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

Automatic Emergency Braking (AEB) systems are becoming mandatory in all new US light vehicles by September 2029 under NHTSA’s FMVSS No. 127. However, current production implementations rely on deterministic thresholds (e.g., stopping distance, Time-to-Collision) that ignore real-world uncertainties in sensing, road conditions, and vehicle dynamics. To address this, researchers from the University of Waterloo (Akshay Karjol and Shadi Alawneh) developed a GPU-accelerated Monte Carlo framework that evaluates AEB performance under uncertainty in real time. The model uses a high-fidelity longitudinal vehicle model with 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, and deterministic CPU-generated sampling ensures bit-exact numerical consistency between CPU and GPU implementations.

Tested across four hardware platforms (laptop GPUs GTX 1650 and RTX 5070, and embedded Jetson Orin Nano and Jetson AGX Orin), the framework achieved peak speedups of 54.57x while maintaining exact numerical agreement with CPU baselines. Crucially, the authors performed a real-time feasibility analysis using a complete AEB timing budget: 700ms human reaction time minus 120ms for perception and 50ms for decision overhead, leaving a 530ms window for computation. On the Jetson AGX Orin, the system can execute approximately 25,000 Monte Carlo samples within that budget, enabling real-time probabilistic AEB evaluation as part of a complete embedded pipeline. This work establishes Monte Carlo-based uncertainty evaluation as a deployable runtime component rather than an offline validation tool, providing quantitative guidance for risk-aware AEB threshold selection under the NHTSA final rule.