Knowledge Distillation Neural Network for Predicting Car-following Behaviour of Human-driven and Autonomous Vehicles

TL;DR

This paper introduces a Knowledge Distillation Neural Network (KDNN) for predicting car-following behavior, demonstrating comparable accuracy to LSTM models while being more efficient and safer in mixed traffic scenarios.

Contribution

The study develops a novel KDNN model that outperforms traditional models and standalone neural networks in predicting vehicle following behavior with improved safety and efficiency.

Findings

KDNN achieves similar accuracy to LSTM in speed prediction.

KDNN outperforms MLP and Gipps model in collision prevention.

KDNN requires less computational power, suitable for real-time applications.

Abstract

As we move towards a mixed-traffic scenario of Autonomous vehicles (AVs) and Human-driven vehicles (HDVs), understanding the car-following behaviour is important to improve traffic efficiency and road safety. Using a real-world trajectory dataset, this study uses descriptive and statistical analysis to investigate the car-following behaviours of three vehicle pairs: HDV-AV, AV-HDV and HDV-HDV in mixed traffic. The ANOVA test showed that car-following behaviours across different vehicle pairs are statistically significant (p-value < 0.05). We also introduce a data-driven Knowledge Distillation Neural Network (KDNN) model for predicting car-following behaviour in terms of speed. The KDNN model demonstrates comparable predictive accuracy to its teacher network, a Long Short-Term Memory (LSTM) network, and outperforms both the standalone student network, a Multilayer Perceptron (MLP), and traditional physics-based models like the Gipps model. Notably, the KDNN model better prevents collisions, measured by minimum Time-to-Collision (TTC), and operates with lower computational power, making it ideal for AVs or driving simulators requiring efficient computing.