自駕車在挑戰性道路條件下的神經網絡微調之研究｜國立臺灣科技大學博碩士論文系統

簡易檢索 / 詳目顯示

回結果列表

研究生：	李謙熙 Qian-Xi Li
論文名稱：	自駕車在挑戰性道路條件下的神經網絡微調之研究 Fine-Tuning Neural Networks for Self-Driving Cars in Challenging Road Conditions
指導教授：	陳郁堂 Yie-Tarng Chen
口試委員:	阮聖彰 Shanq-Jang Ruan 方文賢 Wen-Hsien Fang 賴坤財 Kuen-Tsair Lay 林銘波 Ming-Bo Lin 陳郁堂 Yie-Tarng Chen
學位類別：	碩士 Master
系所名稱：	電資學院 - 電子工程系 Department of Electronic and Computer Engineering
論文出版年：	2024
畢業學年度：	112
語文別：	英文
論文頁數：	32
中文關鍵詞：	自動駕駛車輛、Openpilot 、Supercombo 、適配器、CARLA 、模擬器、多任務學習、分階段訓練
外文關鍵詞：	Autonomous vehicles, Openpilot, Supercombo, Adapter, CARLA, Simulator, Multi-task learning, phase training
相關次數：	點閱：68 下載：4
分享至:	分享至facebook 分享至twitter

查詢本校圖書館目錄查詢臺灣博碩士論文知識加值系統勘誤回報

上一筆

增強自動駕駛汽車在多樣化、高曲率和複雜道路條件下的導航仍然是一項重大挑戰。這個問題至關重要，因為自動駕駛汽車安全有效地透過這些複雜場景的能力直接影響其可靠性和廣泛採用的潛力。在這項研究中，我們透過 Carla-Roach 框架使用從 CARLA LeaderBoard 收集的數據對 Openpilot 的 Supercombo 模型進行微調來應對這一挑戰。我們的方法採用適配器、多任務學習和分階段訓練來優化模型對車道線、領先車輛和駕駛計劃的預測。我們的解決方案在多個領域取得了顯著的改進。適配器顯著減少了所需的微調資料量。多任務學習增強了軌跡和引導車輛的預測，而分階段訓練則在所有任務中提供增量增益，特別是在車道線預測準確性方面。這展示了該模型在各種具有挑戰性的駕駛場景中的穩健性和適應性。

Enhancing the navigation of self-driving cars in diverse, high-curvature, and complex road conditions remains a significant challenge. This issue is crucial because the capability of self-driving cars to safely and effectively maneuver through these intricate scenarios directly impacts their reliability and potential for widespread adoption. In this research, we address this challenge by fine-tuning the Supercombo model from Openpilot using data collected from the CARLA LeaderBoard via the Carla-Roach framework. Our approach employs adapters, multi-task learning, and phased training to optimize the model’s predictions of lane lines, leading vehicle, and driving plans. Our solution achieves notable improvements in several areas. Adapters significantly reduce the amount of fine-tuning data needed. Multi-task learning enhances trajectory and lead vehicle predictions, while phased training provides incremental gains across all tasks, particularly in lane line prediction accuracy. This showcases the model's robustness and adaptability in various challenging driving scenarios.

摘要 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
Acknowledgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Table of contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1 Parameter-efficient Transfer Learning . . . . . . . . . . . . . . . . . . 3
2 Structural Re-parameterization . . . . . . . . . . . . . . . . . . . . . 4
3 RepAdapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4 Supercombo with RepAdapter . . . . . . . . . . . . . . . . . . . . . . 6
Proposed Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1 Preliminary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1 Basic Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2 Key Outputs of Supercombo . . . . . . . . . . . . . . . . . . . 8
1.3 Structure of Comma2k19 . . . . . . . . . . . . . . . . . . . . . 9
2 Data collection from Carla-Roach . . . . . . . . . . . . . . . . . . . . 10
2.1 Lane lines data collection . . . . . . . . . . . . . . . . . . . . . 12
2.2 Plans and Leads data collection . . . . . . . . . . . . . . . . . 12
3 Lane lines Data Pre-processing . . . . . . . . . . . . . . . . . . . . . 14
4 Loss function and Multi-task Training . . . . . . . . . . . . . . . . . . 15
4.1 Loss function of Lane lines and Plans . . . . . . . . . . . . . . 15
4.2 Multi-task Learning . . . . . . . . . . . . . . . . . . . . . . . . 16
5 Phase Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1 Introduction of Evaluation metrics . . . . . . . . . . . . . . . . . . . 18
1.1 Lane lines and Plans . . . . . . . . . . . . . . . . . . . . . . . 18
1.2 Leads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2 Ablation study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1 First part of experiment . . . . . . . . . . . . . . . . . . . . . 19
2.2 Second part of experiment . . . . . . . . . . . . . . . . . . . . 21
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
                                

[1] H. Schafer, E. Santana, A. Haden, and R. Biasini, “A commute in data: The comma2k19 dataset,” 2018.
[2] Glutamat42, “Carla-lane-detection-dataset-generation,” 2021.
[3] L. Chen, T. Tang, Z. Cai, Y. Li, P. Wu, H. Li, J. Shi, J. Yan, and Y. Qiao, “Level 2 autonomous driving on a single device: Diving into the devils of openpilot,” 2022.
[4] A. Kendall, Y. Gal, and R. Cipolla, “Multi-task learning using uncertainty to weigh losses for scene geometry and semantics,” 2018.
[5] G. Luo, M. Huang, Y. Zhou, X. Sun, G. Jiang, Z. Wang, and R. Ji, “Towards efficient visual adaption via structural re-parameterization,” 2023.
[6] D. Vu, B. Ngo, and H. Phan, “Hybridnets: End-to-end perception network,” 2022.
[7] A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y. Lo, P. Dollr, and R. Girshick, “Segment anything,” 2023.
[8] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” 2023.
[9] A. Garcia-Garcia, S. Orts-Escolano, S. Oprea, V. Villena-Martinez, and J. Garcia-Rodriguez, “A review on deep learning techniques applied to semantic segmentation,” 2017.
[10] D. A. Roberts, S. Yaida, and B. Hanin, The Principles of Deep Learning Theory: An Effective Theory Approach to Understanding Neural Networks. Cambridge University Press, May 2022.

簡易檢索 / 詳目顯示

相關論文