Akkumula: Evidence accumulation driver models with Spiking Neural Networks

TL;DR

Akkumula introduces a novel evidence accumulation framework using Spiking Neural Networks to enhance driver models, enabling realistic, adaptable, and efficient simulation of driving behaviors based on perceptual inputs.

Contribution

It is the first framework to integrate Spiking Neural Networks with evidence accumulation for driver modeling, improving realism and scalability.

Findings

Successfully reproduces braking, accelerating, and steering behaviors.

Scales to large datasets and adapts to various driving scenarios.

Maintains transparency of internal decision processes.

Abstract

Processes of evidence accumulation can make driver models more realistic, by explaining how drivers adjust their actions based on perceptual inputs and decision boundaries. The absence of a standard modelling approach limits their adoption; existing methods are hand-crafted, hard to adapt, and computationally inefficient. This paper presents Akkumula, an evidence accumulation modelling framework that uses Spiking Neural Networks and other deep learning techniques. Tested on data from a test-track experiment, the model can reproduce the time course of braking, accelerating, and steering. Akkumula integrates with existing machine learning architectures, scales to large datasets, adapts to different driving scenarios, and keeps its internal logic relatively transparent.