2022

2022
[33]

Learning transferable visual models from natural language supervision

Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pa...

2021
[34]

Uncertainty quantification and deep ensembles

Rahul Rahaman et al. Uncertainty quantification and deep ensembles. Advances in neural information processing systems, 34: 0 20063--20075, 2021

2021
[35]

Ensemble learning: A survey

Omer Sagi and Lior Rokach. Ensemble learning: A survey. Wiley interdisciplinary reviews: data mining and knowledge discovery, 8 0 (4): 0 e1249, 2018

2018
[36]

Nnlander-verif: A neural network formal verification framework for vision-based autonomous aircraft landing

Ulices Santa Cruz and Yasser Shoukry. Nnlander-verif: A neural network formal verification framework for vision-based autonomous aircraft landing. In NASA Formal Methods Symposium, pages 213--230. Springer, 2022

2022
[37]

Neural network verification with branch-and-bound for general nonlinearities

Zhouxing Shi, Qirui Jin, Zico Kolter, Suman Jana, Cho-Jui Hsieh, and Huan Zhang. Neural network verification with branch-and-bound for general nonlinearities. arXiv preprint arXiv:2405.21063, 2024

2024 arXiv
[38]

Daniel C. H. Tan, Fernando Acero, Robert McCarthy, Dimitrios Kanoulas, and Zhibin Li. Value functions are control barrier functions: Verification of safe policies using control theory, 2023. URL https://arxiv.org/abs/2306.04026

2023 arXiv
[39]

Learning a formally verified control barrier function in stochastic environment, 2024

Manan Tayal, Hongchao Zhang, Pushpak Jagtap, Andrew Clark, and Shishir Kolathaya. Learning a formally verified control barrier function in stochastic environment, 2024. URL https://arxiv.org/abs/2403.19332

2024 arXiv
[40]

When are ensembles really effective? Advances in Neural Information Processing Systems, 36, 2024

Ryan Theisen, Hyunsuk Kim, Yaoqing Yang, Liam Hodgkinson, and Michael W Mahoney. When are ensembles really effective? Advances in Neural Information Processing Systems, 36, 2024

2024
[41]

Enforcing safety for vision-based controllers via control barrier functions and neural radiance fields

Mukun Tong, Charles Dawson, and Chuchu Fan. Enforcing safety for vision-based controllers via control barrier functions and neural radiance fields. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 10511--10517. IEEE, 2023

2023
[42]

Deepaccident: A motion and accident prediction benchmark for v2x autonomous driving

Tianqi Wang, Sukmin Kim, Ji Wenxuan, Enze Xie, Chongjian Ge, Junsong Chen, Zhenguo Li, and Ping Luo. Deepaccident: A motion and accident prediction benchmark for v2x autonomous driving. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 5599--5606, 2024

2024
[43]

Neural lyapunov control for discrete-time systems

Junlin Wu, Andrew Clark, Yiannis Kantaros, and Yevgeniy Vorobeychik. Neural lyapunov control for discrete-time systems. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. URL https://openreview.net/forum?id=ArRycLMoUg

2023
[44]

Differentiable control barrier functions for vision-based end-to-end autonomous driving

Wei Xiao, Tsun-Hsuan Wang, Makram Chahine, Alexander Amini, Ramin Hasani, and Daniela Rus. Differentiable control barrier functions for vision-based end-to-end autonomous driving. arXiv preprint arXiv:2203.02401, 2022

2022 arXiv
[45]

Barriernet: Differentiable control barrier functions for learning of safe robot control

Wei Xiao, Tsun-Hsuan Wang, Ramin Hasani, Makram Chahine, Alexander Amini, Xiao Li, and Daniela Rus. Barriernet: Differentiable control barrier functions for learning of safe robot control. IEEE Transactions on Robotics, 39 0 (3): 0 2289--2307, 2023. doi:10.1109/TRO.2023.3249564

2023
[46]

Pre-trained vision models as perception backbones for safety filters in autonomous driving, 2024

Yuxuan Yang and Hussein Sibai. Pre-trained vision models as perception backbones for safety filters in autonomous driving, 2024. URL https://arxiv.org/abs/2410.22585

2024 arXiv

Pith tools

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

[1] [1]

Safe control using vision-based control barrier function (v-cbf)

Hossein Abdi, Golnaz Raja, and Reza Ghabcheloo. Safe control using vision-based control barrier function (v-cbf). In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 782--788. IEEE, 2023

work page 2023

[2] [2]

Introduction to neural network verification

Aws Albarghouthi. Introduction to neural network verification. CoRR, abs/2109.10317, 2021. URL https://arxiv.org/abs/2109.10317

arXiv 2021

[3] [3]

Control barrier function based quadratic programs for safety critical systems

Aaron D Ames, Xiangru Xu, Jessy W Grizzle, and Paulo Tabuada. Control barrier function based quadratic programs for safety critical systems. IEEE Transactions on Automatic Control, 62 0 (8): 0 3861--3876, 2016

2016

[4] [4]

Ames, Samuel Coogan, Magnus Egerstedt, Gennaro Notomista, Koushil Sreenath, and Paulo Tabuada

Aaron D. Ames, Samuel Coogan, Magnus Egerstedt, Gennaro Notomista, Koushil Sreenath, and Paulo Tabuada. Control barrier functions: Theory and applications. In 2019 18th European Control Conference (ECC), pages 3420--3431, 2019. doi:10.23919/ECC.2019.8796030

arXiv 2019

[5] [5]

Formally verified neural network control barrier certificates for unknown systems

Mahathi Anand and Majid Zamani. Formally verified neural network control barrier certificates for unknown systems. IFAC-PapersOnLine, 56 0 (2): 0 2431--2436, 2023. ISSN 2405-8963. doi:https://doi.org/10.1016/j.ifacol.2023.10.1219. URL https://www.sciencedirect.com/science/article/pii/S2405896323016233. 22nd IFAC World Congress

work page doi:10.1016/j.ifacol.2023.10.1219 2023

[6] [6]

Averaged-dqn: Variance reduction and stabilization for deep reinforcement learning

Oron Anschel, Nir Baram, and Nahum Shimkin. Averaged-dqn: Variance reduction and stabilization for deep reinforcement learning. In International conference on machine learning, pages 176--185. PMLR, 2017

2017

[7] [7]

Autonomous driving—a crash explained in detail

Johannes Betz, Alexander Heilmeier, Alexander Wischnewski, Tim Stahl, and Markus Lienkamp. Autonomous driving—a crash explained in detail. Applied Sciences, 9 0 (23), 2019. ISSN 2076-3417. doi:10.3390/app9235126. URL https://www.mdpi.com/2076-3417/9/23/5126

work page doi:10.3390/app9235126 2019

[8] [8]

Guaranteed safe spacecraft docking with control barrier functions

Joseph Breeden and Dimitra Panagou. Guaranteed safe spacecraft docking with control barrier functions. IEEE Control Systems Letters, 6: 0 2000--2005, 2022. doi:10.1109/LCSYS.2021.3136813

arXiv