| 제목 |
심층 강화학습 기반 자율 주행 자동차의 램프구간 주행 정책 연구 |
| 분야 |
전기ㆍ전자ㆍ통신 |
| 언어 |
Korean |
| 저자 |
이재휘(숭실대학교), 엄찬인(숭실대학교), 이동수(숭실대학교 ), 권민혜(숭실대학교) |
| Key Words |
Autonomous driving system(자율주행 시스템), Deep reinforcement learning(심층 강화학습), Partially observable Markov decision process(부분 관측 가능한 마르코프 의사결정), On-ramp merge(램프구간 병합), Decisionmaking( 판단) |
| 초록 |
As autonomous driving systems attract more attention, it is important to develop a driving strategy on complex road conditions, e.g., on-ramp merging scenarios. Deep Reinforcement Learning (RL) is a promising solution for building an autonomous driving policy because it uses neural networks as functional approximators, enabling autonomous vehicles to adapt to dynamic and unpredictable scenarios. In this study, we aimed to develop RL based, on-ramp merging strategies that minimize disruption to traffic flow and ensure safe merging. Specifically, we designed a Partially Observable Markov Decision Process (POMDP), and then trained the driving strategy by using three deep RL algorithms. Simulation results demonstrated that the RL-based driving strategy could outperform the control theoretic strategy, thus improving the traffic flow in the ramp lane by 4.52% and the main lane by 2.13%. |
| 원문(PDF) |
다운로드
Journal Site
|
