TY  - JOUR
T1  - Autonomous Driving Strategy for Bottleneck Traffic with Prioritized Experience Replay
AU  - Eom, Chanin 
AU  - Lee, Dongsu 
AU  - Kwon, Minhae 
JO  - The Journal of Korean Institute of Communications and Information Sciences
PY  - 2023
DA  - 2023/1/14
DO  - 10.7840/kics.2023.48.6.690
KW  - Autonomous driving system
KW  - Bottleneck traffic
KW  - Deep reinforcement learning
KW  - Partially observable Markov decision process
KW  - Twin delayed deep deterministic policy gradient
KW  - Prioritized experience replay
AB  - Recently, the demand for a higher level of autonomous driving technology in complex circumstances has increased. Deep reinforcement learning is gaining attention as a promising solution that enables instant decision-making based on high-dimensional state information. In this study, we propose a Partially Observable Markov Decision Process (POMDP) to train the autonomous driving policy that can successfully navigate bottleneck congestion. Furthermore, we suggest using the prioritized experience replay method in Twin Delayed Deep Deterministic Policy Gradient (TD3) to train the policy. As a result, we confirm that the vehicles trained with the prioritized experience replay method outperform the vehicles trained with the random experience replay method.