Application of Deep Reinforcement Learning in Training Autonomous Vehicles: A bibliometric analysis

Abstract

Deep Reinforcement Learning (DRL), a subset of machine learning, combines reinforcement learning with deep learning by using deep neural networks as function. This study assesses the research undertaken for advancement of autonomous vehicles (AVs) using DRL techniques. A bibliometric analysis was conducted using journal papers from Scopus database. A systematic pre-defined screening methodology was applied for selecting the articles. Initial search provided 5964 articles for AVs and DRL. After screening only 401 articles were retained for further investigation indicating only 6.7% relevant articles. The work in this area showed transformative trajectory from 2017 to 2024, characterized by distinct phases of growth. The keywords analysis showed that "reinforcement learning" and "autonomous vehicles" are central with stronger connectivity. The classification of articles in four main categories indicated that most of the articles were related to safety (44%) and traffic efficiency (30%) improvement indicating gap in the literature in other areas. The outcomes of this analysis can lead future research directions.