Autonomous driving systems (ADS) are at the forefront of transforming the future of transportation promises of increased safety, efficiency, and mobility. Nevertheless, it remains a big problem to ensure the safety and reliability of these systems in complex and dynamic environments. In recent years, due to the efforts of optimal and safe decision-making under certainty from reinforcement learning (RL), especially Proximal Policy Optimization (PPO), the use of reinforcement learning has been shown to be a viable approach to balance performance and safety in Ads. In this review we examine recent advances in RL-based ADS and those in PPO and how it can be coupled with key safety mechanisms. Through a systematic literature review (SLR), this paper stakes out a few high impact research questions relating to safety, decision-making functionality, and limitations of RL-based ADS models. We find that PPO can help ensure stability and performance in dynamic environments and highlight challenges pointing to computational and real-world applicability. The importance of future research to regulatory compliance, computational efficiency and ethical decision-making is also emphasized by the paper. The proposed directions aim to bridge the gap between simulation successes and real-world deployment of RL-based ADS, ensuring their safe and scalable implementation.