arxiv: 2604.27193 · v1 · submitted 2026-04-29 · 💻 cs.RO · cs.CE· cs.DC· cs.SY· eess.SY

Recognition: unknown

Real-Time GPU-Accelerated Monte Carlo Evaluation of Safety-Critical AEB Systems Under Uncertainty

Akshay Karjol , Shadi Alawneh

Authors on Pith no claims yet

Pith reviewed 2026-05-07 09:42 UTC · model grok-4.3

classification 💻 cs.RO cs.CEcs.DCcs.SYeess.SY

keywords automatic emergency brakingMonte Carlo simulationGPU accelerationreal-time embedded systemsvehicle dynamicsuncertainty quantificationsafety-critical controlNHTSA FMVSS

0 comments

The pith

GPU acceleration enables real-time Monte Carlo evaluation of automatic emergency braking on embedded automotive hardware.

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

Traditional AEB systems rely on fixed stopping-distance or time-to-collision thresholds that overlook uncertainties from sensors, road conditions, and vehicle dynamics. This work builds a framework that runs many independent simulations of a detailed longitudinal vehicle model on GPUs to compute probabilistic braking outcomes. The one-thread-per-sample approach produces exact numerical matches with CPU versions and delivers large speedups on both laptop and embedded platforms. On the Jetson AGX Orin, roughly 25,000 samples complete inside the 530 ms window left after subtracting perception and decision time from a 700 ms total budget. This moves Monte Carlo uncertainty analysis from an offline design tool into a feasible runtime component that can inform threshold choices under upcoming NHTSA requirements.

Core claim

The GPU-accelerated Monte Carlo framework, built around a high-fidelity longitudinal vehicle model that includes aerodynamic drag, road grade, brake actuator dynamics, and weight transfer, executes approximately 25,000 independent samples within a 530 ms budget on the Jetson AGX Orin while achieving peak speedups of 54.57x and preserving bit-exact numerical agreement with CPU implementations across tested platforms.

What carries the argument

One-thread-per-sample GPU execution of independent Monte Carlo rollouts on a high-fidelity longitudinal vehicle model incorporating aerodynamic drag, road grade, brake actuator dynamics, and weight transfer.

Load-bearing premise

The high-fidelity longitudinal vehicle model captures the dominant sources of uncertainty relevant to safety-critical AEB decisions.

What would settle it

A side-by-side comparison on a physical test track where the framework's probabilistic braking decisions produce stopping distances that systematically differ from measured real-world outcomes under matched initial conditions and uncertainties.

Figures

Figures reproduced from arXiv: 2604.27193 by Akshay Karjol, Shadi Alawneh.

**Figure 1.** Figure 1: CPU vs. GPU stopping-distance comparison across all four hardware view at source ↗

**Figure 3.** Figure 3: Collision probability P(Dstop > H0) as a function of initial headway, computed from N = 12,000 Monte Carlo samples. Horizontal dashed lines indicate risk tolerance thresholds: aggressive (5%), conservative (1%), and very conservative (0.1%). Annotations show minimum safe headway for each threshold. • 0.1% risk (very conservative): H0 = 122.7 m These headway thresholds translate to TTC criteria via TTC = H0… view at source ↗

**Figure 4.** Figure 4: GPU speedup relative to CPU baseline across four hardware plat view at source ↗

**Figure 5.** Figure 5: GPU execution time vs. sample count across four platforms with view at source ↗

**Figure 6.** Figure 6: Maximum Monte Carlo sample count achievable within the view at source ↗

read the original abstract

Automatic Emergency Braking (AEB) systems represent a safety-critical national interest, with the National Highway Traffic Safety Administration (NHTSA) Federal Motor Vehicle Safety Standard (FMVSS No. 127) requiring AEB in all new light vehicles sold in the United States by September 2029. However, production implementations frequently rely on deterministic stopping-distance or Time-to-Collision (TTC) thresholds that fail to capture uncertainty in sensing, road conditions, and vehicle dynamics. This paper presents a GPU-accelerated Monte Carlo framework for stochastic evaluation of emergency braking performance using a high-fidelity longitudinal vehicle model incorporating aerodynamic drag, road grade, brake actuator dynamics, and weight transfer effects. A one-thread-per-sample execution strategy exploits the independence of Monte Carlo rollouts, while deterministic CPU-generated sampling ensures bit-exact numerical consistency between CPU and GPU implementations. The framework is evaluated across four hardware platforms spanning development and deployment environments: two laptop GPUs (GTX 1650, RTX 5070) and two automotive-grade embedded platforms (Jetson Orin Nano, Jetson AGX Orin). Peak speedups of 54.57x are achieved while maintaining exact numerical agreement. Real-time feasibility analysis with a complete AEB timing budget (700 ms human reaction time minus 120 ms perception and 50 ms decision overhead) demonstrates that the Jetson AGX Orin can execute approximately 25,000 Monte Carlo samples within a 530 ms budget, enabling real-time probabilistic AEB evaluation as part of a complete embedded pipeline. These results establish Monte Carlo-based uncertainty evaluation as a deployable runtime component rather than an offline validation tool and provide quantitative guidance for risk-aware AEB threshold selection under the NHTSA final rule.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The manuscript presents a GPU-accelerated Monte Carlo framework for stochastic evaluation of Automatic Emergency Braking (AEB) systems under uncertainty. It employs a high-fidelity longitudinal vehicle model (aerodynamic drag, road grade, brake actuator dynamics, weight transfer) with a one-thread-per-sample GPU strategy and deterministic CPU sampling for bit-exact consistency. Evaluations on laptop GPUs and Jetson embedded platforms report peak speedups of 54.57x and show that the Jetson AGX Orin can execute ~25,000 samples within a 530 ms real-time budget (after subtracting perception and decision overhead from a 700 ms total), enabling runtime probabilistic AEB assessment.

Significance. If the results hold, the work is significant for embedded robotics and autonomous vehicle safety. It demonstrates that Monte Carlo uncertainty quantification can transition from offline validation to real-time deployment on automotive-grade hardware, providing quantitative support for risk-aware AEB thresholds under NHTSA FMVSS No. 127. Credit is due for the exact numerical agreement between CPU and GPU implementations and the concrete cross-platform timing benchmarks that directly address embedded feasibility.

major comments (1)

[Abstract] Abstract: the claim that the high-fidelity longitudinal vehicle model captures the dominant sources of uncertainty relevant to safety-critical AEB decisions lacks supporting analysis or justification. No evidence is given that lateral dynamics, tire-road friction variability, or correlated sensor noise can be neglected without materially affecting collision probability or stopping-distance tail estimates; if any of these factors are non-negligible, the Monte Carlo output misrepresents risk even when the reported timing and speedup numbers are accurate.

minor comments (1)

[Abstract] Abstract: the one-thread-per-sample execution strategy and deterministic CPU-generated sampling are described at a high level; additional detail on the specific uncertainty distributions and how bit-exact reproducibility is verified across hardware would improve clarity without altering the central claims.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive review and recognition of the significance of demonstrating real-time Monte Carlo feasibility on embedded automotive hardware. We address the single major comment below.

read point-by-point responses

Referee: [Abstract] Abstract: the claim that the high-fidelity longitudinal vehicle model captures the dominant sources of uncertainty relevant to safety-critical AEB decisions lacks supporting analysis or justification. No evidence is given that lateral dynamics, tire-road friction variability, or correlated sensor noise can be neglected without materially affecting collision probability or stopping-distance tail estimates; if any of these factors are non-negligible, the Monte Carlo output misrepresents risk even when the reported timing and speedup numbers are accurate.

Authors: We agree that the abstract wording risks overstating completeness. The manuscript deliberately restricts the vehicle model to longitudinal dynamics because NHTSA FMVSS No. 127 AEB requirements and the associated test procedures focus on straight-line stopping distance and time-to-collision; lateral motion and yaw are handled by separate stability systems. Tire-road friction variability and correlated sensor noise are acknowledged as relevant but would necessitate a full 3-D multi-body model and stochastic perception pipeline, which lies outside the scope of a paper whose primary contribution is proving that Monte Carlo rollouts can run in real time on Jetson-class hardware. We will revise the abstract to state that the model incorporates the principal longitudinal uncertainties (aerodynamic drag, road grade, brake actuator dynamics, and weight transfer) without claiming dominance over all possible sources. We will also insert a concise justification paragraph in the vehicle-model section (Section 3) that cites the regulatory focus on longitudinal performance and explicitly flags lateral/friction/sensor extensions as future work. The timing, speedup, and bit-exact CPU/GPU agreement results remain unchanged by these clarifications. revision: yes

Circularity Check

0 steps flagged

No circularity: performance claims rest on direct hardware measurements and an independent Monte Carlo implementation.

full rationale

The paper's derivation chain is a straightforward description of a GPU-parallel Monte Carlo rollout using an explicitly stated longitudinal vehicle model, followed by empirical timing benchmarks on four hardware platforms. No equations are present that define a quantity in terms of itself, no fitted parameters are relabeled as predictions, and no load-bearing steps rely on self-citations or imported uniqueness theorems. The 25,000-sample / 530 ms claim is presented as a measured outcome under a stated timing budget, not derived from prior results by the same authors. The model assumptions (aerodynamic drag, road grade, etc.) are declared inputs rather than outputs of the framework, so the central timing result does not reduce to its own inputs by construction. This is the common case of a self-contained empirical engineering paper.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The work rests on standard Monte Carlo convergence assumptions and GPU parallel execution models; no new free parameters, ad-hoc axioms, or invented physical entities are introduced in the abstract.

axioms (1)

standard math Monte Carlo sampling yields reliable probability estimates when the number of independent rollouts is sufficiently large.
Invoked implicitly by the claim that 25,000 samples enable real-time probabilistic evaluation.

pith-pipeline@v0.9.0 · 5637 in / 1186 out tokens · 34808 ms · 2026-05-07T09:42:53.939347+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

35 extracted references

[1]

Federal motor vehicle safety standards; automatic emergency braking systems for light vehi- cles: Final rule,

National Highway Traffic Safety Administration, “Federal motor vehicle safety standards; automatic emergency braking systems for light vehi- cles: Final rule,” Federal Register, 89 FR 40118, 2024, fMVSS No. 127, published May 9, 2024, effective September 1, 2029

2024
[2]

Traffic safety facts 2021: A compilation of motor vehicle crash data,

——, “Traffic safety facts 2021: A compilation of motor vehicle crash data,” U.S. Department of Transportation, Washington, DC, Tech. Rep. DOT HS 813 375, 2023

2021
[3]

Time-to-collision and collision avoidance systems,

R. van der Horst and J. Hogema, “Time-to-collision and collision avoidance systems,”Accident Analysis & Prevention, vol. 45, pp. 507– 517, 2012

2012
[4]

Iso 22737: Intelligent transport systems — pedestrian automatic emergency braking systems — performance requirements and test procedures,

International Organization for Standardization, “Iso 22737: Intelligent transport systems — pedestrian automatic emergency braking systems — performance requirements and test procedures,” 2021, standard

2021
[5]

Stochastic situation assessment in advanced driver assistance system for complex multi- objects traffic situations,

A. Berthelot, A. Tamke, T. Dang, and G. Breuel, “Stochastic situation assessment in advanced driver assistance system for complex multi- objects traffic situations,” in2012 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 2012, pp. 1180–1185

2012
[6]

Robust design of an automatic emergency braking system considering sensor mea- surement errors,

C. St ¨ockle, W. Utschick, S. Herrmann, and T. Dirndorfer, “Robust design of an automatic emergency braking system considering sensor mea- surement errors,” in2018 21st International Conference on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 2015–2021

2018
[7]

Prediction-uncertainty- aware threat detection for adas: A case study on lane-keeping assis- tance,

J. Dahl, G. R. de Campos, and J. Fredriksson, “Prediction-uncertainty- aware threat detection for adas: A case study on lane-keeping assis- tance,”IEEE Transactions on Intelligent Vehicles, vol. 8, no. 4, pp. 2914–2925, 2023

2023
[8]

Automated vehicular safety systems: Robust function and sensor design,

C. St ¨ockle, S. Herrmann, T. Dirndorfer, and W. Utschick, “Automated vehicular safety systems: Robust function and sensor design,”IEEE Signal Processing Magazine, vol. 37, no. 4, pp. 24–35, 2020

2020
[9]

C. P. Robert and G. Casella,Monte Carlo Statistical Methods, 2nd ed. New York: Springer, 2004. 10

2004
[10]

Accelerating quantum monte carlo simulations of real materials on gpu clusters,

K. Esler, J. Kim, D. Ceperley, and L. Shulenburger, “Accelerating quantum monte carlo simulations of real materials on gpu clusters,” Computing in Science & Engineering, vol. 14, no. 1, pp. 40–46, 2012

2012
[11]

Safety-dominant stochastic model predictive decision-making considering obstacle tra- jectory uncertainties for intelligent vehicles,

Q. Dai, H. Chen, J. Liu, Q. Meng, H. Guo, and D. Cao, “Safety-dominant stochastic model predictive decision-making considering obstacle tra- jectory uncertainties for intelligent vehicles,”IEEE Transactions on Intelligent Vehicles, vol. 10, no. 3, pp. 1638–1656, 2025

2025
[12]

Gpu computing,

J. D. Owens, M. Houston, D. Luebke, S. Green, J. E. Stone, and J. C. Phillips, “Gpu computing,”Proceedings of the IEEE, vol. 96, no. 5, pp. 879–899, 2008

2008
[13]

Understanding the efficiency of gpu algorithms for monte carlo simulation,

T. Aila and S. Laine, “Understanding the efficiency of gpu algorithms for monte carlo simulation,” inProceedings of High Performance Graphics, 2009, pp. 87–94

2009
[14]

Robust function and sensor design considering sensor measurement errors applied to automatic emergency braking,

C. St ¨ockle, W. Utschick, S. Herrmann, and T. Dirndorfer, “Robust function and sensor design considering sensor measurement errors applied to automatic emergency braking,” in2019 IEEE Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 2284–2289

2019
[15]

An efficient approach to simulation-based robust function and sensor design applied to an automatic emergency braking system,

M. L. Leyrer, C. St ¨ockle, S. Herrmann, T. Dirndorfer, and W. Utschick, “An efficient approach to simulation-based robust function and sensor design applied to an automatic emergency braking system,” in2020 IEEE Intelligent Vehicles Symposium (IV). IEEE, 2020, pp. 1440–1447

2020
[16]

Uncertainty-aware stopping distance prediction for automated driving using stochastic simulation,

J. Norden, C. Sch ¨oller, and C. Stachniss, “Uncertainty-aware stopping distance prediction for automated driving using stochastic simulation,” IEEE Transactions on Intelligent Vehicles, 2023, early access/details as available in IEEE Xplore

2023
[17]

A multi-gpu approach for the exchange monte carlo method,

C. A. Navarro, H. Wei, and Y . Deng, “A multi-gpu approach for the exchange monte carlo method,” in2015 34th International Conference of the Chilean Computer Science Society (SCCC). IEEE, 2015, pp. 1–9

2015
[18]

Resource-constrained machine learning for adas: A systematic review,

J. Borrego-Carazo, D. Castells-Rufas, E. Biempica, and J. Carrabina, “Resource-constrained machine learning for adas: A systematic review,” IEEE Access, vol. 8, pp. 40 573–40 605, 2020

2020
[19]

Performance analysis and optimization of automotive gpus,

F. Mazzocchetti, P. Benedicte, H. Tabani, L. Kosmidis, J. Abella, and F. J. Cazorla, “Performance analysis and optimization of automotive gpus,” in2019 31st International Symposium on Computer Architecture and High Performance Computing (SBAC-PAD). IEEE, 2019, pp. 96– 103

2019
[20]

Evaluation of thermal imag- ing on embedded gpu platforms for application in vehicular assistance systems,

M. A. Farooq, W. Shariff, and P. Corcoran, “Evaluation of thermal imag- ing on embedded gpu platforms for application in vehicular assistance systems,”IEEE Transactions on Intelligent Vehicles, vol. 8, no. 2, pp. 1130–1143, 2023

2023
[21]

Cell-based update algorithm for occupancy grid maps and hybrid map for adas on embedded gpus,

J. Fickenscher, J. Schlumberger, F. Hannig, J. Teich, and M. E. Bouzouraa, “Cell-based update algorithm for occupancy grid maps and hybrid map for adas on embedded gpus,” in2018 Design, Automation & Test in Europe Conference (DATE), 2018, pp. 443–448

2018
[22]

The Wiener-Askey polynomial chaos for stochastic differential equations,

D. Xiu and G. E. Karniadakis, “The Wiener-Askey polynomial chaos for stochastic differential equations,”SIAM Journal on Scientific Computing, vol. 24, no. 2, pp. 619–644, 2002

2002
[23]

Uncertainty quantification using generalized polynomial chaos for online simulations of automotive propulsion systems,

H. Yang, N. Kidambi, Y . Fujii, A. Gorodetsky, and K.-W. Wang, “Uncertainty quantification using generalized polynomial chaos for online simulations of automotive propulsion systems,” in2020 American Control Conference (ACC). IEEE, 2020, pp. 2427–2432

2020
[24]

C. E. Rasmussen and C. K. I. Williams,Gaussian Processes for Machine Learning. Cambridge, MA: MIT Press, 2006

2006
[25]

Gaussian process based model predictive control for overtaking scenarios at highway curves,

W. Liu, Y . Zhai, G. Chen, and A. Knoll, “Gaussian process based model predictive control for overtaking scenarios at highway curves,” in2022 IEEE Intelligent Vehicles Symposium (IV). IEEE, 2022, pp. 1–6

2022
[26]

Proba- bilistic metamodels for an efficient characterization of complex driving scenarios,

M. Winkelmann, M. Kohlhoff, H. H. Tadjine, and S. M ¨uller, “Proba- bilistic metamodels for an efficient characterization of complex driving scenarios,”IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 12, pp. 23 896–23 908, 2022

2022
[27]

Analytical derivation of performance bounds of autonomous emergency brake systems,

J. E. Stellet, P. V ogt, J. Schumacher, W. Branz, and J. M. Z ¨ollner, “Analytical derivation of performance bounds of autonomous emergency brake systems,” in2016 IEEE Intelligent Vehicles Symposium (IV). IEEE, 2016, pp. 220–226

2016
[28]

A survey on motion prediction and risk assessment for intelligent vehicles,

S. Lef `evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk assessment for intelligent vehicles,”Robotics and Autonomous Systems, vol. 58, no. 1, pp. 61–71, 2014

2014
[29]

J. Y . Wong,Theory of Ground Vehicles, 5th ed. Hoboken, NJ: Wiley, 2022

2022
[30]

Rajamani,Vehicle Dynamics and Control, 2nd ed

R. Rajamani,Vehicle Dynamics and Control, 2nd ed. New York, NY: Springer, 2012

2012
[31]

G. F. Franklin, J. D. Powell, and A. Emami-Naeini,Feedback Control of Dynamic Systems, 8th ed. Upper Saddle River, NJ: Pearson, 2019

2019
[32]

J. C. Butcher,Numerical Methods for Ordinary Differential Equations, 3rd ed. Chichester, UK: Wiley, 2016

2016
[33]

How long does it take to stop? methodological analysis of driver perception-brake times,

M. Green, “How long does it take to stop? methodological analysis of driver perception-brake times,”Transportation Human Factors, vol. 2, no. 3, pp. 195–216, 2000

2000
[34]

A survey of motion planning and control techniques adopted in self-driving vehicles,

B. Paden, M. ˇC´ap, S. Z. Yong, D. Yershov, and E. Frazzoli, “A survey of motion planning and control techniques adopted in self-driving vehicles,” IEEE Transactions on Intelligent Vehicles, vol. 1, no. 1, pp. 33–55, 2016

2016
[35]

Lane change maneuvers for automated vehicles,

J. Nilsson, M. Br ¨annstr¨om, E. Coelingh, and J. Fredriksson, “Lane change maneuvers for automated vehicles,”IEEE Transactions on Intel- ligent Transportation Systems, vol. 18, no. 5, pp. 1087–1096, 2017. Akshay Karjol(Senior Member, IEEE) received the M.S. degree in auto- motive engineering from Clemson University–International Center for Auto- motive R...

2017