Adaptive Autopilot: Constrained DRL for Diverse Driving Behaviors

TL;DR

This paper presents an adaptive autopilot framework using constrained deep reinforcement learning to emulate human driving behaviors safely and effectively across different driving styles.

Contribution

It introduces a novel combination of style classification, neural acceleration prediction, and C-DRL for human-like autonomous driving.

Findings

Rule-based classifier successfully distinguishes driving styles

Neural network accurately predicts human-like acceleration

C-DRL learns safe, human-like driving policies across styles

Abstract

In pursuit of autonomous vehicles, achieving human-like driving behavior is vital. This study introduces adaptive autopilot (AA), a unique framework utilizing constrained-deep reinforcement learning (C-DRL). AA aims to safely emulate human driving to reduce the necessity for driver intervention. Focusing on the car-following scenario, the process involves (i) extracting data from the highD natural driving study and categorizing it into three driving styles using a rule-based classifier; (ii) employing deep neural network (DNN) regressors to predict human-like acceleration across styles; and (iii) using C-DRL, specifically the soft actor-critic Lagrangian technique, to learn human-like safe driving policies. Results indicate effectiveness in each step, with the rule-based classifier distinguishing driving styles, the regressor model accurately predicting acceleration, outperforming traditional car-following models, and C-DRL agents learning optimal policies for humanlike driving across styles.