Reinterpreting Safety Thresholds as Neuron Spiking Thresholds

Researchers propose a biologically-inspired approach to safety thresholds in autonomous driving by modeling Surrogate Safety Measures (SSMs) as leaky integrate-and-fire neuron spiking thresholds within a spiking neural network. Trained on human braking data from controlled experiments, the SNN captures dynamic safety responses that fixed thresholds miss, potentially bridging the gap between objective risk metrics and subjective human perception.

This research addresses a fundamental limitation in autonomous vehicle safety assessment: fixed safety thresholds fail to account for how humans actually respond to sustained borderline risks or sudden peaks. Traditional SSM evaluations use static trigger points that don't reflect the nuanced temporal dynamics of human decision-making. By reinterpreting safety thresholds through the lens of neuronal spiking dynamics, the researchers create a system that learns when and how to respond to multiple safety signals simultaneously.

The study builds on established computational neuroscience concepts, leveraging leaky integrate-and-fire models that have proven valuable in understanding biological neural computation. The experiment setup using CARLA simulation and a motion platform represents a sophisticated methodology for collecting realistic human response data. Across different participants, the analysis reveals that while learned input thresholds remain relatively consistent—suggesting generalizable safety principles—decay factors vary, indicating individual differences in temporal sensitivity to risk.

For the autonomous vehicle industry, this approach offers practical value by improving safety prediction models beyond simple threshold crossing logic. Better alignment between objective metrics and human behavior perception reduces both false positives that erode user trust and false negatives that pose genuine safety risks. This work demonstrates how biologically-inspired computational models can enhance safety-critical systems by capturing human-like adaptive responses.

Future research should validate these findings across diverse driving scenarios, weather conditions, and vehicle types. The framework could extend beyond passenger vehicles to commercial autonomous systems, and similar approaches might apply to other domains requiring human-machine safety alignment.

• →Spiking neural networks model safety thresholds dynamically, capturing human braking behavior better than fixed-threshold approaches
• →Learned decay factors vary across individuals, encoding different temporal sensitivities to safety measures
• →The approach bridges objective safety metrics with subjective human perception in autonomous driving contexts
• →Biologically-inspired models improve prediction accuracy for critical safety events by accounting for sustained risk conditions
• →This framework could improve user trust in autonomous systems by reducing false alarms and safety gaps