- 學位論文
基於最佳化回授雙向快速搜尋隨機樹*與停車導向模型預測控制之人性化自駕車停車運動規劃與車輛控制
Human-like Motion Planning and Vehicle Control for Autonomous Parking Using Optimize-feedback Bidirectional Rapidly-exploring Random Trees* and Parking-oriented Model Predictive Control
摘要
時至今日，自駕車技術致力於為日常生活提供便捷。然而，現存先進駕駛輔助系統 (ADAS)的應用只能被歸類在自駕等級1至4，而非全自動等級5。基於危險與環境的考量，自動停車可能成為第一個自駕車的全自動應用。大部分自駕停車所需要的技術可從現有技術中選擇，例如：感測結合、SLAM和周圍感知等。另一方面，某些技術應該依據停車場景與考量來做開發，例如：停車格探測、運動規劃與車輛控制等。 在本篇論文中，有兩項用於自駕停車技術的關鍵方法，分別是運動規劃法與車輛控制器： 就運動規劃而言，一個基於隨機取樣的運動規劃法被提出用來從任何可行的起始點有效率地產生仿人類且免於碰撞的停車路徑於日常生活中的停車情境。此外，被提出的方法其目的是為要解決基於隨機取樣的運動規劃法的共同缺點，以維持每次執行結果的高度路徑品質與一致性。 就車輛控制而言，一個停車導向的模型預測控制器被提出來同時控制方向盤與速度以達到精確且順暢的停車路徑追蹤。更進一步來說，被提出的車輛控制器致力於解決實作問題，例如：車輛考量、即時控制與訊號延遲。因此，實作表現上得以接近在理想狀態下的表現。 為了證實兩項本篇所提出方法的效果，本篇論文中的模擬與實驗分別被執行在常見與嚴苛的停車情境中，例如：垂直停車、平行停車、斜向停車與U字型迴轉。模擬結果不僅驗證了本篇研究中所提出兩項方法中技術元素的效果，也展示處理多樣停車情境的能力。最重要的是，實車實驗充分展現了該兩項方法是可被確實實踐在日常生活中的，而非僅僅是理論上的模擬。
並列摘要
Nowadays, techniques of autonomous vehicles are devoted to providing convenience for people in everyday life. However, existing applications of Advanced Driver Assistance Systems (ADAS) can just be categorized into autonomous driving level 1-4 but not full automation in level 5. With the considerations of risks and environments, autonomous parking may be the first fully autonomous applications for autonomous vehicles. Most required techniques for autonomous parking can be chosen from existing ones, like sensor fusion, SLAM, and surrounding perception. On the other hand, some required techniques should be developed based on the parking scenarios and considerations, such as parking slot detection, motion planning, and vehicle control. In this thesis, there are two key methods proposed for the autonomous parking technique, that is, a motion planner and a vehicle controller: For motion planning, a sampling based motion planner is proposed to generate a human-like and collision-free parking path efficiently with any feasible start and goal configurations for parking scenarios in everyday life. Moreover, the proposed method aims to deal with the common defects of sampling based motion planners to maintain high path quality and consistency for each execution. For vehicle control, a parking-oriented model predictive controller is proposed to control steering and speed simultaneously for accurate and smooth parking path tracking. Furthermore, the proposed vehicle controller is dedicated to working around the practical problems, such as vehicle considerations, real-time control, and signal delay. As a result, the practical performance can approach the performance in ideal condition. To verify the effects of both proposed methods, the simulations and experiments in the thesis are conducted in common and strict parking scenarios, such as perpendicular parking, parallel parking, angle parking, and U-turn. The simulation results not only verify the effects of each technical element in both proposed methods, but also show the ability to deal with the various parking scenarios. The most important of all, the on-car experiments sufficiently demonstrate that both proposed methods can be actually implemented in everyday life instead of theoretical simulations.
參考文獻
延伸閱讀
- 許珮筠（2022）。結合序列學習與動作狀態估測技術應用於自駕車行駛周圍之即時物件軌跡預測〔碩士論文，國立臺灣師範大學〕。華藝線上圖書館。https://doi.org/10.6345/NTNU202200179
- 謝旭昇、夏晧清、葉光毅（2015）。考量社會互動下之機慢車停車地點選擇行為模式之建構。都市與計劃，42(3)，325-362。https://doi.org/10.6128/CP.42.3.325
- 程凱驛（2011）。基於視訊場景資料蒐集與訓練之自適應車流估計機制〔碩士論文，國立中央大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0031-1903201314425521
- 林敬智、李則霖、張峻豪、連豐力（2021）。自動停車路徑規劃與決策系統開發。機械工業雜誌，(457)，60-69。https://www.airitilibrary.com/Article/Detail?DocID=P20171221002-202104-202104060022-202104060022-60-69
- 吳明軒（2016）。Research on Modularized Control Structure Combined with Lane Keeping Control and Tire Anti-slip Strategy for an Automated Vehicle〔碩士論文，國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342/NTU201602891