Biologically Inspired Spiking Neural Network for Traffic Risk Assessment

A recent study suggests a novel approach to interpreting Surrogate Safety Measures (SSMs) by viewing them as neuron spiking thresholds through a spiking neural network (SNN), aimed at improving the understanding of human braking reactions. This SNN utilizes leaky integrate-and-fire (LIF) neurons, processing various SSM inputs and is trained to synchronize spikes with the onset of human braking. The training data was collected from a controlled car-following study conducted on the 3D-CoAutoSim platform, which incorporates CARLA/Unreal and a 6-DOF motion platform, where critical events were artificially created. Findings indicate that the learned spiking activity closely resembles braking behavior in different scenarios, providing a more dynamic option compared to traditional fixed-threshold SSM assessments.