Multi-actor multi-critic method for traffic accident anticipation교통사고 예측을 위한 멀티 액터 멀티 크리틱 방법론

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 234
  • Download : 0
DC FieldValueLanguage
dc.contributor.advisorHar, Dongsoo-
dc.contributor.advisor하동수-
dc.contributor.authorCho, Injoon-
dc.date.accessioned2023-06-26T19:32:06Z-
dc.date.available2023-06-26T19:32:06Z-
dc.date.issued2023-
dc.identifier.urihttp://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=1033003&flag=dissertationen_US
dc.identifier.urihttp://hdl.handle.net/10203/309644-
dc.description학위논문(석사) - 한국과학기술원 : 조천식모빌리티대학원, 2023.2,[iii, 22 p. :]-
dc.description.abstractAs the demand for autonomous driving increases, it is paramount to ensure safety. Accident prediction using deep learning methods for safety in autonomous driving has recently received a lot of attention. In this paper, we propose an accident anticipation method using reinforcement learning. The accident anticipation platform used in this paper learns accident prediction and fixation point prediction by using dashcam video as input. For the first time in this accident prediction platform, the double actors with regularized critics (DARC) method are used. DARC is a state-of-the-art reinforcement learning model that uses two actors in a continuous action space and is applied to the accident anticipation platform to obtain 5\% faster predictions than existing algorithms. Furthermore, in this work, for the first time, a multi-actor multi-critic method is proposed. The proposed multi-actor multi-critic method allows stable learning by regularizing the critics while increasing the agent's exploration with a multi-actor structure. The proposed method is used for accident anticipation and the optimal number of actor-critic pairs is found. Experimental results demonstrate that the algorithm performs best among reinforcement learning methods, and suggest the possibility that it can be used on other platforms in the future.-
dc.languageeng-
dc.publisher한국과학기술원-
dc.subjectAccident anticipation▼aDeep learning▼aReinforcement learning-
dc.subject사고 예측▼a심층 학습▼a강화 학습-
dc.titleMulti-actor multi-critic method for traffic accident anticipation-
dc.title.alternative교통사고 예측을 위한 멀티 액터 멀티 크리틱 방법론-
dc.typeThesis(Master)-
dc.identifier.CNRN325007-
dc.description.department한국과학기술원 :조천식모빌리티대학원,-
dc.contributor.alternativeauthor조인준-
Appears in Collection
GT-Theses_Master(석사논문)
Files in This Item
There are no files associated with this item.

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0