國立虎尾科技大學 |

Language: English

Back to Search results for [ subject:"深度強化學習." ]

基於深度強化學習之Sim2Real車道維持的研究 = = Researc...

許聖杰

基於深度強化學習之Sim2Real車道維持的研究 = = Research on Sim to Real Lane Keeping Based on Deep Reinforcement Learning /

Record Type:	Language materials, printed : Monograph/item
Title/Author:	基於深度強化學習之Sim2Real車道維持的研究 =/ 許聖杰.
Reminder of title:	Research on Sim to Real Lane Keeping Based on Deep Reinforcement Learning /
remainder title:	Research on Sim to Real Lane Keeping Based on Deep Reinforcement Learning.
Author:	許聖杰
Published:	雲林縣 :國立虎尾科技大學 , : 民113.06.,
Description:	[8], 58面 :圖, 表 ; : 30公分.;
Notes:	指導教授: 鄭佳炘 , 黃國鼎.
Subject:	Hough Transform. -
Online resource:	電子資源

基於深度強化學習之Sim2Real車道維持的研究 = = Research on Sim to Real Lane Keeping Based on Deep Reinforcement Learning /
許聖杰

基於深度強化學習之Sim2Real車道維持的研究 =Research on Sim to Real Lane Keeping Based on Deep Reinforcement Learning /Research on Sim to Real Lane Keeping Based on Deep Reinforcement Learning.許聖杰. - 初版. - 雲林縣 :國立虎尾科技大學 ,民113.06. - [8], 58面 :圖, 表 ;30公分.

指導教授: 鄭佳炘 , 黃國鼎.

碩士論文--國立虎尾科技大學電機工程系碩士班.

含參考書目.

隨著時代的快速進步，自駕車相關研究逐漸受到熱烈關注，因此在汽車產業領域都大量投入於自駕車技術，這促使自駕車技術的快速演進，但在訓練時需大量數據資料，若完全投入在真實場域訓練需耗費龐大的時間及成本。為了解決以上問題，開發人員開始在電腦裡開始架設與真實場域環境相符的虛擬環境，自駕車通過虛擬的方式來訓練真實場域的行駛，由於目前真實環境的多樣化，虛擬環境難以模擬真實場域的各種變化，例如天氣、路況、光線等因素，這些差異已經成為提升自駕車整體性能和安全性的必要研究。本論文提出了一種將深度強化學習中的Dueling DQN與Cycle GAN相結合的方法，以實現在虛擬環境中進行Sim2Real車道維持訓練。此外，論文還引入了霍夫轉換車道辨識技術，以增強車道維持的穩定性。同時，為了減少環境光對模型的影響，論文還引入了自適應車燈技術，確保自駕車在各種光線條件下都能夠維持最佳亮度行駛。這些方法的目標是降低虛擬環境中訓練的模型轉移到真實環境時所帶來的差異，並進一步提高車道維持的穩定性。最後，本論文還建立了一個與虛擬環境相符的真實環境，用於驗證所提出的方法的可行性及穩定性。.

(平裝)Subjects--Topical Terms:

1451875
Hough Transform.

基於深度強化學習之Sim2Real車道維持的研究 = = Research on Sim to Real Lane Keeping Based on Deep Reinforcement Learning /
LDR:04227cam a2200241 i 4500 001 1129793
008 241015s2024 ch ak erm 000 0 chi d
035 $a (THES)112NYPI0441020
040 $a NFU $b chi $c NFU $e CCR
041 0 # $a chi $b chi $b eng
084 $a 008.165M $b 0814 113 $2 ncsclt
100 1 $a 許聖杰 $3 1448844
245 1 0 $a 基於深度強化學習之Sim2Real車道維持的研究 = $b Research on Sim to Real Lane Keeping Based on Deep Reinforcement Learning / $c 許聖杰.
246 1 1 $a Research on Sim to Real Lane Keeping Based on Deep Reinforcement Learning.
250 $a 初版.
260 # $a 雲林縣 : $b 國立虎尾科技大學 , $c 民113.06.
300 $a [8], 58面 : $b 圖, 表 ; $c 30公分.
500 $a 指導教授: 鄭佳炘 , 黃國鼎.
500 $a 學年度: 112.
502 $a 碩士論文--國立虎尾科技大學電機工程系碩士班.
504 $a 含參考書目.
520 3 $a 隨著時代的快速進步，自駕車相關研究逐漸受到熱烈關注，因此在汽車產業領域都大量投入於自駕車技術，這促使自駕車技術的快速演進，但在訓練時需大量數據資料，若完全投入在真實場域訓練需耗費龐大的時間及成本。為了解決以上問題，開發人員開始在電腦裡開始架設與真實場域環境相符的虛擬環境，自駕車通過虛擬的方式來訓練真實場域的行駛，由於目前真實環境的多樣化，虛擬環境難以模擬真實場域的各種變化，例如天氣、路況、光線等因素，這些差異已經成為提升自駕車整體性能和安全性的必要研究。本論文提出了一種將深度強化學習中的Dueling DQN與Cycle GAN相結合的方法，以實現在虛擬環境中進行Sim2Real車道維持訓練。此外，論文還引入了霍夫轉換車道辨識技術，以增強車道維持的穩定性。同時，為了減少環境光對模型的影響，論文還引入了自適應車燈技術，確保自駕車在各種光線條件下都能夠維持最佳亮度行駛。這些方法的目標是降低虛擬環境中訓練的模型轉移到真實環境時所帶來的差異，並進一步提高車道維持的穩定性。最後，本論文還建立了一個與虛擬環境相符的真實環境，用於驗證所提出的方法的可行性及穩定性。.
520 3 $a With the rapid advancement of technology, research related to autonomous vehicles has gradually attracted significant attention. Consequently, there has been a considerable investment in autonomous driving technology within the automotive industry, leading to the rapid evolution of autonomous vehicle technology. However, training requires a large amount of data, and conducting training solely in real-world environments would entail significant time and cost. To address these challenges, developers have begun to create virtual environments in computers that mimic real-world settings. Autonomous vehicles are trained in simulated environments to replicate real-world driving scenarios. However, due to the diversity of real-world environments, it is difficult for simulated environments to accurately mimic various factors such as weather conditions, road conditions, and lighting. These differences have become a necessary focus of research to enhance the overall performance and safety of autonomous vehicles. This paper proposes a method that combines Dueling DQN from deep reinforcement learning with Cycle GAN to achieve Sim2Real lane-keeping training in virtual environments. Additionally, the paper introduces the Hough transform lane detection technique to enhance the stability of lane-keeping. Furthermore, to mitigate the impact of environmental lighting on the model, the paper introduces adaptive headlights technology to ensure optimal brightness for autonomous driving under various lighting conditions. The goal of these methods is to reduce the disparities between models trained in simulated environments and their performance in real-world environments, thereby further improving the stability of lane-keeping. Finally, the paper establishes a real-world environment that matches the virtual environment for validating the feasibility and stability of the proposed methods..
563 $a (平裝)
650 # 4 $a Hough Transform. $3 1451875
650 # 4 $a Lane Keeping. $3 1451874
650 # 4 $a Sim2Real. $3 1451873
650 # 4 $a Deep Reinforcement Learning. $3 1346275
650 # 4 $a 霍夫轉換. $3 1420632
650 # 4 $a 車道維持. $3 1418221
650 # 4 $a 深度強化學習. $3 1346271
856 7 # $u https://handle.ncl.edu.tw/11296/nkh2gv $z 電子資源 $2 http