Pith. sign in

REVIEW 4 major objections 5 minor 3 cited by

CRASH: Challenging Reinforcement-Learning Based Adversarial Scenarios For Safety Hardening

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read This paper shows that an adversarial reinforcement-learning agent can induce crashes in both rule-based and learning-based autonomous driving planners over 90% of the time, and that retraining the planner against a pool of such…

desk verdict A clear falsification demo plus a useful iterative hardening idea, but the headline 26% improvement is under-supported because only crash rates are reported. read the letter →

arxiv 2411.16996 v1 pith:CDEO623R submitted 2024-11-26 cs.LG cs.RO

Amar Kulkarni , Shangtong Zhang , Madhur Behl This is my paper · ORCID
classification cs.LGcs.RO
keywords adversarialreinforcementlearningautonomousvehiclesafetymotionplanningfalsificationhardeningscenariogenerationcollisionratedeepQ-networks
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

CRASH is a two-part framework for simulation-based safety testing of autonomous vehicle motion planners. First, a Deep Q-Network agent controls an adversarial non-player vehicle, rewarded by a signed time-to-collision signal, and learns to collide with the Ego vehicle in more than 90% of episodes against both a rule-based planner and a learning-based planner. Second, the Ego planner is retrained against the adversarial agents over five alternating cycles; sampling adversaries uniformly from a model pool of past opponents cuts the Ego's average collision rate from 64% to 38%—a 26% reduction. The paper's point is that failure scenarios, once automatically discovered, can be fed back into planner training instead of being discarded as one-off corner cases.

What carries the argument

The load-bearing object is the adversarial reward function $r_t = w_1 r_c + w_2 r_x + w_3 r_y$, where $r_c$ is a sparse collision bonus and $r_x, r_y$ are sigmoid-shaped rewards derived from the signed time-to-collision $\lambda_x = \Delta x/\Delta v_x$ and $\lambda_y = \Delta y/\Delta v_y$. The sign of $\lambda$ encodes whether the NPC is approaching the Ego, turning a sparse collision signal into a dense shaping reward that drives learning. This reward feeds a Double DQN with prioritized experience replay for the NPC; the same DQN architecture is reused for the Ego during hardening. Carrying the safety-hardening claim is the bi-level alternation—falsification of the current Ego, then retraining of the Ego against frozen adversaries—instantiated in three sampling strategies: local (only the most recent opponent), uniform model pool, and Elo-prioritized model pool with round-robin tournament evaluation.

What would settle it

Measure the hardened Ego policies' forward progress and lane-change frequency during the collision-rate evaluation: if forward progress collapses in adversarial episodes, the reported 26% reduction reflects avoidance, not hardening; equivalently, if the reduction vanishes against a novel NPC outside the training pool, the effect is overfitting.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a single DQN-based NPC, trained with a collision bonus plus dense sigmoid rewards on signed longitudinal and lateral time-to-collision, can falsify both a rule-based planner (IDM/MOBIL) and a learning-based DQN planner in a two-lane, two-vehicle highway scenario, with crash rates of 97% and 90% respectively. The complementary discovery is that the Ego planner becomes harder to crash when it is iteratively retrained against a pool of all previously trained adversarial NPCs, rather than only the most recent one: uniform pool sampling reduces the Ego's mean crash rate by 26% over five cycles, while local hardening and Elo-prioritized sampling do not improve as consistently. The paper frames falsification and hardening as a bi-level optimization in which the NPC maximizes the collision indicator and the Ego minimizes it, alternating between the two objectives.

Load-bearing premise

The argument assumes that the measured drop in crash rate reflects a genuinely safer driving policy, but the paper reports only collision rates and never measures whether the hardened Ego still moves forward and changes lanes sensibly; if the Ego simply brakes or avoids engaging, the 26% reduction would not demonstrate safety hardening.

Editorial extensions

If this is right

  • Adversarial falsification with signed TTC rewards can automate the discovery of planner failure cases that would otherwise require manual scenario design.
  • The uniform model-pool hardening result implies that, in this setting, remembering how to handle all past adversaries is more effective than specializing against the latest or strongest one.
  • The bi-level alternation formulates AV safety testing as a minimax game, providing a template for iterating between scenario generation and planner improvement.
  • If the 26% reduction transfers to richer scenarios, the same loop could reduce reliance on on-road testing for safety validation.

Reading between the lines

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

  • A natural extension is to replace the binary collision indicator with fault labels, so falsification targets crashes where the AV is clearly responsible; the paper itself notes that not all collisions are equally informative.
  • The signed-TTC reward could serve directly as a dense reward for training defensive Ego policies, not just for adversarial NPCs, since it encodes approach direction rather than mere proximity.
  • The surprising underperformance of Elo-prioritized sampling relative to uniform sampling suggests that, in low-dimensional scenarios, rating noise or overfitting to the strongest opponent hurts generalization; this could be tested by varying the Elo gain and scaling factor.
  • Because the paper enumerates only eight initial configurations, a stronger test of the hardening claim would be to sample initial states continuously and check whether the uniform-pool reduction persists outside the enumerated set.
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

4 major / 5 minor

Summary. The paper introduces CRASH, a two-stage framework for autonomous-vehicle motion planning safety. In the first stage, a Double-DQN adversary controls an NPC in the highway-env simulator to induce collisions with an Ego vehicle, with a shaped reward based on collision and time-to-collision terms. In the second stage, the Ego planner is iteratively retrained against adversarial NPCs using one of three schemes: local safety hardening, uniform model-pool sampling, or prioritized model-pool sampling with Elo ratings. Experiments in a two-lane, two-agent scenario report falsification crash rates of 97% against the rule-based IDM/MOBIL planner and 90% against a learning-based DQN planner, and a claimed 26% reduction in Ego crash rate under uniform model-pool safety hardening.

Significance. If the results hold, the paper makes a useful empirical contribution: it demonstrates that a relatively simple DQN-based adversary can reliably falsify both rule-based and learning-based planners in a controlled highway setting, and it compares three iterative retraining strategies for improving the Ego planner. The falsification results are measured over ten trials and are internally consistent between Figure 8 and Table 2. The model-pool comparison is also a reasonable attempt to study catastrophic forgetting in adversarial training. The main claims are, however, limited by the very small scenario and by the fact that the hardening evaluation reports only crash rates. The paper would be strengthened by behavioral metrics for the hardened Ego, statistical error bars, control retraining baselines, and a clearer statement of what the '26% reduction' means. As an early-stage proof of concept, the work is relevant to simulation-based AV safety testing, but it does not yet support a general claim about safety hardening beyond the specific two-lane, two-agent environment.

major comments (4)
  1. [Section 5.3, Table 2] The safety-hardening claim rests entirely on collision-rate numbers, and the paper does not report any measure of whether the hardened Ego policy still drives: no speed, distance-traveled, lane-change count, or forward-progress metric is given for E0 through E5. Section 3 states that the learning-based planner's reward encourages forward motion, but the hardening evaluation never verifies this, so the 26% reduction could be produced by an Ego that brakes or stops whenever an adversarial NPC approaches. This is a load-bearing confound for the headline hardening result and should be addressed with explicit behavioral metrics from the evaluation episodes, not just crash counts.
  2. [Section 5.3] There is no control condition for retraining the Ego. Because the Ego is retrained against an adversary and then evaluated against the same adversary pool, the 0.64 to 0.38 improvement could reflect generic retraining effects such as additional experience, better DQN tuning, or adaptation to the fixed eight-state distribution rather than the effect of adversarial falsification. The paper should compare against retraining with random-action NPCs, non-adversarial IDM NPCs, or a non-adversarial DQN NPC under the same cycle structure; without such a baseline, the causal attribution of the crash-rate reduction to CRASH's adversarial hardening is not established.
  3. [Table 2 and Figure 9] Table 2 and Figure 9 report single point estimates from 100 episodes with no standard errors, confidence intervals, or number of seeds for the safety-hardening experiments; the 'ten trials' mentioned in Section 5.2 concern the falsification curves only. The '26% reduction' is also an absolute percentage-point change (0.64 to 0.38), not a relative reduction, which would be 40.6%. The text should state this distinction explicitly, especially because the mean rows show large variation across NPC columns, for example Local E5 ranges from 0.20 against V5 to 0.83 against V1.
  4. [Section 4.1 and Figure 7] The adversarial reward weights w1=400, w2=4, w3=1 are selected by inspecting accumulated-reward learning curves on the same falsification task that is later used to report the 97% and 90% crash rates. The reported falsification numbers are therefore conditional on this tuning, and the paper should include a sensitivity analysis or at least state explicitly that the weights were fixed before the final evaluation episodes were run.
minor comments (5)
  1. [Section 3, Eq. (1)] The falsification objective is written with an undefined summation index j and an ambiguous uppercase E; it should be written as a sum over episodes with the terminal collision indicator.
  2. [Section 4.1, Eq. (6)] The Bellman loss appears to swap the value and target network parameters: the target uses θval and the predicted Q-value uses θadv, while the subsequent target update again mixes θval and θadv. This makes the training equations difficult to reproduce.
  3. [Figure 7 (bottom)] The four reward-curve experiments are not labeled in the legend of the bottom panel; the text should identify which plotted curve corresponds to which weight setting.
  4. [Section 5.2] There is an incomplete cross-reference in the sentence 'the DQN implementation in Section for automatic falsification'; the referenced section number is missing.
  5. [Throughout] The paper does not provide a hyperparameter table, environment version, or code/data availability statement, which limits reproducibility of the reported crash rates.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: CRASH's claims are empirical measurements of collision rates, not derivations from fitted equations or self-citations.

full rationale

The paper's central claims are experimental measurements in a fixed simulator environment: an adversarially trained NPC achieves 97% and 90% collision rates against rule-based and learning-based planners, and uniform model-pool safety hardening reduces the average Ego crash rate from 64% to 38% over five cycles. These numbers are read out from simulator episodes, not derived from assumed equations, so there is no equation-level circularity. The NPC's objective and reward are explicitly defined to maximize collisions, but the high collision rates are an empirical result of optimization, not a logical consequence of the definition—the paper shows that sparse rewards fail, demonstrating that success is not forced. The reward weights w1=400, w2=4, w3=1 were selected from learning curves on the same task, which is hyperparameter tuning rather than a circular prediction; the subsequent crash rates are still measured outcomes. Safety hardening's 26% reduction is computed directly from the reported Table 2 means and could have gone the other way, as the local and prioritized baselines show. There are no load-bearing self-citations: the references to Leurent's highway-env, Mnih's DQN, and Treiber's IDM are external and independent. The paper's acknowledged limitations—a simple two-lane scenario, an enumerable initial-state set, and crash-only falsification—concern scope and external validity, not circular derivation. The absence of forward-progress metrics for hardened Ego policies is a correctness/evaluation gap, but it does not make the reported crash-rate reduction circular. Overall, the derivation chain is self-contained and empirical.

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

The quantitative claims depend on a small simulator, tuned reward weights, and unreported RL hyperparameters. The paper introduces no new physical entities. The eight-initial-state enumeration is explicitly a toy assumption, so the results should be read as a method demonstration rather than a general safety guarantee.

free parameters (4)
  • Adversarial reward weights w1, w2, w3 = w1=400, w2=4, w3=1
    Chosen by comparing NPC learning curves in Figure 7 (bottom), not derived from a principle, and retained for all subsequent experiments.
  • Sigmoid scaling parameters a, b in TTC reward = not reported
    Tuned to balance reward sensitivity in Section 4.1, but the values are not given.
  • Elo rating hyperparameters zeta, K, beta = not reported
    Introduced in Section 4.2, these control opponent sampling and directly influence the reported crash rates, yet no values are stated.
  • Standard DQN training hyperparameters = not reported
    Batch size, replay buffer size, target update frequency, epsilon schedule, discount factor, and tau are not reported, and they affect both falsification and hardening results.
assumptions (5)
  • domain assumption State of every vehicle is fully known, with no sensor noise or uncertainty.
    Stated in Section 3 as an assumption, but real AV planning receives noisy perception.
  • domain assumption The state transition model is a black box governed by the simulator, and collision detection is exact and binary.
    Stated in Section 3; the central metric of collision depends entirely on the simulator's collision model.
  • ad hoc to paper The set of initial states is finite and enumerable, namely 8 configurations.
    Stated in Section 3 and acknowledged in Limitations; this only holds for the two-vehicle toy scenario.
  • domain assumption The learning-based Ego planner is re-trainable and its reward encourages both collision avoidance and forward motion.
    Assumed in Section 3; the paper does not verify that hardened policies maintain forward progress or lane discipline.
  • domain assumption The highway-env dynamics provide a realistic enough model for safety conclusions.
    Implicit in the experimental design; the paper only evaluates inside this simulator and provides no validation against real traffic data.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CRASH: Challenging Reinforcement-Learning Based Adversarial Scenarios For Safety Hardening." pith.science (2026). https://pith.science/paper/CDEO623R

@misc{pith2026241116996,
  author       = {Pith},
  title        = {Pith review of: CRASH: Challenging Reinforcement-Learning Based Adversarial Scenarios For Safety Hardening},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CDEO623R}},
  note         = {Machine review of arXiv:2411.16996}
}
read the original abstract

Ensuring the safety of autonomous vehicles (AVs) requires identifying rare but critical failure cases that on-road testing alone cannot discover. High-fidelity simulations provide a scalable alternative, but automatically generating realistic and diverse traffic scenarios that can effectively stress test AV motion planners remains a key challenge. This paper introduces CRASH - Challenging Reinforcement-learning based Adversarial scenarios for Safety Hardening - an adversarial deep reinforcement learning framework to address this issue. First CRASH can control adversarial Non Player Character (NPC) agents in an AV simulator to automatically induce collisions with the Ego vehicle, falsifying its motion planner. We also propose a novel approach, that we term safety hardening, which iteratively refines the motion planner by simulating improvement scenarios against adversarial agents, leveraging the failure cases to strengthen the AV stack. CRASH is evaluated on a simplified two-lane highway scenario, demonstrating its ability to falsify both rule-based and learning-based planners with collision rates exceeding 90%. Additionally, safety hardening reduces the Ego vehicle's collision rate by 26%. While preliminary, these results highlight RL-based safety hardening as a promising approach for scenario-driven simulation testing for autonomous vehicles.

Figures

Figures reproduced from arXiv: 2411.16996 by the authors.

Figure 1
Figure 1. CRASH is a framework that iteratively tests and improves an AV motion planner. First, Automatic Falsifica￾tion uses adversarial NPCs, guided by a reinforcement learn￾ing policy designed to induce collisions with the Ego vehi￾cle. Then, Safety Hardening retrains the Ego motion planner to enhance its robustness against these adversarial scenarios. fornia DMV and NHTSA (Team 2021; NHTSA 2024) pro￾vide limited insights,… view at source ↗
Figure 2
Figure 2. (a) Coordinate system: the blue vehicle ( [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Local Safety Hardening over two cycles: In cy [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (4 more)
Figure 6
Figure 6. Figure 6: Example of cth safety hardening cycle with priori￾tized sampling and tournament evaluation: (1) NPC Vc−1 is trained against EP c−1 , producing Vc. (2) Vc is initialized with 1000 Elo and competes in a round-robin tournament with EP c−1 to update scores. (3) Vc is added…
Figure 5
Figure 5. Figure 5: Uniform Sampling for model pool-based safety [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 8
Figure 8. Figure 8: Average reward and crash rates for Automatic Fal [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]
Figure 9
Figure 9. Figure 9: The average crash rate across each variant of [PITH_FULL_IMAGE:figures/full_fig_p007_9.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. Sutured manifold hierarchies and the Thurston norm

    math.GT 2026-04 unverdicted novelty 6.0 of 10

    Maw dual graphs extract the Thurston norm from taut sutured hierarchies, yielding pretzel-link computations that show wrapping number is not always a seminorm.

  2. Simulating the Unseen: Crash Prediction Must Learn from What Did Not Happen

    cs.LG 2025-05 conditional novelty 4.0 of 10

    Crash prediction should learn from near-miss events and synthetic counterfactual scenarios, not just recorded crashes.

  3. Less is More: A Stealthy and Efficient Adversarial Attack Method for DRL-based Autonomous Driving Policies

    cs.LG 2024-12 conditional novelty 4.0 of 10

    A reinforcement-learning-based adversary learns to trigger collisions in DRL autonomous-driving policies with about three small, well-timed input perturbations per episode.

Reference graph

Works this paper leans on

34 extracted references · 23 canonical work pages · cited by 3 Pith papers

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    ISO 26262: Road vehicles - Functional safety

    2018. ISO 26262: Road vehicles - Functional safety. International Organization for Standardization

    work page 2018

  4. [4]

    Elo, A. E. 1967. The Proposed USCF Rating System, Its Development, Theory, and Applications. Chess Life, XXII(8): 242--247

    work page 1967

  5. [5]

    Exponent. 2024. Quinn Emanuel Report re Cruise. Technical report, Exponent. Prepared for Quinn Emanuel Urquhart & Sullivan, LLP

    work page 2024

  6. [6]

    FARS. 2023. F atality A nalysis R eporting S ystem ( F A R S ) | N H T S A --- nhtsa.gov. https://www.nhtsa.gov/research-data/fatality-analysis-reporting-system-fars. [Accessed 06-11-2023]

    work page 2023

  7. [7]

    Feng, S.; Sun, H.; Yan, X.; Zhu, H.; Zou, Z.; Shen, S.; and Liu, H. X. 2023. Dense reinforcement learning for safety validation of autonomous vehicles. Nature, 615(7953): 620--627. Number: 7953 Publisher: Nature Publishing Group

    work page 2023

  8. [8]

    Feng, S.; Yan, X.; Sun, H.; Feng, Y.; and Liu, H. X. 2021. Intelligent driving intelligence test for autonomous vehicles with naturalistic and adversarial environment. Nature Communications, 12(1): 748. Number: 1 Publisher: Nature Publishing Group

    work page 2021

Show all 34 references
  1. [9]

    J.; Pouget-Abadie, J.; Mirza, M.; Xu, B.; Warde-Farley, D.; Ozair, S.; Courville, A

    Goodfellow, I. J.; Pouget-Abadie, J.; Mirza, M.; Xu, B.; Warde-Farley, D.; Ozair, S.; Courville, A. C.; and Bengio, Y. 2014. Generative Adversarial Nets. In Neural Information Processing Systems

    2014

  2. [10]

    Kalra, N.; and Paddock, S. M. 2016. Driving to safety: How many miles of driving would it take to demonstrate autonomous vehicle reliability? Transportation Research Part A: Policy and Practice, 94: 182--193

    2016

  3. [11]

    S.; Worrall, S.; and Nebot, E

    Karunakaran, D.; Berrio, J. S.; Worrall, S.; and Nebot, E. 2022. Critical concrete scenario generation using scenario-based falsification. ArXiv:2208.13329 [cs]

    2022 arXiv

  4. [12]

    Kesting, A.; Treiber, M.; and Helbing, D. 2007. General Lane-Changing Model MOBIL for Car-Following Models. Transportation Research Record, 1999(1): 86--94

    2007

  5. [13]

    P.; and Ba, J

    Kingma, D. P.; and Ba, J. 2014. Adam: A Method for Stochastic Optimization. CoRR, abs/1412.6980

    2014 arXiv

  6. [14]

    Klischat, M.; and Althoff, M. 2023. Falsifying Motion Plans of Autonomous Vehicles With Abstractly Specified Traffic Scenarios . IEEE Transactions on Intelligent Vehicles, 8(2): 1717--1730. Conference Name: IEEE Transactions on Intelligent Vehicles

    2023

  7. [15]

    D.; Scanlon, J

    Kusano, K. D.; Scanlon, J. M.; Chen, Y.-H.; McMurry, T. L.; yu Chen, R.; Gode, T.; and Victor, T. 2023. Comparison of Waymo Rider-Only Crash Data to Human Benchmarks at 7.1 Million Miles. ArXiv, abs/2312.12675

    2023 arXiv

  8. [16]

    Leurent, E. 2018. An Environment for Autonomous Driving Decision-Making. https://github.com/eleurent/highway-env

    2018

  9. [17]

    D.; and Belta, C

    Mehdipour, N.; Althoff, M.; Tebbens, R. D.; and Belta, C. 2023. Formal methods to comply with rules of the road in autonomous driving: State of the art and grand challenges. Automatica, 152: 110692

    2023

  10. [18]

    A.; Veness, J.; Bellemare, M

    Mnih, V.; Kavukcuoglu, K.; Silver, D.; Rusu, A. A.; Veness, J.; Bellemare, M. G.; Graves, A.; Riedmiller, M.; Fidjeland, A. K.; Ostrovski, G.; Petersen, S.; Beattie, C.; Sadik, A.; Antonoglou, I.; King, H.; Kumaran, D.; Wierstra, D.; Legg, S.; and Hassabis, D. 2015. Human-leve...

    2015

  11. [19]

    Neelofar; and Aleti, A. 2022. Identifying Safety -critical Scenarios for Autonomous Vehicles via Key Features . ArXiv:2212.07566 [cs]

    2022 arXiv

  12. [20]

    NHTSA . 2024. ADS Crash Report ID 12345. Technical report, U.S. Department of Transportation. Report retrieved from the Standing General Order on Crash Reporting

    2024

  13. [21]

    Pan, X.; Seita, D.; Gao, Y.; and Canny, J. 2019. Risk Averse Robust Adversarial Reinforcement Learning. arXiv:1904.00511

    2019 arXiv

  14. [22]

    Pinto, L.; Davidson, J.; Sukthankar, R.; and Gupta, A. 2017. Robust Adversarial Reinforcement Learning. arXiv:1703.02702

    2017 arXiv

  15. [23]

    J.; Fidler, S.; and Litany, O

    Rempe, D.; Philion, J.; Guibas, L. J.; Fidler, S.; and Litany, O. 2022. Generating Useful Accident-Prone Driving Scenarios via a Learned Traffic Prior. In Conference on Computer Vision and Pattern Recognition (CVPR)

    2022

  16. [24]

    Rengarajan, D.; Vaidya, G.; Sarvesh, A.; Kalathil, D.; and Shakkottai, S. 2022. Reinforcement Learning with Sparse Rewards using Guidance from Offline Demonstration. arXiv:2202.04628

    2022 arXiv

  17. [25]

    Schaul, T.; Quan, J.; Antonoglou, I.; and Silver, D. 2016. Prioritized Experience Replay . ArXiv:1511.05952 [cs]

    2016 arXiv

  18. [26]

    Sharif, A.; and Marijan, D. 2022. Adversarial Deep Reinforcement Learning for Improving the Robustness of Multi-agent Autonomous Driving Policies. In 2022 29th Asia-Pacific Software Engineering Conference ( APSEC ) . IEEE

    2022

  19. [27]

    Sun, H.; Feng, S.; Yan, X.; and Liu, H. X. 2021. Corner Case Generation and Analysis for Safety Assessment of Autonomous Vehicles. arXiv:2102.03483

    2021 arXiv

  20. [28]

    Team, S. D. C. 2021. Metadata record for: Crash and disengagement data of autonomous vehicles on public roads in California

    2021

  21. [29]

    Treiber, M.; Hennecke, A.; and Helbing, D. 2000. Congested traffic states in empirical observations and microscopic simulations. Physical Review E, 62(2): 1805--1824

    2000

  22. [30]

    Turlej, W.; and Pankiewicz, N. 2021. Adversarial Trajectories Generation for Automotive Applications. In 2021 25th International Conference on Methods and Models in Automation and Robotics (MMAR), 115--120

    2021

  23. [31]

    ULSE. 2020. UL 4600: Standard for Evaluation of Autonomous Products . Standard for safety. Underwriters Laboratories

    2020

  24. [32]

    van Hasselt, H.; Guez, A.; and Silver, D. 2015. Deep Reinforcement Learning with Double Q-learning. arXiv:1509.06461

    2015 arXiv

  25. [33]

    u tt, B.; Utesch, F.; Kramer, B.; Gutenkunst, C.; and B \

    Westhofen, L.; Neurohr, C.; Koopmann, T.; Butz, M.; Sch \" u tt, B.; Utesch, F.; Kramer, B.; Gutenkunst, C.; and B \" o de, E. 2021. Criticality Metrics for Automated Driving: A Review and Suitability Analysis of the State of the Art. CoRR, abs/2108.02403

    2021 arXiv

  26. [34]

    Yan, X.; Zou, Z.; Feng, S.; Zhu, H.; Sun, H.; and Liu, H. X. 2023. Learning naturalistic driving environment with statistical realism. Nature Communications, 14(1): 2037. Number: 1 Publisher: Nature Publishing Group

    2023

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.