智能駕駛泊車路徑規劃算法研究綜述

【摘要】為了提高泊車路徑規劃的效率和質量，對智能駕駛泊車路徑規劃算法進行研究。通過梳理近年來國內外學術研究的泊車規劃方法，根據規劃算法特性將泊車規劃算法分為圖搜索算法、空間采樣算法、智能算法、曲線插值算法和最優控制算法。分析5類算法的優缺點，并探討融合算法在特定環境下的應用，結合規劃求解效率和規劃路徑曲率等指標，對現有方案的合理有效性進行評價，并對泊車路徑規劃算法的未來發展進行展望。研究表明：（1）圖搜索算法具備全局路徑最優、實時性較好等優點，但存在路徑連續性差、高維空間復雜度高的問題；（2）空間采樣算法具有概率完備和高維空間搜索效率高的優勢，但內存消耗大、隨機性大且路徑曲率無法保證；（3）智能算法利用樣本學習能力強、迭代性強，但訓練成本高、動態適應性差、實時性差；（4）基于優化的曲線插值算法路徑易計算，但無法保證曲率的連續性；（5）基于優化的最優控制算法能夠處理復雜優化問題，但時效性無法保證且易陷入局部最小值。融合算法通過優勢互補，能夠更好地適應車輛自身約束和環境約束，實現高效、安全的泊車路徑規劃。

關鍵詞：智能駕駛；泊車路徑規劃；規劃效率；路徑曲率

中圖分類號：U471.15" "文獻標志碼：A" DOI： 10.19822/j.cnki.1671-6329.20230244

Overview on Research on Path Planning Algorithms for Intelligent Driving Parking

Wang Yanyan， Zha Yunfei

（Fujian University of Technology， Fuzhou 350118）

【Abstract】 In order to improve the efficiency and quality of parking path planning， a study on intelligent driving parking path planning algorithms has been conducted. By reviewing the parking planning methods from recent domestic and international academic research， based on their characteristics the parking path planning algorithms are categorized into 5 types： graph search algorithms， sampling-based algorithms， intelligent algorithms， curve interpolation algorithms， and optimal control algorithms. The advantages and disadvantages of these 5 types of algorithms are analyzed， and the application of fusion algorithms in specific environments is explored. The rationality and effectiveness of existing solutions are evaluated in terms of planning efficiency and path curvature. Furthermore， future development trends in parking path planning algorithms are discussed. The study concludes the following： （1） Graph search algorithms offer the advantages of globally optimal paths and good real-time performance but suffer from poor path continuity and high complexity in high-dimensional spaces. （2） Sampling-based algorithms have the advantages of probabilistic completeness and high search efficiency in high-dimensional spaces but have large memory consumption， significant randomness， and cannot guarantee path curvature. （3） Intelligent algorithms have strong learning capabilities based on samples and strong iteration capabilities but have high training costs， poor dynamic adaptability， and poor real-time performance. （4） Curve interpolation algorithms based on optimization are easy to calculate but cannot guarantee the continuity of curvature. （5） Optimal control algorithms based on optimization can handle complex optimization problems but cannot guarantee timeliness and are prone to falling into local minima. Fusion algorithms through complementary advantages can better adapt to vehicle constraints and environmental constraints， achieving efficient and safe parking path planning.

Key words： Intelligent driving， Parking path planning， Planning efficiency， Path curvature

0 引言

隨著汽車保有量大幅上升，“泊車難”成為汽車用戶出行的痛點之一。智能泊車技術可以協助駕駛員完成泊車操作、降低泊車難度、提升泊車安全性和舒適性[1]?！?br>