IOP Conference Series: Earth and Environmental Science

PAPER • OPEN ACCESS An overview of path planning algorithms

To cite this article: Zhuozhen Tang and Hongzhong Ma 2021 IOP Conf. Ser.: Earth Environ. Sci. 804 022024

View the article online for updates and enhancements.

This content was downloaded from IP address 170.106.35.76 on 23/09/2021 at 23:53

6th International Conference on Energy Science and Applied Technology IOP Publishing IOP Conf. Series: Earth and Environmental Science 804 (2021) 022024 doi:10.1088/1755-1315/804/2/022024

An overview of path planning algorithms

Zhuozhen Tang *, Hongzhong Ma Hohai university, Nanjing 211100, Jiangsu, China

*Corresponding author Email: [email protected]

Abstract. This paper reviews the basic concepts of path planning, classifies environmental modeling methods, analyzes the significance of V2X environment modeling, and summarizes the existing path planning algorithms. Different algorithm can be adjusted in time according to different environments to improve the efficiency of path planning. In addition, according to the advantages and disadvantages of different algorithms, each algorithm is fused, which can effectively avoid the shortcomings of each algorithm and improve the efficiency of the planning algorithm.

Keywords: environmental modelling; V2X environmental; path planning algorithm.

1. Introduction The path planning part belongs to the control or decision-making part of the unmanned vehicle and vessel architecture system (Fig. 1), and it is also a key subject to be solved urgently in the unmanned vehicle technology. The performance of path planning module is directly related to the advantages and disadvantages of vehicle driving path selection and driving fluency. With the wide application of intelligent vehicles, ships and mobile robots in different fields, more effective path planning algorithms have been proposed by researchers.

environment

sensors V2X actuators

Perceptions Planning Control Environment Target Action Model Figure 1. Vehicle and vessel architecture system

2. Basic concepts of path planning At present, the concept of path planning that is relatively recognized in the academic field is as follows: According to different degrees of understanding of environmental information, there are two kinds of global path planning, that environmental information is fully known and local path planning where

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. Published under licence by IOP Publishing Ltd 1

6th International Conference on Energy Science and Applied Technology IOP Publishing IOP Conf. Series: Earth and Environmental Science 804 (2021) 022024 doi:10.1088/1755-1315/804/2/022024

environmental information is partially unknown or completely unknown. According to whether the obstacle is moving in the environment, it can be divided into static planning with static environment unchanged and dynamic planning with obstacle movement. According to whether the number of controllable variables in the mobile robot system is less than the dimension of its attitude space, it can be divided into the motion planning of the non-holonomic system, the motion planning of the holonomic system, and the path planning of the kinematics and dynamics system considering both kinematics and dynamics constraints. [6] Global path planning, also known as global navigation planning. Pure geometric path planning from the starting point to the target point, independent of time series and vehicle dynamics; Local path planning is also called obstacle avoidance planning, dynamic path planning, or real-time navigation planning. Mainly to detect obstacles, and the Moving trajectory Tracking of obstacles (Moving Object Detection and Tracking, usually abbreviated MODAT) to make the next step may position, finally draw a picture contains collision risk existing and potential collision risk map of obstacles, the potential risk warning level is 100 milliseconds, the future need to be further improved, the sensor, the algorithm of computing power efficiency and the processor is a great challenge, consider not only the obstacle avoidance planning space is also considering time series, in downtown complex computation. It can be more than 30TFLOPS, which is the most difficult part of the smart car.

3. Environment Modelling To carry out path planning, first, a moving environment model of the mobile machine should be established. With the environment model (different from the understanding of the environment), path planning is constituted by the path search. Environment representations can be divided into static and dynamic environments. The static environment is relatively simple, but in practical application, it is more dynamic environment that changes in real time. In the complex unknown environment, the ground unmanned vehicle usually lacks the global environmental information, but the V2X environmental model can provide rich environmental information, which also makes the vehicle trajectory obtained by the initial value and constraints more. This also puts forward high real-time and robust requirements for path planning algorithm in complex environment. Therefore, it is of great significance to study the path planning algorithm of flexible strain in complex environment. At present, the existing environmental modeling methods are mainly responsible for collecting the road and lane information around the self-driving vehicle through the road map subsystem, and expressing it in the map with geometric and topological characteristics, including interconnection and restriction. The main content of road mapping subsystem is map representation and map creation. The simplest way to create a road map is to extract a manual annotation of the road shape from an aerial image. For dynamic environment, MOT subsystem (also known as Multi-Target Detection and Tracking -DATMO) is mainly used to detect and track the posture of moving obstacles in the environment around the autonomous vehicle.

3.1. Traditional environmental modeling methods The traditional MOT method mainly includes three steps: data segmentation, data association and filtering. In the data segmentation stage, clustering or pattern recognition technology is used to segment the sensor data. In the data association step, data association techniques are used to associate data segments with targets (moving obstacles). In the filtering phase, for each target, the position is estimated by taking the geometric mean of the data assigned to the target. The position estimate is usually updated by Kalman filter or particle filter.

3.2. Model-based approach Model-based approaches infer directly from sensor data, using a physical model of the sensor and a geometric model of the object, and using non-parametric filters (such as particle filters). The data

2

6th International Conference on Energy Science and Applied Technology IOP Publishing IOP Conf. Series: Earth and Environmental Science 804 (2021) 022024 doi:10.1088/1755-1315/804/2/022024

segmentation and association steps are not required because the geometric object model associates the data to the target.

3.3. Based on stereo vision method The method relies on the color and depth information provided by stereo image pair to detect and track the moving obstacles in the environment. Ess et al. proposed an obstacle detection and identification method using only synchronous video from a forward-looking stereo camera. Their work focuses on obstacle tracking based on the output of pedestrian and vehicle detectors per frame. For obstacle detection, they used a support vector machine (SVM) classifier with directional gradient histogram (HOG) feature to classify each image area as obstacle or non-obstacle.

3.4. Based on raster map method Based on the method of raster map, the occupancy raster map of dynamic environment is first constructed. The map-building step is followed by data segmentation, data association, and filtering steps to provide an object-level representation of the scene. Nguyen et al. proposed a grid-based method for detecting and tracking moving objects in stereo cameras. Their work focuses on pedestrian detection and tracking. Reconstruction of 3D points from stereo image pairs. Using the inverse sensor model, the occupancy probability of each cell in the grid map is estimated based on the related three-dimensional points.

3.5. Based on sensor fusion method Sensor fusion based approaches fuse data from various sensors, such as lidar, radar, and cameras, to explore their respective characteristics and improve environmental awareness. The MOT subsystem is divided into two layers. The sensor layer extracts feature from the sensor data that can be used to describe the moving obstacle hypothesis according to the point model or box model. The sensor layer also attempts to associate features with current prediction assumptions from the fusion layer. Functions that cannot be associated with existing assumptions are used to generate new recommendations. Generate observations for each feature associated with a given hypothesis, encapsulating all the information needed to update the hypothesis state estimate. The fusion layer selects the best tracking model for each hypothesis based on the recommendations and observations provided by the sensor layer, and uses the Kalman filter to estimate (or update) the estimate of the hypothesis state.

3.6. Based on deep learning The method based on deep learning uses deep neural network to detect the position and geometric features of moving obstacles, and tracks their future state based on the current camera data.

3.7. V2X-based environment modeling method V2X (Vehicle to X) is the key technology of the future intelligent transportation system. It enables communication between cars, between cars and base stations, and between base stations. In this way, a series of traffic information such as real-time road conditions, road information and pedestrian information can be obtained to improve driving safety, reduce congestion and improve traffic efficiency. V2X, or Vehicle-to-Everything, is one of the supporting technologies for intelligent vehicles and intelligent transportation. As shown in Figure 2, V2X includes various application communication scenarios such as vehicle-to-vehicle V2V(vehicle-to-vehicle), vehicle-to-infrastructure V2I (vehicle-to- infrastructure), vehicle-to-pedestrian V2P(vehicle-to-pedestrian), and vehicle-to-external Network V2N (vehicle-to-network).At present, based on V2V communication vehicles can realize forward collision warning, lane change assistance, left turn assistance, cooperative adaptive cruise control, etc., based on V2I communication vehicles can realize speed suggestion, traffic priority, road condition warning, red light running warning, current weather influence warning, parking space and charging piles location search and other applications..

3

6th International Conference on Energy Science and Applied Technology IOP Publishing IOP Conf. Series: Earth and Environmental Science 804 (2021) 022024 doi:10.1088/1755-1315/804/2/022024

4. Algorithm Concept Global path planning algorithm can be roughly divided into three categories: the first kind of the traditional path planning algorithm based on map traditional algorithm (Dijkstra algorithm, A * algorithm, etc.), the second category of intelligent path planning algorithm based on bionics (PSO algorithm, genetic algorithm and reinforcement learning, etc.), the third type of path planning algorithm based on sampling (RRT) [32].

4.1. Traditional algorithm of map-based path planning In the traditional path planning algorithms, the implementation principle and application scope of various algorithms are very different, but the following five algorithms can be regarded as A class (Dijkstra, A*, D*, LPA*, D* Lite). The basic principles of each algorithm are described below, and the search principle and application scenarios are compared and distinguished. The salient feature of map- based path planning is that it is necessary to build the environment model before the algorithm can be carried out.

4.1.1. Dijkstra algorithm. Dijkstra algorithm was proposed by E.W.Dijkstra in 1959. The algorithm traverses the most subnodes one by one through the greedy principle, and then uses the relaxation method to optimize the path selection. Finally, the optimal path is stored in the readable list, so as to solve the optimal path planning problem. Dijkstra algorithm has a better planning result when the map data volume level is small, which can meet the requirements. However, when the map data volume level is large, the planning result is poor, which cannot meet the requirements of planning. In the calculation process of shortest path, in order to improve Dijkstra algorithm, priority queue and reverse N-tree were combined to achieve a downgratable priority queue [35]. Zhang Yin et al. proposed selection search within a limited range, which effectively reduced the search scope and search times of the algorithm, and greatly improved the search efficiency of the algorithm [36].

4.1.2. A * algorithm. A* algorithm is a heuristic search algorithm, as an improvement of Dijkstra algorithm. In the process of Dijkstra algorithm search, heuristic function is added. It can be said that A* algorithm is the most representative among heuristic search algorithms [37,38]. Heuristic search refers to the establishment of heuristic search rules in the search process, in order to measure the distance relationship between the real-time search location and the target location, so that the search direction is the direction of the target location first, and finally achieve the effect of improving the search efficiency. The basic idea of A* algorithm is as follows: the estimation function f(x) of the current node x is introduced, and the estimation function of the current node x is defined as: f (x)= g(x)+h(x) Where g(x) is the actual distance from the starting point to the current node x (which can be replaced by the distance between two points in the code); H (x) is the minimum distance estimated from node x to the end point. The form of H (x) can be selected from either the Euclidean distance or the Manhattan distance. A* algorithm mainly adds the evaluation function. Compared with Dijkstra algorithm, the advantage of A* algorithm lies in the introduction of heuristic function when exploring the next node. The search times of the algorithm are greatly reduced and the search speed of the algorithm is further improved. Therefore, A* algorithm is more efficient and has been widely used. When the distance from the target point is relatively close, the search range of the algorithm is narrowed, so that the target point can be more easily searched when approaching the target [39].Xu Tangjian et al proposed the bidirectional A* algorithm to search simultaneously from two directions to improve the efficiency of the traditional A* algorithm [40]. Wei Shan etal. proposed a design method of heuristic function to improve A* algorithm, and adopted the weighted heuristic function to select the appropriate weighted Manhattan distance weight, effectively improving the efficiency of path search [41].

4

6th International Conference on Energy Science and Applied Technology IOP Publishing IOP Conf. Series: Earth and Environmental Science 804 (2021) 022024 doi:10.1088/1755-1315/804/2/022024

4.1.3. D * algorithm. Based on the A* algorithm, Anthony Stentz proposed the Dynamic A* algorithm, namely the D* algorithm, in 1994. D* algorithm is a reverse incremental search algorithm. Reverse means that the algorithm searches gradually from the target point to the starting point. Incremental search, the algorithm calculates each node in the process of searching the distance metric information H (x), in a dynamic environment if appear obstacles cannot continue along the path search in advance, algorithm based on each point has been the original distance measure information at the current state of path planning again, need not to planning from the target point. Wherein, the distance measurement information H(x)=H(y)+C(y, x), where H(y) represents the distance measurement from point x to the target point, and C(y, x) represents the distance measurement from point y to point x, which can be replaced by the actual distance between two points in the algorithm.

4.1.4. LPA * algorithm. Life Planning A* algorithm, jointly proposed by Sven Koenig and Maxim Likhachev in 2001, is an incremental heuristic search algorithm based on A* algorithm. LPA* algorithm implementation principle: The starting point of search is set as the starting point (forward search), and the size of the Key value is taken as the principle of search progress. The planning is completed when the target point is the next search point. The Key value contains the heuristic function H term as the heuristic principle to influence the search direction; In a dynamic environment, LPA* can adapt to the change of obstacles in the environment without recalculating the entire environment by using the G value obtained from the previous search twice during the current search to re-plan the path. Where Key[] is a two-dimensional array:

G (n) represents the distance metric from the starting point to the current point. RHS (n) for min (g (n) + c (n, n), n 'n's parent, h (n, goal) as inspiration; The search principle is as follows: judge the size of K1 first; if K1 is small, then traverse first; if K1 = K2, then select the point with smaller K2.

4.1.5. D * Lite algorithm. D* Lite algorithm is a path planning algorithm proposed by Koenig S and Likhachev M based on LPA_STAR algorithm. The main difference between D* Lite and LPA* is the search direction, which replaces the target point goal in the Key definition with the corresponding information for the starting point start. The D*_lite algorithm first searches backward in a given map set and finds an optimal path. In the process of approaching the target point, it deals with the emergence of dynamic obstacle points by searching in the local scope. The advantage of the incremental algorithm is that the path search of each point has been completed, and when the obstacle point cannot continue to approach according to the original path, the data of incremental search can be reused to directly replan an optimal path at the current position of the obstruction, and then continue to move forward.

4.1.6. Comparison of traditional algorithms based on maps. Table 1 shows the comparison of the five algorithms in search direction, heuristic, incremental, applicable scope and practical application. The efficiency and scope of application of the above five algorithms are different. High efficiency and a wide range of problem solving do not necessarily mean a wide range of applications. At present, the actual applied algorithms should try their best to give play to the expertise of one algorithm and constantly optimize the algorithm performance on the basis of the basic functions.

5

6th International Conference on Energy Science and Applied Technology IOP Publishing IOP Conf. Series: Earth and Environmental Science 804 (2021) 022024 doi:10.1088/1755-1315/804/2/022024

Table 1. Comparison of performance and application of traditional algorithms Search Heuristic Incremental Scope of application direction Dijkstra F F positive Global information is known, static planning A* T F positive Global information is known, static planning D* F T reverse Some information is known, Dynamic programming Part of the information is known, assuming the rest of the LPA* F T positive free path, dynamic programming Part of the information is known, assuming the rest of the D*lite T T reverse free path, dynamic programming

1) Pros and cons of search direction: When the global map information is known in a static environment, either forward search or reverse search can play an effective role. However, in the dynamic environment, facing the unknown map, to get the shortest path needs to keep trying, forward search is easy to produce the phenomenon of running counter to the optimal path, and the reverse search algorithm can deal with this situation well. Reverse search combined with incremental search enables D* Lite algorithm to constantly update the optimal path from the current point to the target point in the dynamic obstacle diagram by using the node distance information generated in the previous iteration. In the forward search, the incremental algorithm can only provide the distance information from the current point to the starting point and the heuristic estimation information to the target point, but cannot guarantee the accessibility of the unsearched area. 2) Heuristic and non-heuristic: Heuristic algorithm can guide the search direction to the target point in each search, replacing the limitation of non-heuristic algorithm to traverse the periphery without rules. Under normal circumstances, it can greatly improve the search efficiency. But when the heuristic path is blocked, the search effect will be counterproductive 3) Heuristic and Incremental: Heuristic search is to use heuristic function to guide search, so as to achieve efficient search, heuristic search is A kind of "intelligent" search, typical algorithms such as A* algorithm, genetic algorithm, etc. Incremental search is the reuse of previous search result information to achieve efficient search, greatly reducing the search scope and time, typical such as LPA*, D* Lite algorithm, etc. The positive and negative search direction is related to whether it can deal with dynamic programming. Heuristic search brings the improvement of time efficiency, avoid global blind search; Incremental search represents the secondary use of iterative information and is often used to improve the efficiency of the algorithm.

4.2. Path planning algorithm based on bionics With the continuous in-depth study of the nature, human beings have designed different bionics algorithms based on bionics, mainly including neural network, ant colony algorithm, Wolf colony algorithm, heritage algorithm, etc. [41].

4.2.1. Genetic algorithm. Genetic Algorithm (GA) is a method to search for the optimal solution by simulating the natural evolution process. As a computational model of biological evolution process that simulates natural selection and Genetic mechanism of Darwin's biological evolution. The main characteristic is that the structure object is operated directly without the limitation of derivation and function continuity. It has inherent implicit parallelism and better global optimization ability. Using the probabilistic optimization method, the optimal search space can be obtained and guided automatically without definite rules, and the search direction can be adjusted adaptively.

6

6th International Conference on Energy Science and Applied Technology IOP Publishing IOP Conf. Series: Earth and Environmental Science 804 (2021) 022024 doi:10.1088/1755-1315/804/2/022024

4.2.2. Neural network algorithm. Artificial Neural Network (ANN) has been a hot topic in the field of Artificial intelligence since the 1980s.It abstracts the neural network of human brain from the perspective of information processing, establishes some simple model, and forms different networks according to different connection modes. In engineering and academic circles, it is often referred to as neural network or quasi-neural network. Neural network is an operation model, which is composed of a large number of nodes (or neurons) connected with each other. Each node represents a specific output function, known as the activation function. Each connection between two nodes represents a weighted value for the signal that passes through the connection, called the weight, which is equivalent to the memory of an artificial neural network.

4.2.3. Ant Colony Algorithm. In 1991, Italian scientist Marco Dorigo et al. first drew inspiration from ant colony foraging and proposed ant colony algorithm [43-45]. Ant colony algorithm is a common intelligent algorithm at present, but the basic ant colony algorithm also has its inherent shortcomings, because of the number of iterations, slow convergence speed, easy to fall into local optimal. Zhao Hongcai et al. used fuzzy algorithm to improve ant colony algorithm, such as the number of iterations and the slow convergence rate, and proposed to combine ant colony algorithm and fuzzy algorithm to optimize the trajectory planning and improve the efficiency of ant colony algorithm [43-45]. Fei Xiaofang etal. proposed that based on ant colony algorithm, a new theory was adopted to update pheromone concentration to improve the search speed of the algorithm and monitor all pheromone concentrations at the same time, effectively avoiding the stagnation phenomenon of the algorithm in the search process [47]. Qu Xiaokang etal. combined ant colony algorithm with genetic algorithm, and obtained the optimal path of this iteration through selection crossover mutation of genetic algorithm [48].

4.3. Sampling-Based Path Planning Algorithm. Sampling-based path planning algorithms have been used in vehicle path planning for a long time. Probabilistic Road Map (PRM) and Rapidly exploring Random Tree (RRT) are the most common sampling-based algorithms.

4.3.1. The PRM algorithm. Probability graph algorithm is to randomly select N nodes in the planned space, and then connect each node, and remove the connection with the obstacle, so as to get a feasible path. Obviously, when the sampling points are too few or the distribution is unreasonable, the PRM algorithm is incomplete. However, sampling points can be added to make the algorithm complete, so the PRM is probabilistically complete but not optimal.

4.3.2. RRT algorithm. Rapidly exploring Random Tree (RRT) was proposed by Lavalle [49-51]. Its advantage is that it does not need to model the planned space. It has a high coverage rate in the search space and a wide search range, so it can explore the unknown area as much as possible. But at the same time, there is also the problem that the calculation cost of algorithm is too high. Researchers have proposed various modifications of RRT to solve such problems. For example, Goal-Bias RRT algorithm, Bi-RRT algorithm, RRT-Connect algorithm, Extend RRT algorithm, Local-Tree-RRT algorithm, Dynamic RRT algorithm and so on. The goal-bias algorithm takes the target node as the sampling point, and the probability of the target point can be controlled in the algorithm. The Dynamic RRT algorithm proposes pruning and merging operations to remove invalid nodes before continuing the search [52-54]. Nowadays, although a variety of path planning methods for robots emerge in an endless stream, most of the path planning algorithms still stay in the static environment and ignore the complexity of the dynamic environment. The path planning algorithms for the dynamic environment are not mature enough, and the problems considered in the research also have some deficiencies. Of course, the research on real-time planning in dynamic environment is also a focus of research because of its challenging and practical application to technology and life.

7

6th International Conference on Energy Science and Applied Technology IOP Publishing IOP Conf. Series: Earth and Environmental Science 804 (2021) 022024 doi:10.1088/1755-1315/804/2/022024

5. Conclusion As an important part of unmanned driving technology, path planning has been proposed by domestic and foreign researchers in order to improve the quality and speed of path planning.This paper reviews the existing path planning algorithms. Different algorithms have their own advantages and disadvantages. Therefore, for the advantages and disadvantages of different algorithms, each algorithm can be fused in the later stage.In addition, the algorithm can be adjusted in time according to different environments to improve the efficiency of path planning.

References [1] LIU Zuofeng, ZHANG Jinyou. Development Status and Trend of Driverless Vehicle [J]. Agricultural Machinery, 2020, (01): 56. [2] LI Deyi. Future Traffic: Autonomous Driving and Intelligent Network [J]. Robot Industry, 2019, (06): 48-54. [3] PENG Xiaoyan, XIE Hao, Jing. Research on Local Path Planning Algorithm for Unpilotless Vehicle [J]. Automotive Engineering, 2020, 42(01): 1-10. [4] Journal of Tongji University (Natural Science Edition), 2019, 47(12): 1785-1790. (in Chinese with English abstract) [4] CHEN J Y, LI R B, XING X Y, et al. A review on intelligent evaluation of autonomous vehicles [J]. [5] QU C Z, Gai Q W, Zhang J, Zhong M Y. Novel Hybrid Grey Wolf Optimizer for Unmanned Aerial Vehicle (UAV) Path Planning [J]. Knowledge Based Systems,2020,194. [6] ZHU Linfeng, YANG Jiafu, SHI Yangyang, et al. Research progress on lateral control strategy of unmanned vehicle [J]. World Science and Technology Research and Development, 2018, 40(05): 506-518. [7] ZHOU Jingjing, SU Zhiyuan, WANG Rendong. Development Status and Enlightenment of Foreign Military Unmanned Ground Vehicle [J]. Journal of Military Transportation Institute, 2018, 20(05): 39-44. [8] PAN Fuquan, QI Rongjie, ZHANG Xuan, et al. Research Overview and Development Prospect of Driverless Vehicle [J]. Science and Technology Innovation and Application, 2017, (02): 27-28. [9] XIE R L, MENG Z J, ZHOU Y M, et al. Heuristic Q-learning based on experience replay for three-dimensional path planning of the unmanned aerial vehicle. 2020, 103(1) [10] ZHAO Yang. Development Status and Trend of Military Unmanned Ground Platform [J]. Commercial Culture (First Half), 2011, (05): 310. [11] FAZLOLLAHTABAR H, HASSANLI S. Hybrid cost and time path planning for multiple autonomous guided vehicles [J]. Applied Intelligence, 2017, 48 (1):482-498. [12] SONG J, HAO C, SU J C. Path planning for unmanned surface vehicle based on predictive artificial potential field. 2020, 17(2) [13] WU Junta. Research on Unmanned Driving Strategy Based on Integrated Multi-Depth Deterministic Strategy Ggradient [D]. Shenzhen: University of Chinese Academy of Sciences (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences), 2019. [14] XU Z, ZHANG E,CHEN Q W. Rotary unmanned aerial vehicles path planning in rough terrain based on multi-objective particle swarm optimization[J].Journal of Systems Engineering andElectronics, 2020, 31 (01) : 130-141. [15] MA Teng. Research on Hierarchical Coordinated Control Method of Unpiloted Vehicle Trajectal Tracking [D]. Liaoning: Dalian University of Technology, 2017. [16] SONG Feiyang. Driverless Vehicle and Its Development [J]. China High-tech, 2019, (05): 24-27. [17] DONG Bin. Research on the Function and Morphology Design of Automobile Led by Internet Thinking [D]. : Beijing Institute of Technology, 2015. [18] CHEN Yanshou. Research Trend of Driverless Vehicle [J]. Automobile & Accessories, 2017, (06): 44-48. [19] HE Jia, RONG Hui, WANG Wenyang, et al. The Development Review of Baidu Google

8

6th International Conference on Energy Science and Applied Technology IOP Publishing IOP Conf. Series: Earth and Environmental Science 804 (2021) 022024 doi:10.1088/1755-1315/804/2/022024

Autonomous Vehicle [J]. Automotive Electrical Appliance, 2017, (12): 19-21. [20] DU Kaiyue. Research on Road Information Sensing Technology of Unpilotless Vehicle [D]. Jilin: Changchun University of Science and Technology, 2018. [21] XUE Z B, LIU J C, WU Z G,et al. Development and path planning of a novel unmanned surface vehicle system and its application to exploitation of Qarhan Salt Lake[J].Science China(InformationSciences), 2019 on conversion (8) : 195-197. [22] JIN Yueqiang. Path Planning Model of Shortest Time for Obstacle Avoidance of Robot in Static Field [J]. Machine Tool & Hydraulics, 2018, 46(15): 88-94. [23] QI Zhi. Research on Lane Changing and Overtaking Control Method of Undriverless Vehicle [D]. Hebei: Yanshan University, 2017. [24] XU Yang, LU Liping, CHU Duanfeng, et al. Unified Modeling Method for Intelligent Vehicle Trajectional Planning and Tracking Control [J]. Acta Automatica Sinica, 2019, 45(04): 799- 807. [25] Zhuoping, LI Yishan, XIONG Lu. Review of motion planning algorithms for intelligent vehicles [J]. Journal of Tongji University (Natural Science Edition), 2017, 45(08): 1150-1159. [26] WANG J Q, WU J, YANG L. The driving safety field based on driver vehicle road interactions [C]. //IEEE Transactions on Intelligent Transportation Systems, 2015, 16(4):2203-2214. [27] CAO H T, SONG X L,HUANG Z Y. Simulation research on emergency path planning of an active collision avoidance system combined with longitudinal control for an autonomous vehicle [J]. Proceedings of the Institution ofMechanical Engineers, Part D: Journal of Automobile Engineering,2016, 230(12): 1624−1654. [28] JI J, KHAJEPOUR A,MELEK W W. Path Planning and Tracking for Vehicle Collision Avoidance Based on Model Predictive Control with Multiconstraints [C]. //IEEE Transactions on Vehicular Technology,2017, 66 (2) : 952-964 [29] ABBAS M A, MILMAN R,EKLUND J M. Obstacle avoidance in real time with Nonlinear Model Predictive Control of autonomous vehicles [J]. Canadian Journal of Electrical and Computer Engineering, 2017, 40(1): [30] Aconcave M, Heshmati Z, Aconcave M, et al. Autonomous vehicle convex optimization at length [C]. //In:Proceedings of the 21st Iranian Conference on Electrical Engineering (ICEE). Mashhad, Iran: IEEE, 2014.1−7. [31] Huang Z C, Wu Q, Ma J, Fan S Q. Vehicular communication technology in vehicular communication systems using APF and MPC [J]. Journal of Vehicular Technology, 2016,89(2016):232-242. [32] Yanjie Li,Wu Wei,Yong Gao,Dongliang Wang, Hun Fan. A new algorithm for mobile robots based on multi-path optimization [J]. IEEE Transactions on Robotics and Automation, 2015, 22 (2) : 123-135. [33] WANG X, WANG X, WANG X, et al. A Study on the Path Selection Method Based on Dynamic Trip Time Reliability [J]. Transportation Research Part B, 2014, 44 (1): 69-74. [34] WANG Z G, WANG X H, LI Y. An improved algorithm based on Dijkstra shortest path algorithm [J]. Journal of Inner Mongolia Normal University (Natural Science Edition), 2012, 41(02): 195-200. [35] LI Ruimin, LI Ruimin, LI Maolin, et al. An improved Dijkstra algorithm for route selection for avoiding satellite detection [J]. [36] LV Zhigang, LI Lin, YU Wenchaopeng, et al. Research on path planning of heuristic search algorithm [J]. Foreign Electronic Measurement Technology, 2018, 37(06): 16-21. [37] IEEE Transactions on Intelligent Transportation Systems, 2016, 26(11): 153-156. (in Chinese with English abstract) [38] WANG J, WANG J B, ZHANG X, et al. Journal of Intelligent Transportation Systems, 2014, 23 (1): 1-10. [38] Wang W, Wang D, Feng Z. An improved A~* algorithm for shortest path planning of mobile robots [J]. Journal of Computer Applications, 2018, 38(05): 1523-1526. [39] LI X, LI X, LI X, et al. A new method for the optimization of A novel path finding algorithm [J].

9

6th International Conference on Energy Science and Applied Technology IOP Publishing IOP Conf. Series: Earth and Environmental Science 804 (2021) 022024 doi:10.1088/1755-1315/804/2/022024

IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, (1): 63-74. [40] QIAO Huifen. Research on Robot Path Planning Algorithm [D]. : North University of China, 2015. [41] Zhuge Chengchen, Zhu Gengchen, Zhu Gengchen, Wang Xiaofeng, Wang Xiaofeng. Research on Vehicle Path Planning in Complex Environment [D] [42] DOROGO M, CAGO G D. The modified swarm optimization metaheuristic [M] //COME D, MDORIGO F, Editors. New Ideas in Optimization. Mc London, UK; Graw Hill,1999; 11-32. [43] He Yu, Zhang Zhian, Han Mingming, et al. Ant Colony Algorithm Based on Omnidirectional Mobile Robot Path Planning [J]. Journal of Test Technology, 2018, 32(05): 374-380. [44] LI L, LI H, YAN B, et al. Path planning with multi-heuristic factor improved ant colony algorithm [J]. Computer Engineering and Applications, 2019, 55(05): 219-225. [45] ZHAO Hongcai, GUO Jiale, XU Xiaojian, et al. Research on trajectory planning of mobile robot based on fuzzy ant colony algorithm [J]. Computer Simulation, 2018, 35(05): 318-321. [46] IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 23 (4): 417-423. [47] KOMAZ M, KOMAZ M, KOMAZ M, et al. A new method for the estimation of the performance of an adaptive adaptive ant colony algorithm [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014, 23 (4): 417-423. [47] [QU X K, RUI X P, HAN Y, et al.Improved ant colony algorithm for distance calculation of grid cost [J]. Journal of Geo-information Science, 2016, 18(08): 1052-1059.] [48] LAVALLE S M. Rapidly-Exploring Random Trees: A New Tool for Path Planning [R]. Ames, USA: Computer Science Department Iowa State University, 1998. [49] KUFFNER J J , LAVALLE S M. RRT-connect: An efficient approach to single-query path planning [C]. //IEEE international Conference on Robotics and Automation. Piscataway, USA :IEEE, 2000:995-1001. [50] KARAMAN S, WALTER M R, PEREZ A, et al. Anytime Motion Planning using the RRT [C]. //IEEE International Conference on Robotics and Automation. Piscataway, USA: IEEE, 2011: 1478-1484. [51] GAMMELL J D,SRINIVASA S S,Informed RRT*: Optimal path planning focused via direct sampling of an admissible ellipsoidal heuristic [C]//Proceedings of the 2014 IEEE/RSJInternational Conference on Intelligent Robots and Systems. Piscataway, NJ: IEEE, 2014:2997-3004. [52] DOSHI A, POSTULA A J, FLETCHER A, et al. Development of micro-UAV with integrated motion planning for open-cut mining surveillance [J]. Microprocessors and Microsystems, 2015, 33 (8) 6:829-835. [53] Journal of Northwest University (Natural Science Edition), 2018, 48(05): 651-658. (in Chinese with English abstract) [54] Shi K, Denny J, Amato N M. Spark PRM:Using RRTs within PRMs to efficiently explore narrow passages[C]//2014 IEEE International Conference on Robotics and Automation, ICRA 2014. Piscataway: IEEE, 2014:4659-4666.

10