Artificial potential field method vehicle obstacle avoidance control method based on PSO optimization potential field parameters
By adjusting the key parameters of the artificial potential field method online using the particle swarm optimization algorithm, the problems of parameter sensitivity and local minima in the traditional method in dynamic environments are solved, and smooth, stable and energy-saving obstacle avoidance control of autonomous vehicles is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JILIN NORMAL UNIV
- Filing Date
- 2026-02-09
- Publication Date
- 2026-05-08
AI Technical Summary
Traditional artificial potential field methods are highly sensitive to parameters in autonomous driving, making it difficult to adapt to complex and ever-changing dynamic environments. This leads to path oscillations, planning failures, local minima, and target unreachability issues, especially with insufficient environmental adaptability under low-cost LiDAR perception conditions.
The Particle Swarm Optimization (PSO) algorithm is used to optimize key parameters in the Artificial Potential Field (APF) method online. By establishing multiple potential field models and designing fitness functions, the parameters are adaptively adjusted. Combined with the A* algorithm, global path planning and local rolling optimization are performed to improve obstacle avoidance performance.
It effectively overcomes local minima, improves path smoothness and driving stability, significantly reduces steering angle and path curvature fluctuations, shortens path length, and improves driving energy efficiency, making it suitable for autonomous driving applications under low-cost perception conditions.
Smart Images
Figure CN121994243A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of autonomous driving and intelligent vehicle motion planning technology. Background Technology
[0002] In autonomous driving systems, real-time and reliable local path planning and dynamic obstacle avoidance capabilities are core to achieving safe autonomous navigation. The artificial potential field method is widely used in real-time obstacle avoidance planning due to its advantages such as simple model, high computational efficiency, and ease of implementation. This method constructs a virtual potential field, guiding the vehicle's movement under the combined force of the attraction at the target point and the repulsion from obstacles.
[0003] However, the traditional artificial potential field method has obvious limitations: 1) High parameter sensitivity: Key parameters such as the potential field gain coefficient usually rely on empirical settings. Fixed parameters are difficult to adapt to complex and ever-changing dynamic environments, which can easily lead to path oscillation and planning failure; 2) Local minima problem: Vehicles are prone to getting stuck in local minima where attraction and repulsion are balanced and cannot reach the target; 3) Target unreachability problem: When obstacles are close to the target point, excessive repulsion may prevent the vehicle from reaching the target.
[0004] To overcome the aforementioned problems, researchers have proposed various improvement schemes, such as introducing virtual target points, combining velocity potential fields, and improving repulsion functions. While these methods have shown some effectiveness, their performance remains heavily dependent on parameter configuration and lacks adaptability to dynamic environmental changes. Particle swarm optimization (PSO), as a highly efficient swarm intelligence optimization algorithm, features fewer parameters, faster convergence, and suitability for online optimization, providing a new approach to solving the adaptive adjustment problem of APF parameters. However, existing research largely focuses on theoretical analysis and simulation verification. Deeply integrating PSO and APF to achieve real-time, robust online parameter optimization and obstacle avoidance control in dynamic environments under limited perception conditions still requires further exploration.
[0005] Especially in specific scenarios such as industrial parks and closed parks, vehicle perception systems are often limited by low-cost LiDAR, which has problems such as limited perception range and lack of semantic information, thus placing higher demands on the environmental adaptability and reliability of obstacle avoidance algorithms. Summary of the Invention
[0006] The purpose of this invention is to achieve online adaptive optimization of key APF parameters, thereby effectively overcoming local minima, improving path smoothness and driving stability under limited perception conditions, and realizing energy-saving driving through an artificial potential field method for vehicle obstacle avoidance control based on PSO optimized potential field parameters.
[0007] The steps of this invention are: S1. Establish a vehicle motion environment model that integrates multiple potential fields: S11, Target gravitational potential field: (1) In the formula, Gravitational gain coefficient; Indicates the vehicle's current location To sub-target Euclidean distance; To point to sub-target Unit direction vector; S12, Repulsive Potential Field of Obstacles: (2) In the formula, The obstacle repulsion gain coefficient; The minimum distance from the vehicle to the obstacle; Distance threshold affected by obstacles; This is the unit direction vector from the obstacle to the vehicle; The target adjustment factor; and The Euclidean distances for the current target adjustment and the previous target adjustment, respectively. ; To point to sub-target Unit direction vector; S13, Repulsive potential field at the road boundary: (3) In the formula, This is the boundary repulsion gain coefficient; For vehicle location To the nearest boundary The distance; The corresponding boundary repulsion is the negative gradient of the potential field: (4) In the formula, This is the unit direction vector pointing from the boundary to the vehicle; Represents the repulsive potential field function at the road boundary At vehicle location The gradient vector at that point; S14, Velocity potential field repulsive force: (5) In the formula, The gain coefficient of the velocity potential field repulsion force; This is the velocity vector of the vehicle relative to the dynamic obstacle. The corresponding velocity potential field repulsive force is: (6) In the formula, Represents the velocity repulsive potential field function At vehicle location The gradient vector at that point; S15, Resultant force on the vehicle The vector sum of all attractive and repulsive forces: (7) In the formula, This represents the number of obstacles detected at the current moment. For the first The repulsive force generated by the obstacle; S2. Constructing a PSO-APF fusion optimization strategy Encode the five key parameters in the APF into position vectors for PSO particles: (8) In the formula, This is the gravitational gain coefficient. The obstacle repulsion gain coefficient. The boundary repulsion gain coefficient. The velocity potential field repulsion gain coefficient, Distance threshold affected by obstacles; The position of each particle represents a set of APF parameters, and its velocity and position update formulas are as follows: (9) (10) In the formula, The particle velocity vector Let be the particle position vector. For inertial weights, , As a learning factor, , A uniformly random number within the range [0,1]; For particles In the number of iterations The velocity vector at time; For particles In the number of iterations The updated velocity vector at that time; For particles In the number of iterations Position vector at time; For particles In the number of iterations Position vector at time; For particles The individual's historical optimal position vector; The global optimal position vector of the group; Design a fitness function To comprehensively evaluate parameter performance, its function fusion path length , smoothness Safety and penalties for failure : (11) in, This is the path length weighting coefficient; This is the smoothness weighting coefficient; This is a safety weighting coefficient; This is the weighting coefficient for failure penalty; S3. Implement a global-local rolling optimization planning framework. 1) Global path planning stage: The A* algorithm is used to generate the optimal reference path from the starting point to the destination based on prior environmental information; 2) PSO online parameter optimization phase: The PSO optimizer is started within a local scrolling window, with the global path as a reference; 3) Local obstacle avoidance execution phase: Initialize the APF model using optimized parameters.
[0008] The advantages of this invention compared to the prior art are: 1. Strong adaptability: Through PSO online optimization, the APF parameters can dynamically adapt to different obstacle distributions, motion states and scene changes, overcoming the shortcomings of fixed traditional APF parameters; 2. Superior performance: Effectively solves the problems of local minima and target unreachability, and the planned path in dynamic environments is smoother and more stable, significantly reducing the fluctuation of steering angle and path curvature; 3. High performance: The algorithm has low dependence on sensors, making it particularly suitable for limited perception conditions such as LiDAR, and it is easy to integrate and deploy in real autonomous driving platforms (such as systems based on ROS and Autoware). 4. Potential: The optimized route reduces the average route length by about 5% while ensuring safety, achieving higher driving efficiency. Attached Figure Description
[0009] Figure 1 This is a schematic diagram of the PSO-APF algorithm control system architecture; Figure 2 This is a comparison diagram of vehicle movement paths using two different methods in a two-person pedestrian crossing scenario; Figure 3 Comparison of vehicle movement paths in a low-speed three-vehicle overtaking scenario; Figure 4 Comparison of overtaking trajectories of actual vehicles at medium speeds on campus. Detailed Implementation
[0010] This invention proposes a vehicle obstacle avoidance control method based on PSO-optimized potential field parameters (PSO-APF method), specifically involving a real-time vehicle obstacle avoidance control method based on the particle swarm optimization (PSO) algorithm to optimize key parameters of the artificial potential field (APF) online.
[0011] The technical solution adopted in this invention includes the following steps: I. The vehicle motion environment model integrating multiple potential fields is established as follows: First, based on the environmental point cloud information obtained by vehicle sensors (such as LiDAR), a comprehensive potential field model containing the following components is established: 1) Target gravitational potential field: guides the vehicle toward the sub-target point. (1) In the formula, Gravitational gain coefficient; Indicates the vehicle's current location To sub-target Euclidean distance; To point to sub-target Unit direction vector.
[0012] 2) Obstacle repulsive potential field: A piecewise function is used to prevent vehicles from colliding with static obstacles. (2) In the formula, The obstacle repulsion gain coefficient; The minimum distance from the vehicle to the obstacle; The radius of influence of the obstacle; This is the unit direction vector from the obstacle to the vehicle; The target adjustment factor (a constant with a value greater than 0, usually 2 or 3); and The Euclidean distances for the current target adjustment and the previous target adjustment, respectively. It is used to adjust the intensity of the target guiding effect in the repulsive field; To point to sub-target Unit direction vector; Used to ensure that when the vehicle arrives at the target point The repulsive force is zero, which solves the problem of the goal being unattainable.
[0013] 3) Road boundary repulsive potential field: Ensures vehicles travel within the feasible area. (3) In the formula, This is the boundary repulsion gain coefficient; For vehicle location To the nearest boundary The distance.
[0014] The corresponding boundary repulsion is the negative gradient of the potential field: (4) In the formula, This is the unit direction vector pointing from the boundary to the vehicle; Represents the repulsive potential field function at the road boundary At vehicle location The gradient vector at that point.
[0015] 4) Velocity potential field repulsion: Based on relative velocity, it prevents high-speed collisions with dynamic obstacles. (5) In the formula, The gain coefficient of the velocity potential field repulsion force; This is the velocity vector of the vehicle relative to the dynamic obstacle.
[0016] The corresponding velocity potential field repulsive force is: (6) In the formula, Represents the velocity repulsive potential field function At vehicle location The gradient vector at that point.
[0017] 5) The resultant force acting on the vehicle The vector sum of all attractive and repulsive forces: (7) in, This represents the number of obstacles detected at the current moment. For the first The repulsive force generated by the obstacle.
[0018] II. The present invention constructs a PSO-APF fusion optimization strategy, as detailed below: Encode the five key parameters in the APF into position vectors for PSO particles: (8) in, This is the gravitational gain coefficient. The obstacle repulsion gain coefficient. The boundary repulsion gain coefficient. The velocity potential field repulsion gain coefficient, The distance threshold is affected by obstacles.
[0019] The position of each particle represents a set of APF parameters, and its velocity and position update formulas are as follows: (9) (10) In the formula, The particle velocity vector Let be the particle position vector. For inertial weights, , As a learning factor, , A uniformly random number within the range [0,1]; For particles In the number of iterations The velocity vector at time; For particles In the number of iterations The updated velocity vector at that time; For particles In the number of iterations Position vector at time; For particles In the number of iterations Position vector at time; For particles The individual's historical optimal position vector; The global optimal position vector of the population. PSO tracks individual extreme values. and global extrema Achieve rapid convergence of the parameter space.
[0020] Design a fitness function To comprehensively evaluate parameter performance, its function fusion path length , smoothness Safety and penalties for failure : (11) in, This is the path length weighting coefficient; This is the smoothness weighting coefficient; This is a safety weighting coefficient; This is the failure penalty weighting coefficient. Smoothness. Defined as the sum of continuous changes in steering angle along the path, safety It is determined by the minimum distance between the vehicle and the obstacle.
[0021] The PSO algorithm iteratively updates the particle's velocity and position, tracking the individual's historical best position. and the global optimal position of the group The final output is the parameter combination that makes the fitness function optimal.
[0022] III. Implement the "global-local" rolling optimization planning framework. The specific design steps are as follows: This method adopts a hierarchical planning architecture, and the specific process includes three stages: 1) Global Path Planning Phase: Global planning algorithms such as A* are used to generate the optimal reference path from the starting point to the destination based on prior environmental information. The A* algorithm, or global path planning algorithm, is a heuristic search algorithm that evaluates a function... Seeking excellence, among which For the actual cost, To estimate the cost.
[0023] 2) PSO Online Parameter Optimization Phase: Using the global path as a reference, the PSO optimizer is launched within a local scrolling window. Utilizing current perception information, the local obstacle avoidance performance under different parameter combinations is simulated and evaluated, and the current optimal APF parameters are optimized and output in real time.
[0024] 3) Local obstacle avoidance execution phase: The APF model is initialized with optimized parameters. Within each control cycle, the APF calculates the resultant force based on real-time perceived obstacle information, determines the vehicle's next movement direction and adaptive step size (speed), achieving smooth and safe real-time obstacle avoidance. After the vehicle reaches the current sub-target, it updates the information and enters the next cycle until it reaches the final target.
[0025] To verify the effectiveness of the PSO-APF method proposed in this invention, system tests were conducted on a high-fidelity virtual simulation platform and in a real campus road environment. The software platform adopts a certain communication framework, integrates a certain autonomous driving software stack, and constructs a three-dimensional simulation environment containing dynamic pedestrians and vehicles in a certain simulation environment; the hardware platform is based on a real vehicle equipped with a 16-line LiDAR, a combined navigation system, and an on-board industrial control computer, and is tested on structured campus roads.
[0026] 1) In Gazebo simulations, a two-person obstacle avoidance scenario and a low-speed three-vehicle continuous overtaking scenario were selected for verification. Experimental results show that, compared with the traditional APF, the PSO-APF reduces the standard deviation of path curvature by 23.3% and the standard deviation of steering angle by 16.8% in the two-person crossing scenario; in the three-vehicle overtaking scenario, the standard deviation of steering angle is reduced by 22.2% and the standard deviation of path curvature is reduced by 21.1%, and the average path length is shortened by about 5%, demonstrating better smoothness, stability and traffic efficiency.
[0027] 2) In real-vehicle verification on campus, a medium-speed three-vehicle overtaking scenario test was conducted. Real-vehicle data showed that the trajectory planned by PSO-APF was significantly smoother, the standard deviation of the steering angle was reduced by 28.4%, the standard deviation of the path curvature was reduced by 44.1%, and the path length was shortened by 5.5%, achieving safe, smooth, and energy-efficient obstacle avoidance control in a real dynamic environment.
[0028] In summary, this invention effectively overcomes the problems of local minima and parameter sensitivity of traditional methods by optimizing APF parameters online through PSO, thereby improving the obstacle avoidance performance and driving efficiency of vehicles in dynamic environments with limited perception.
[0029] This invention discloses a vehicle obstacle avoidance control method based on particle swarm optimization of artificial potential field parameters, belonging to the field of autonomous driving and intelligent vehicle motion planning technology. Addressing the problems of traditional artificial potential field methods in terms of local minima, parameter sensitivity, and path oscillation, this invention proposes a real-time obstacle avoidance strategy that integrates particle swarm optimization and artificial potential field methods. This method optimizes five key parameters in the artificial potential field online using particle swarm optimization: the gravitational gain coefficient, obstacle repulsion gain coefficient, boundary repulsion gain coefficient, velocity potential field repulsion gain coefficient, and obstacle influence distance threshold. It also designs a fitness function that comprehensively considers path length, smoothness, and safety to achieve adaptive adjustment of the potential field parameters. Simulation and real-vehicle test results show that, compared with the traditional artificial potential field method, this invention can effectively avoid local minima, significantly improve path smoothness and driving stability, and achieve energy savings of over 5% in various dynamic scenarios, demonstrating good environmental adaptability and engineering practicality.
Claims
1. A vehicle obstacle avoidance control method based on PSO-optimized potential field parameters using an artificial potential field, characterized in that: The steps are as follows: S1. Establish a vehicle motion environment model that integrates multiple potential fields: S11, Target gravitational potential field: （1） In the formula, Gravitational gain coefficient; Indicates the vehicle's current location To sub-target Euclidean distance; To point to sub-target Unit direction vector; S12, Repulsive potential field of obstacle: （2） In the formula, The obstacle repulsion gain coefficient; The minimum distance from the vehicle to the obstacle; Distance threshold affected by obstacles; This is the unit direction vector from the obstacle to the vehicle; The target adjustment factor; and The Euclidean distances for the current target adjustment and the previous target adjustment, respectively. ; To point to sub-target Unit direction vector; S13, Repulsive potential field at the road boundary: （3） In the formula, This is the boundary repulsion gain coefficient; For vehicle location To the nearest boundary The distance; The corresponding boundary repulsion is the negative gradient of the potential field: （4） In the formula, This is the unit direction vector pointing from the boundary to the vehicle; Represents the repulsive potential field function at the road boundary At vehicle location The gradient vector at that point; S14, Velocity potential field repulsive force: （5） In the formula, The gain coefficient of the velocity potential field repulsion force; This is the velocity vector of the vehicle relative to the dynamic obstacle. The corresponding velocity potential field repulsive force is: （6） In the formula, Represents the velocity repulsive potential field function At vehicle location The gradient vector at that point; S15, Resultant force on the vehicle The vector sum of all attractive and repulsive forces: （7） In the formula, This represents the number of obstacles detected at the current moment. For the first The repulsive force generated by the obstacle; S2. Constructing a PSO-APF fusion optimization strategy Encode the five key parameters in the APF into position vectors for PSO particles: （8） In the formula, This is the gravitational gain coefficient. The obstacle repulsion gain coefficient. The boundary repulsion gain coefficient. The velocity potential field repulsion gain coefficient, Distance threshold affected by obstacles; The position of each particle represents a set of APF parameters, and its velocity and position update formulas are as follows: （9） （10） In the formula, The particle velocity vector Let be the particle position vector. For inertial weights, , As a learning factor, , A uniformly random number within the range [0,1]; For particles In the number of iterations The velocity vector at time; For particles In the number of iterations The updated velocity vector at that time; For particles In the number of iterations Position vector at time; For particles In the number of iterations Position vector at time; For particles The individual's historical optimal position vector; The global optimal position vector of the group; Design a fitness function To comprehensively evaluate parameter performance, its function fusion path length , smoothness Safety and penalties for failure : （11） in, This is the path length weighting coefficient; This is the smoothness weighting coefficient; This is a safety weighting coefficient; This is the weighting coefficient for failure penalty; S3. Implement a global-local rolling optimization planning framework. 1) Global path planning stage: The A* algorithm is used to generate the optimal reference path from the starting point to the destination based on prior environmental information; 2) PSO online parameter optimization phase: The PSO optimizer is started within a local scrolling window, with the global path as a reference; 3) Local obstacle avoidance execution phase: Initialize the APF model using optimized parameters.