Intelligent vehicle local obstacle avoidance path planning method and system, storage medium and software product

Abstract

The invention relates to the technical field of intelligent vehicle autonomous navigation, in particular to an intelligent vehicle local obstacle avoidance path planning method and system, a storage medium and a software product. The invention provides a dynamic scene-oriented improved IAPF (artificial potential field) local obstacle avoidance path planning method and a dynamic scene-oriented improved IAPF local obstacle avoidance path planning system which are strongly coordinated with a global path, a gravitation source is reconstructed into a next key point of the global path, 'distance to target 'modulation is introduced into a repulsive force potential, and radial / tangential component decomposition is performed; in combination with dynamic obstacle online prediction, kinematics constraint shaping of resultant force and a double-attraction-point loop strategy when no obstacle exists, the problems of local minimum and target inreachability of a traditional APF are fundamentally relieved, obstacle avoidance oscillation is reduced, the minimum safe distance is guaranteed, and the target-free obstacle avoidance performance is improved. Real-time, safe, smooth and executable local path planning of the moving body in a complex mixed environment is realized.

Description

Technical Field

[0001] This invention relates to the field of autonomous navigation technology for intelligent vehicles, and in particular to a method, system, storage medium, and software product for local obstacle avoidance path planning for intelligent vehicles. Background Technology

[0002] With the deployment of mobile robots, autonomous vehicles, and service robots in scenarios such as warehousing and logistics, park inspection, intelligent manufacturing, and special operations, local obstacle avoidance path planning has become a key link in ensuring system safety and efficiency. Typical application processes usually adopt a two-layer "global-local" architecture: offline or low-frequency updated global planners (such as A*, D*, RRT / RRT*, etc.) provide a reference path from the starting point to the destination; online, high-frequency executed local planners generate safe trajectories that can be followed in real time within a local area based on static / dynamic obstacle information perceived by sensors, and revert to the global path when necessary. The local planner needs to simultaneously satisfy three constraints: ① safety (maintaining minimum safe distance, avoiding collisions); ② feasibility (satisfying kinematic / dynamic constraints such as speed and turning angle); ③ stability and efficiency (smooth trajectory, no large oscillations, strong real-time performance). Among many local planning methods, the artificial potential field method, as an efficient local path planning method, has advantages such as fast response and ease of implementation, but it also suffers from problems such as path oscillation and susceptibility to local minima.

[0003] The traditional Artificial Potential Field (APF) method is a local path planning algorithm first proposed by Japanese scholar Khatib in 1986. This method is based on the potential field theory in physics, treating the agent as a moving object in a virtual environment. The target point and obstacles in the moving environment are considered as sources of different forces, and the robot is guided along a safe and efficient path by the synthesis of these forces.

[0004] like Figure 1 As shown, in the artificial potential field method, the target point passes through a gravitational potential field. Apply attraction to robots To guide the robot toward the target point, attraction It typically increases with the distance from the target point, and the obstacle passes through a repulsive potential field. Generate repulsive force To prevent the robot from colliding with obstacles, the repulsive force increases as the distance from the obstacle decreases, forcing the robot to move away from the obstacle.

[0005] The following are the more specific steps and details of the artificial potential field method:

[0006] (1) Initialization: Set the initial position of the intelligent vehicle Target point location Location of obstacles Gravitational coefficient repulsion coefficient Range of repulsive force from static obstacles Maximum number of iterations Parameters;

[0007] (2) Define the artificial potential field: Define a two-dimensional space based on the environment in which the intelligent vehicle is located, and set an attraction field in the space with the target point as the gravitational source. Setting up a repulsive field using obstacles as the repulsive force source ;

[0008] (3) Calculate the gravitational potential field and gravitational force: Determine the gravitational force based on the distance and direction between the current position of the intelligent vehicle and the target point. Size and direction are usually determined using a formula similar to Coulomb's law, with the attraction weakening as the distance increases.

[0009] (4) Calculate the repulsive potential field and repulsive force value: Determine the repulsive force based on the current position of the intelligent vehicle, the distance to each obstacle, and the range of influence of the obstacles. The magnitude and direction of the repulsive potential field are usually calculated using a formula that is inversely proportional to the distance; the closer the distance, the stronger the repulsive force.

[0010] (4) Calculate the resultant force: Add the force vectors in the gravitational field generated by the target point and the repulsive field generated by the obstacle to obtain the resultant force on the intelligent vehicle;

[0011] (5) Update the position of the intelligent vehicle: Update the position and speed of the intelligent vehicle according to the magnitude and direction of the resultant force;

[0012] (6) Determine the termination condition: if the intelligent vehicle reaches the target point or reaches the maximum number of iterations. The algorithm terminates.

[0013] (7) Iteration: If the termination condition is not met, return to step (3) until the termination condition is met.

[0014] Artificial potential field method is suitable for path planning in real-time dynamic environment due to its simple algorithm and high computational efficiency. However, the traditional artificial potential field method has certain limitations, which may cause the robot to get stuck in local minimum potential field in complex environment, resulting in stagnation or ineffective detour. Specifically, it manifests as: (1) the problem of unreachable target. In the traditional artificial potential field method, the intelligent vehicle is affected by both attraction and repulsion. When there is an obstacle near the target point, as the intelligent vehicle approaches the target, the attraction weakens and the repulsion strengthens, causing the vehicle to oscillate near the target point and fail to reach it stably; (2) the problem of getting stuck in local minimum. When the attraction and repulsion cancel each other out to form a local minimum point with zero resultant force, the intelligent vehicle will stagnate or wander after entering the point, resulting in the interruption of path planning. The above problems often occur when the intelligent vehicle, the target point and the obstacle are arranged linearly, or in non-convex obstacle environment, such as Figure 2 As shown. Summary of the Invention

[0015] The technical objective of this invention is to propose an improved artificial potential field (IAPF) local obstacle avoidance path planning method and system that is strongly coordinated with the global path and oriented towards dynamic scenarios. By reconstructing the gravity source as the "next key point of the global path", introducing the "distance to the target" modulation into the repulsive potential and performing radial / tangential component decomposition, combining online prediction of dynamic obstacles and kinematic constraint shaping of the resultant force, and the dual attraction point loop strategy when there are no obstacles, this invention fundamentally alleviates the local minima and target unreachability problems of traditional APF, reduces obstacle-around oscillations and ensures the minimum safe distance, and realizes real-time, safe, smooth and executable local path planning for moving bodies in complex mixed environments.

[0016] To achieve the above objectives, the present invention adopts the following technical solution:

[0017] A local obstacle avoidance path planning method for intelligent vehicles based on improved artificial potential field (IAPF) includes the following steps:

[0018] S1. Environment and parameter initialization: Construct a grid and / or continuous hybrid map, obtain the current position and / or attitude, global path, destination and obstacle status, and set the gravity / repulsion weight, detection radius, minimum safe distance, and heading and turning angle limits;

[0019] S2, Target Attraction Source Reconstruction: Replace the single endpoint with the global path "next key point" as the attraction source, set an arrival threshold ε, and switch to the subsequent key point when the distance is less than ε in order to suppress local minima and path oscillation;

[0020] S3. Improved repulsive force modeling: Within the radius of the obstacle, introduce a "distance to the endpoint" adjustment factor to construct repulsive potential energy, and calculate the negative gradient of the distance and tangential variables to obtain radial and / or tangential repulsive force components, so that the repulsive force and attraction force at the endpoint approach zero simultaneously.

[0021] S4. Dynamic obstacle prediction: Updates obstacle positions based on observation speed and recalculates distance matrix, refreshes repulsion force and set of possible actions;

[0022] S5. Resultant force solution and constraint shaping: Summing the gravitational force and each repulsive force vector to obtain the direction and magnitude of the resultant force, and combining kinematics and rotation constraints to limit the heading increment and speed amplitude, forming a one-step executable control quantity;

[0023] S6. Return to global path strategy: When there are no obstacles within the detection radius, construct a first attraction point and a forward second attraction point on the global path that are consistent with the current heading, and superimpose the attraction potential to limit the return angle and accelerate the fit to the global path.

[0024] S7. Path generation and key point switching: Update the pose based on the control variables and write the local path sequence; when the switching condition is met, iterate to the next key point until the destination.

[0025] S8. Termination and Safety Constraints: Termination occurs when the endpoint is reached or the maximum number of iterations is reached; maintain a minimum safe distance from obstacles throughout the process; if a violation occurs, a deceleration and / or detour / back-off mechanism will be triggered.

[0026] Preferably, the gravitational potential field function in step S2 is:

[0027] ,

[0028] in, : in position gravitational potential energy; Gravitational potential field adjustment factor; Current location To the target point The Euclidean distance; : The current position of the moving object; Target point location;

[0029] gravitational The negative gradient of the gravitational potential field function:

[0030] ,

[0031] Distance pairs The gradient, with direction from point to The unit vector.

[0032] As a preferred option, the improved repulsive potential field function in step S3 for:

[0033] ,

[0034] in, for The point is affected by the first The repulsive potential energy value generated by the obstacle As a regulating factor for the repulsive potential field, The Euclidean distance from the current location of the intelligent vehicle to the point affected by the repulsive force emanating from the obstacle. ,in, , As a distance adjustment factor, Let the radius be the repulsive potential field of the obstacle;

[0035] When the intelligent vehicle is within the repulsive potential field of the obstacle, the repulsive force... To each , Find the negative differential and obtain the improved result. Two components and Specifically, it is expressed as follows:

[0036] ,

[0037] ,

[0038] ,

[0039] in, Indicating intelligent vehicles The point is affected by the first The repulsive force value generated by each obstacle The direction of the vector is from the obstacle towards the intelligent vehicle. The direction of the vector is the direction with the smallest angle to the direction of movement of the intelligent vehicle when the detection area of ​​the intelligent vehicle is tangent to the obstacle.

[0040] As a preferred option, the resultant force in step S5 The calculation model is as follows:

[0041]

[0042] in, For smart cars The resultant force of the attractive and repulsive forces acting on a point. The attraction of intelligent vehicles to the target, Demonstrating intelligent vehicles The point is affected by the first The repulsive force value generated by each obstacle This represents the total number of repulsive forces experienced by the intelligent vehicle.

[0043] The potential field relationship is as follows:

[0044] ,

[0045] ,

[0046] ;

[0047] in, The gradient of the overall potential; , The gradient between the gravitational potential and the repulsive potential. Total force; , Attraction and total repulsion; Total resultant force; Attraction symbol; : Regarding the first The repulsive force symbol of an obstacle.

[0048] Preferably, step S6 introduces a heading decision angle smaller than the maximum turning angle of the intelligent vehicle. And set gravity points on two global paths. When the detection radius When there are no obstacles inside, the repulsive potential field generated by the obstacles is closed, and the intelligent vehicle is only subject to the attractive force from the target position;

[0049] First, draw an angle from the location of the intelligent vehicle to the global path. A ray, defined as the point where the ray intersects the global path. Secondly, based on this point, define the distance towards the target location. The point at is Third, definition and The new attractive potential field function for:

[0050] ,

[0051] in, , This represents the Euclidean distance between the location of the intelligent vehicle and the point of attraction, representing gravity. The negative gradient of the gravitational potential field function :

[0052] ,

[0053] Intelligent vehicle detection radius When there are no obstacles inside, it will be affected by the global path. The gravitational pull is proportional to the distance between the smart car and the gravitational point, and the smart car and... The distance between points is often greater than The point, plus the direction of the resultant force, always tends towards The dot points to the global path.

[0054] Preferably, the evaluation metrics of the method include at least one of the following metrics:

[0055] 1) Planning and obstacle avoidance success rate

[0056] The dynamic programming success rate metric measures the algorithm's planning and obstacle avoidance success rate as the proportion of times the algorithm successfully reaches the target point without colliding with obstacles out of the total number of tests. It is applicable to complex environments with both static and dynamic obstacles, and its calculation formula is as follows:

[0057] ,

[0058] in, The number of times the intelligent vehicle successfully avoids obstacles and completes path planning. This represents the total number of path planning iterations. To improve the success rate of obstacle avoidance.

[0059] 2) Path target value

[0060] 2.1) The target path length is: i = 2, 3, 4…m

[0061] ;

[0062] in: This is the total length; The Euclidean distance between adjacent waypoints; Number of waypoints;

[0063] 2.2) The path smoothness objective is: i=2,3,4…m

[0064] ,

[0065] ;

[0066] in: This represents the cumulative change in heading; For the first Section heading angle; The difference in heading between adjacent segments;

[0067] 2.3) Piecewise risk function is used for driving safety:

[0068] i=2,3,4…m,

[0069] ;

[0070] in: For the first Point of risk; The distance from the vehicle's center of gravity to the nearest obstacle's center of gravity; To achieve the minimum safety and maximum radius of influence;

[0071] 3) Total path time

[0072] The total path time starts from the starting point. After dynamic obstacle avoidance, the target point is reached. To calculate the total time, this paper often uses Euclidean distance to calculate and sum the distances between adjacent points:

[0073] ,

[0074] ,

[0075] in, This indicates that the intelligent vehicle has reached the previous node. At that moment, This indicates that the intelligent vehicle has reached the current node. At that moment, This represents the total time it takes for the intelligent vehicle to reach the current node after traversing the previous section. Represents starting point To the target point Total time required;

[0076] 4) Minimum distance from obstacles

[0077] The calculation formula is as follows:

[0078] ,

[0079] ,

[0080] in, Let i be the minimum distance between point i and the obstacle. Let be the distance matrix between point i and each obstacle.

[0081] Furthermore, the present invention also provides a local obstacle avoidance path planning system for intelligent vehicles based on an improved artificial potential field. This system is used to implement the method, including:

[0082] The perception and mapping module is used to obtain the grid environment containing static and / or dynamic obstacles, as well as the current position and attitude information of the intelligent vehicle;

[0083] The global path management module is used to provide the global path and output the "next key point" as the current gravity source point and its arrival determination threshold;

[0084] The gravity calculation module is used to obtain the gravity vector by taking the negative gradient of the current key point based on the modified gravitational potential field function.

[0085] The repulsion calculation module is used to calculate the negative gradient of each obstacle within the detection radius based on the repulsion potential energy function containing the "distance adjustment factor to the target point", and obtain the radial repulsion component and the tangential repulsion component.

[0086] The resultant force and trajectory update module is used to sum the gravitational and repulsive force vectors to obtain the total resultant force, and update the position of the intelligent vehicle, record the path point sequence and the change in heading angle accordingly;

[0087] The dynamic obstacle prediction module is used to update the position of dynamic obstacles in real time based on their speed and calculate the distance matrix between them and the intelligent vehicle to refresh the repulsive force.

[0088] Return to the global path strategy module, which is used to construct two attraction points on the global path and generate corresponding attraction potential fields based on the heading decision angle when there are no obstacles within the detection radius, so as to limit the return heading angle and guide the fast loop.

[0089] Preferably, the repulsion calculation module superimposes the obstacle repulsion effect from the detection boundary to improve safety redundancy; the global path management module automatically switches to the next key point on the global path until the destination when it detects that the current key point has been reached; the global path return strategy module uses the gravitational superposition of the two attraction points to make the resultant force direction deviate towards the global path and satisfy the vehicle's maximum turning angle constraint; the resultant force and trajectory update module is configured to support iteration termination conditions, including reaching the destination or reaching the maximum number of iterations.

[0090] Furthermore, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method.

[0091] Furthermore, the present invention also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the method.

[0092] This invention, by employing the aforementioned technical solution, achieves quantifiable comprehensive technical effects through integrated potential field modeling of "path-pointed gravitational source, target distance-modulated repulsive force, and radial / tangential decomposition," combined with dynamic obstacle online prediction, resultant force limiting and kinematic constraint shaping such as turning angle / angular velocity, and a dual-attraction point loop strategy in obstacle-free situations. These effects include: firstly, significantly reducing the probability of local minima and target unreachability, enabling the continuous generation of directional resultant forces and smooth convergence to the endpoint even in narrow channels, concave regions, and target-adjacent obstacle scenarios; and secondly, suppressing back-and-forth oscillations and sawtooth trajectories during obstacle avoidance, reducing instantaneous... The system features five key features: First, it improves trajectory smoothness and controllability by abruptly changing direction (curvature and directional acceleration are more controllable); second, it maintains a minimum safe distance margin for static / dynamic obstacles, initiating repulsion at the early detection radius to reduce critical approach and emergency braking; third, it quickly conforms to the global path after obstacle avoidance, shortening ineffective detours and backtracking time, reducing path redundancy and total travel distance; and fourth, it maintains computational complexity on par with the classic APF (approximately linear with respect to the number of obstacles) while ensuring real-time performance, thereby achieving stable, smooth, short-range, and safe local path planning in complex and dynamic mixed environments. Attached Figure Description

[0093] Figure 1 This is a diagram illustrating human labor.

[0094] Figure 2 This illustrates the unattainability of the objective and the local optimum.

[0095] Figure 3 For the correction of the gravitational source.

[0096] Figure 4 A schematic diagram of the repulsive force of an improved artificial potential field.

[0097] Figure 5 This is a strategy for returning the global path.

[0098] Figure 6 This is a map for local path planning.

[0099] Figure 7 This is a diagram of the artificial resultant potential field.

[0100] Figure 8 This is the path planning effect in a static environment with no unknown obstacles.

[0101] Figure 9 This describes the path planning performance in environments with static, unknown obstacles.

[0102] Figure 10 This describes the path planning performance in environments with unknown dynamic obstacles. Detailed Implementation

[0103] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0104] I. Modification and Improvement of the Artificial Potential Field Method

[0105] In traditional artificial potential field algorithms, the intelligent vehicle is influenced by the repulsive force of obstacles and the attractive force of the target point, avoiding obstacles along the path to reach the target point. However, during obstacle avoidance, the intelligent vehicle is affected by the repulsive forces of all obstacles within the repulsive force range, causing unnecessary steering. After obstacle avoidance, the intelligent vehicle deviates significantly from the initial course, but the gravitational force always points towards the target point, preventing the intelligent vehicle from returning to the initial path and causing it to get stuck in a local minimum state. To improve the path-keeping ability of the intelligent vehicle during obstacle avoidance and solve the local minimum problem of traditional artificial potential field methods, an improved algorithm based on the gravitational source and repulsive function is proposed.

[0106] 1.1 Modification and Improvement of the Artificial Potential Field Method

[0107] Traditional artificial potential field methods typically use a gravitational potential field originating from the target endpoint, which can easily lead to local minima in complex environments. Furthermore, the gravitational effect changes too rapidly with distance, affecting path smoothness and motion stability. This paper, in local dynamic obstacle avoidance, shifts the source of gravity from the global endpoint to the next critical point on the global path. This allows the intelligent vehicle to follow the gravity of global path points to reach the endpoint, avoiding local optima and path oscillations, and reducing non-optimal path selection caused by the gravity of a single target point. It also enhances adaptability in dynamic environments, enabling the intelligent vehicle to quickly return to the global path after obstacle avoidance, improving the global rationality and robustness of path planning. Finally, it eliminates abrupt deflections caused by the gravity at the endpoint, improving the smoothness and consistency of path following.

[0108] The gravitational potential field function in this application is:

[0109]

[0110] in, for The gravitational potential energy of a point. As a regulating factor for the gravitational potential field, The Euclidean distance between the current point of the intelligent vehicle and the target point. .gravitational The negative gradient of the gravitational potential field function :

[0111]

[0112] Figure 3 This is a revised schematic diagram of the gravitational source. , Represents the starting point and the target point. , Represents the intermediate point on the global path. The threshold for determining whether an intelligent vehicle has reached a node in the global path is the minimum value at which gravity exerts its effect. For example... Figure 3 As shown, when the location of the intelligent vehicle is close to the target point of the current path... The distance is less than When, it indicates that the intelligent vehicle has arrived. Near the point, No gravity is applied to the intelligent vehicle; at this point, the current target point is set as... Current location distance The position of the point is greater than , This will generate a gravitational force that attracts the smart car towards it. Proceed forward, iterating in sequence until the target point is reached.

[0113] 1.2 Improved IAPF repulsive potential field function

[0114] A repulsive potential field helps the robot stay away from obstacles. However, when using artificial potential fields for local collision avoidance in complex environments, the intelligent vehicle has a limited detection radius. To enhance the safety of local path planning, a feature is set up where the intelligent vehicle detects the detection range. When obstacles exert repulsive forces within the memory, this paper superimposes the repulsive force generated by the obstacle from the detection edge onto the intelligent vehicle. This paper introduces a distance adjustment factor from the current point to the target point into the repulsive potential field. This indicates that the repulsive and attractive forces only decrease to zero simultaneously when the intelligent vehicle reaches the target point, thus solving the problems of local minima and target unreachability in traditional artificial potential field methods. The improved repulsive potential field function... for:

[0115]

[0116] in, for The point is affected by the first The repulsive potential energy value generated by the obstacle As a regulating factor for the repulsive potential field, The Euclidean distance from the current location of the intelligent vehicle to the point affected by the repulsive force emanating from the obstacle. ,in, , As a distance adjustment factor, Let be the radius of the repulsive potential field of the obstacle. To address the local minima and target inaccessibility issues inherent in traditional artificial potential fields, when the intelligent vehicle is within the range of the obstacle's repulsive potential field, the repulsive force... To each , Find the negative differential and obtain the improved result. Two components and Specifically, it is expressed as follows:

[0117]

[0118]

[0119]

[0120] in, Indicating intelligent vehicles The point is affected by the first The repulsive force value generated by each obstacle The direction of the vector is from the obstacle towards the intelligent vehicle. The direction of the vector is the direction with the smallest angle to the direction of motion of the intelligent vehicle when the detection area of ​​the intelligent vehicle is tangent to the obstacle. Specifically, as shown below... Figure 4 As shown.

[0121] 1.3 Total Potential Function of IAPF

[0122] When a smart car is within the influence range of multiple obstacles in an artificial potential field, it will be subject to the gravitational pull from the target. Repulsive force from multiple obstacles Therefore, the net force it experiences for:

[0123]

[0124] in, For smart cars The resultant force of the attractive and repulsive forces acting on a point. The attraction of intelligent vehicles to the target, Demonstrating intelligent vehicles The point is affected by the first The repulsive force value generated by each obstacle This represents the total number of repulsive forces experienced by the intelligent vehicle.

[0125]

[0126]

[0127]

[0128] Table 1 shows the core pseudocode of the IAPF algorithm.

[0129]

[0130] 1.4 Return to Global Path Strategy

[0131] Local path planning can cause intelligent vehicles to deviate from the global path, but existing path planning algorithms rarely consider the angle at which the vehicle returns to the global path after obstacle avoidance. Therefore, this paper introduces a heading decision angle smaller than the vehicle's maximum turning angle. ( ), and set gravity points on two global paths. To solve the above problems.

[0132] When the detection radius When there are no obstacles inside, the repulsive potential field generated by the obstacles is closed, and the intelligent vehicle is only subject to the attractive force from the target location, such as... Figure 5 As shown, firstly, draw an angle θ from the location of the intelligent vehicle to the global path. A ray, defined as the point where the ray intersects the global path. Secondly, based on this point, define the distance towards the target location. The point at is Third, definition and The new attractive potential field function for:

[0133]

[0134] in, , This represents the Euclidean distance between the location of the intelligent vehicle and the point of attraction, representing gravity. The negative gradient of the gravitational potential field function :

[0135]

[0136] Intelligent vehicle detection radius When there are no obstacles inside, it will be affected by the global path. The gravitational pull is proportional to the distance between the smart car and the gravitational point, and the smart car and... The distance between points is often greater than The point, plus the direction of the resultant force, always tends towards The point points to the global path, thus enabling intelligent vehicles to... and The points between them can safely and quickly return to the global path, and the heading angle of the return is therefore limited, thus satisfying the kinematic constraints of the intelligent vehicle.

[0137] II. Simulation Experiments and Result Analysis

[0138] 2.1 Evaluation Indicators

[0139] In research on dynamic obstacle avoidance, planning and obstacle avoidance success rate, path target value, computation time, and distance to obstacles are key indicators for evaluating the reliability and effectiveness of algorithms, each providing insights into the capabilities and limitations of dynamic obstacle avoidance algorithms from different perspectives. Specifically, planning and obstacle avoidance success rate reflects the reliability and stability of dynamic obstacle avoidance algorithms in complex environments, path length reflects the overall optimality of the planned path, and computation time measures the algorithm's real-time performance and computational efficiency. By comprehensively utilizing these evaluation indicators, the performance of dynamic obstacle avoidance algorithms in different scenarios can be comprehensively and objectively assessed, providing a scientific basis for research on dynamic planning and real-time obstacle avoidance for intelligent vehicles.

[0140] (1) Planning and obstacle avoidance success rate

[0141] The dynamic programming success rate metric measures the stability and reliability of an algorithm under different environments. The planning and obstacle avoidance success rate is defined as the proportion of times the algorithm successfully reaches the target point without colliding with obstacles out of the total number of tests. It is applicable to complex environments with both static and dynamic obstacles. The calculation formula is as follows:

[0142]

[0143] in, The number of times the intelligent vehicle successfully avoids obstacles and completes path planning. This represents the total number of path planning iterations. To improve the success rate of obstacle avoidance.

[0144] (2) Path target value

[0145] Path length, safety, and smoothness are key performance indicators for path planning algorithms. In path planning tasks, the total path length reflects the algorithm's search efficiency; a shorter path indicates a stronger optimization capability in generating the path from the starting point to the destination. Path safety measures the distance relationship between the path and obstacles; higher safety indicates a lower collision risk during navigation. Path smoothness reflects the spatial continuity and feasibility of the path, significantly impacting the stability of intelligent vehicle operation and the feasibility of the path. The specific calculation formula is shown in the Chinese patent application (application number: application date:).

[0146] (3) Total path time

[0147] Total path time measures the total time required for an intelligent vehicle to travel from its starting point to its destination in an environment with dynamic obstacle interference. This metric directly reflects the algorithm's real-time performance and computational efficiency. A shorter total path time indicates that the algorithm can complete obstacle avoidance planning and return to the global path to reach the destination more quickly, thus improving the overall system efficiency. Total path time is calculated from the starting point... After dynamic obstacle avoidance, the target point is reached. To calculate the total time, this paper often uses Euclidean distance to calculate and sum the distances between adjacent points:

[0148]

[0149]

[0150] in, This indicates that the intelligent vehicle has reached the previous node. At that moment, This indicates that the intelligent vehicle has reached the current node. At that moment, This represents the total time it takes for the intelligent vehicle to reach the current node after traversing the previous section. Represents starting point To the target point Total time required.

[0151] (4) Minimum distance from the obstacle

[0152] Minimum obstacle distance is a crucial indicator for measuring path safety. If certain points on the path are too close to obstacles, the risk of collision increases. The minimum distance to obstacles is defined as the minimum distance from the current point on the path to each static and dynamic obstacle. A higher value indicates a greater distance between the intelligent vehicle and obstacles, resulting in a lower risk of collision. The calculation formula is as follows:

[0153]

[0154]

[0155] in, Let i be the minimum distance between point i and the obstacle. Let be the distance matrix between point i and each obstacle.

[0156] 2.2 Simulation Map and Configuration

[0157] The simulation variables for local path planning involve the motion and obstacle avoidance capabilities of the intelligent vehicle. To ensure the simulation environment is as objective and realistic as possible, this paper incorporates the principle of the artificial potential field method to specifically consider parameters such as the vehicle's size, steering angle, and obstacle detection range during the simulation process. Two maps, ranging from simple to complex, are used for simulation tests in both static and dynamic environments.

[99] The proposed enhanced artificial potential field method, the traditional artificial potential field method, and the literature are compared.

[100] The obstacle avoidance process of the proposed A*-DWA algorithm is compared to verify the performance and path optimization effect of the proposed algorithm. During the simulation, when the distance between the intelligent vehicle and the target point is less than or equal to 0.05m, it is considered to have reached the target point. Other relevant parameter settings are shown in Table 2.

[0158] Table 2 Local obstacle avoidance variables and their values ​​under different simulation maps

[0159]

[0160] 2.3 Simulation Experiment in a Static Environment Without Unknown Obstacles

[0161] Refer to the references

[99] Construct 20×20 and 30×30 grid maps to verify the path planning performance of the improved dynamic window method in a static environment. Set the starting coordinates of Map1 to (0.5, 19.5) and the target coordinates to (19.5, 0.5), and set the starting coordinates of Map2 to (0.5, 29.5) and the target coordinates to (29.5, 0.5). The starting positions are represented by blue circles, and the target positions by red pentagrams. The potential fields of Map1 and Map2 are shown below. Figure 7 As shown in (a~b).

[0162] Simulation results in a static environment without unknown obstacles, as shown below Figure 7 , Figure 8 As shown in Table 3. Figure 8 In the diagram, the dark blue line represents the actual path of the IAPF algorithm in this paper, the magenta line represents the actual path of the traditional APF, and the bright blue line represents the actual path of A*-DWA.

[0163] Table 3 shows the simulation performance results of traditional APF, A*-DWA and IAPF algorithms in the static environment without unknown obstacles in the Map1 and Map2 map environments.

[0164] Table 3. Simulation Results Analysis in a Static Environment Without Unknown Obstacles

[0165]

[0166] Depend on Figure 7 , Figure 8As shown in Table 3 and other simulation results of path planning in static environments without unknown obstacles, the following can be observed:

[0167] (1) The overall resultant potential field is highest at the starting point, lowest near the end point, and higher in the obstacle section. The constructed artificial potential field conforms to the planning principle of APF based on gradient descent.

[0168] (2) Both the enhanced IAPF and A*-DWA algorithms can effectively avoid local optima problems and successfully plan a collision-free path from the starting point to the target point. The traditional APF algorithm is not suitable for global planning in complex maps, but is more suitable for local path planning and global planning in simple map environments that do not contain door structure obstacles, U-shaped obstacles, etc.

[0169] (3) In terms of performance indicators, the IAPF algorithm decreased by an average of 7.758% in path smoothness compared to the A*-DWA algorithm, but performed better in path length and safety, with improvements of 0.566% and 14.601% respectively. In terms of planning efficiency, the total planning time of IAPF was 15.478% higher than that of A*-DWA based on the number of iterations. From the analysis of the minimum obstacle distance index, it can be seen that the path generated by the IAPF algorithm maintains a larger minimum distance between the path and the obstacle, effectively reducing the risk of collision, and verifying that IAPF is superior to the A*-DWA algorithm in terms of path safety.

[0170] In summary, the enhanced IAPF algorithm has higher path planning efficiency in static environments and exhibits superior performance in key metrics such as path length and safety.

[0171] 2.4 Simulation Experiment in an Environment with Static Unknown Obstacles

[0172] In real-world applications, unknown static obstacles may appear. To verify the algorithm's adaptability to unexpected situations, different numbers of unknown obstacles are added to different maps based on section 2.3. Due to the similarity between the paths planned by the IAPF and A*-DWA algorithms in Map1, this section adds two unknown static obstacles to Map1 based on the planned paths, located at (2.8, 14.0) and (14.1, 6.0). Figure 9 As shown by the magenta square in (a), two static unknown obstacles are added to the path planned by IAPF on Map2, with positions (15.1, 22.7) and (21.4, 11.7), as follows. Figure 9 As shown by the magenta square in (b), two static unknown obstacles are added to the path planned by DWA, with positions (11.5, 19.0) and (23.0, 10.0). Figure 9As shown in the golden square in (b), its influence range is consistent with the influence range of other static obstacles. The planning result is as follows. Figure 9 As shown in Table 4.

[0173] Table 4. Simulation Results Analysis in the Environment with Static Unknown Obstacles

[0174]

[0175] Simulation results in environments with static unknown obstacles, such as Figure 9 As shown in Table 4. Figure 9 In the diagram, the dark blue line represents the actual path of the IAPF algorithm in this paper, the magenta line represents the actual path of the traditional APF algorithm, and the bright blue line represents the actual path of the A*-DWA algorithm. The results show that:

[0176] (1) In static environments with unknown obstacles, traditional APF cannot complete the path planning task. IAPF and DWA can avoid static unknown obstacles when following the initial path. IAPF and DWA have some improvements over traditional algorithms.

[0177] (2) In terms of following effect, the DWA algorithm is guided and controlled by the target point and the initial path direction, and it moves along the initial path direction but has a small overlap with the initial path. The following is unstable. The IAPF algorithm is guided by the path point and returns to the original path more quickly after avoiding obstacles.

[0178] (3) Regarding the minimum distance to obstacles, both IAPF and A*-DWA are greater than the minimum safe distance, but IAPF has an average minimum distance that is 33.910% higher than A*-DWA, and the collision risk is lower.

[0179] In summary, the enhanced IAPF algorithm proposed in this paper can avoid unknown obstacles more efficiently and safely, and quickly return to the initial path after avoiding unknown static obstacles, with better path following and obstacle avoidance performance.

[0180] 2.5 Simulation Experiment in an Environment with Unknown Dynamic Obstacles

[0181] In real-world environments, non-static obstacles may appear. Due to the limitations of static environments, they cannot accurately simulate the actual conditions of real-world environments. To comprehensively enhance the obstacle avoidance performance of the IAPF algorithm in dynamic obstacle environments, this section sets dynamic obstacles on the original paths of IAPF and A*-DWA in the basic static map and performs simulations. Dynamic obstacles are represented by dark blue hexagonal stars, their influence range is represented by pink circles, and their trajectories are represented by yellow dashed lines.

[0182] In the 20×20 map Map1, set dynamic obstacle O1 on the initial DWA path with an initial position of (2.8, 14.0) and a speed of (-1.0 m / s, 0.0 m / s). On the initial IAPF path, set dynamic obstacle O2 with an initial position of (7.3, 8.0) and a speed of (0.0 m / s, 0.6 m / s). In the 30×30 map Map1, set dynamic obstacle O3 on the initial DWA path with an initial position of (21.0, 7.0) and a speed of (0.0 m / s, 0.6 m / s). On the initial IAPF path, set dynamic obstacle O4 with an initial position of (21.0, 17.0) and a speed of (0.0 m / s, 0.4 m / s). Since the traditional APF algorithm failed to complete the planning task in the two static environments mentioned above, the dynamic obstacle avoidance experiment still requires a collision-free path from the starting point to the target point. Therefore, the traditional APF algorithm will not be tested in the dynamic obstacle avoidance experiment in this section.

[0183] Table 5. Statistics of Dynamic Programming Indicators in the Presence of Dynamic Obstacles

[0184]

[0185] (1) At the core function level of dynamic obstacle avoidance, both the IAPF algorithm and the A*-DWA algorithm can enable the intelligent vehicle to avoid dynamic obstacles in real time while following the initial path, and return to the initial path to reach the target point after obstacle avoidance.

[0186] (2) Combination Figure 10 In the 20×20 simulation environment data (a~b), when the moving obstacle is in the turning area of ​​the path, both algorithms use a detour strategy to complete obstacle avoidance, but the detour path generated by the A*-DWA algorithm is longer. By comparing the results of different maps, the IAPF algorithm maintains stable path following performance in all scenarios, while the A*-DWA algorithm performs better in the complex Map2 environment than Map1, and its adaptability and stability to the environment are slightly inferior to the IAPF algorithm.

[0187] (3) In terms of performance indicators, in environments with dynamic obstacles, the average path length and smoothness index of the DWA algorithm are 1.204% and 6.766% lower than those of the IAPF algorithm, respectively. While shortening the path, the smoothness is improved. IAPF has good safety performance, and the planning time is shortened by 14.761% compared with the DWA algorithm, while the minimum distance to the obstacle is increased by 14.891%.

[0188] In summary, the traditional APF method, primarily focused on single-objective optimization, guides the agent to the endpoint solely through potential field descent, making it prone to getting trapped in local minima, path oscillations, and exhibiting poor adaptability to dynamic environments. The Enhanced Artificial Potential Field (IAPF) method, in its local planning, uses a global path reference point instead of the endpoint as the attraction source, improving the radial and tangential repulsion functions. This avoids local optima and path oscillations, eliminates abrupt deflections caused by endpoint attraction, and improves the smoothness of path following. Simulation results comparing the IAPF algorithm with the traditional APF and A*-DWA algorithms demonstrate that the improved IAPF algorithm reduces path search time and exhibits superior overall performance in dynamic environments.

[0189] In summary, this invention focuses on the application of the Enhanced Artificial Potential Field Method (IAPF) in local obstacle avoidance path planning. It systematically analyzes the basic principles of the traditional artificial potential field method and its limitations in practical applications, and constructs a more guiding and secure total potential field function, effectively solving the problem of traditional algorithms easily getting trapped in local minima in complex environments. Based on this, this application designs a return strategy that incorporates global path information, enabling local path planning to maintain effective approximation to the global goal during dynamic adjustments. Comparative experiments were conducted on the improved algorithm in various simulation environments, including static (no unknown obstacles), static (with unknown obstacles), and dynamic (with unknown obstacles). Experimental results show that the IAPF algorithm outperforms the traditional APF algorithm and the A*-DWA reference algorithm in terms of path length, safety, and planning efficiency, demonstrating good environmental adaptability and path optimization capabilities, providing theoretical and methodological support for its subsequent application in real-world environments.

[0190] The foregoing description of embodiments of the present invention, through which those skilled in the art are able to implement or use the present invention, will be readily apparent to those skilled in the art. Various modifications to these embodiments will be readily apparent to those skilled in the art. The general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novelty disclosed herein.

Claims

1. A local obstacle avoidance path planning method for intelligent vehicles based on improved artificial potential field (IAPF), characterized in that, The method includes the following steps: S1. Environment and parameter initialization: Construct a grid and / or continuous hybrid map, obtain the current position and / or attitude, global path, destination and obstacle status, and set the gravity / repulsion weight, detection radius, minimum safe distance, and heading and turning angle limits; S2. Target attraction source reconstruction: Replace the single endpoint with the global path "next key point" as the attraction source, set the arrival threshold ε, and switch to the subsequent key point when the distance is less than ε in order to suppress local minima and path oscillation. S3. Improved repulsive force modeling: Within the radius of the obstacle, introduce a "distance to the endpoint" adjustment factor to construct repulsive potential energy, and calculate the negative gradient of the distance and tangential variables to obtain radial and / or tangential repulsive force components, so that the repulsive force and attraction force at the endpoint approach zero simultaneously. S4. Dynamic obstacle prediction: Updates obstacle positions based on observation speed and recalculates distance matrix, refreshes repulsion force and set of possible actions; S5. Resultant force solution and constraint shaping: Summing the gravitational force and each repulsive force vector to obtain the direction and magnitude of the resultant force, and combining kinematics and rotation constraints to limit the heading increment and speed amplitude, forming a one-step executable control quantity; S6. Return to global path strategy: When there are no obstacles within the detection radius, construct a first attraction point and a forward second attraction point on the global path that are consistent with the current heading, and superimpose the attraction potential to limit the return angle and accelerate the fit to the global path. S7. Path generation and key point switching: Update the pose based on the control variables and write the local path sequence; when the switching condition is met, iterate to the next key point until the destination. S8. Termination and Safety Constraints: Termination occurs when the endpoint is reached or the maximum number of iterations is reached; maintain a minimum safe distance from obstacles throughout the process; if a violation occurs, a deceleration and / or detour / back-off mechanism will be triggered.

2. The method according to claim 1, characterized in that, The gravitational potential field function in step S2 is: in, At position p i gravitational potential energy; μ a >0: Gravitational potential field adjustment factor; ρ(p i ,p D )=‖p i -p D ‖2: Current position p i to target point p D The Euclidean distance; The current position of the moving object; Target point location; gravitational The negative gradient of the gravitational potential field function: Distance to p i The gradient, with direction from p i Point to p D The unit vector.

3. The method according to claim 2, characterized in that, The improved repulsive potential field function U in step S3 ri (p i )for: Among them, U rij (p i ) is p i The repulsive potential energy value of point j caused by the j-th obstacle, μ r As a regulating factor for the repulsive potential field, The Euclidean distance from the current location of the intelligent vehicle to the point affected by the repulsive force emanating from the obstacle. in, ρ n (p i ,p D ) is the distance adjustment factor, and ρ0 is the radius of the repulsive potential field of the obstacle; When the intelligent vehicle is within the repulsive potential field of the obstacle, the repulsive force F rij (p i ) respectively ρ(p i ,p D Find the negative differential to obtain the improved F. rij (p i The two components of ) and Specifically, it is expressed as follows: Among them, F rij (p i ) indicates that the intelligent vehicle is in p i The repulsive force exerted on point j by the j-th obstacle. The direction of the vector is from the obstacle towards the intelligent vehicle. The direction of the vector is the direction with the smallest angle to the direction of movement of the intelligent vehicle when the detection area of ​​the intelligent vehicle is tangent to the obstacle.

4. The method according to claim 3, characterized in that, In step S5, the resultant force F ci (p i The calculation model is as follows: Among them, F ci (p i For intelligent vehicles in p i The resultant force of the attractive and repulsive forces acting on a point, F ai (p i F is the gravitational pull on the intelligent vehicle. rij (p i (This indicates that the intelligent vehicle is in p) i The repulsive force generated by the j-th obstacle at point j, where x is the total number of obstacles that generate repulsive forces on the intelligent vehicle. The potential field relationship is as follows: F=F a +F r , in, The gradient of the overall potential; Gradient between gravitational and repulsive potentials, F: total force; F a F r Attraction and total repulsion; F total (p i ): Total resultant force; F attr (p i ): Attraction symbol; The repulsive force symbol for the j-th obstacle.

5. The method according to claim 4, characterized in that, In step S6, a heading decision angle δ smaller than the maximum turning angle of the intelligent vehicle is introduced, and two gravity points B on the global path are set. i When the detection radius d r When there are no obstacles inside, the repulsive potential field generated by the obstacles is closed, and the intelligent vehicle is only subject to the attractive force from the target position; First, draw a ray with an angle δ from the location of the intelligent vehicle to the global path, and define the point where this ray intersects the global path as B1. Second, based on this point, define a point at a distance r towards the target location as B2. Third, define new attractive potential field functions for B1 and B2. for: Among them, B i ∈{B1,B2},ρ(p i B i ) represents the Euclidean distance between the location of the intelligent vehicle and the point of attraction, gravity. The negative gradient of the gravitational potential field function Intelligent vehicle detection radius d r When there are no obstacles inside, it will be affected by B from the global path. i The gravitational force is proportional to the distance between the intelligent vehicle and the gravitational point, and the distance between the intelligent vehicle and point B2 is often greater than that between point B1. In addition, the direction of the resultant force is always biased towards point B2 and points to the global path.

6. The method according to claim 1, characterized in that, The evaluation metrics of the method shall include at least one of the following metrics: 1) Planning and obstacle avoidance success rate The dynamic programming success rate metric measures the algorithm's planning and obstacle avoidance success rate as the proportion of times the algorithm successfully reaches the target point without colliding with obstacles out of the total number of tests. It is applicable to complex environments with both static and dynamic obstacles, and its calculation formula is as follows: Where, N success N represents the number of times the intelligent vehicle successfully avoids obstacles and completes path planning. total Let η be the total number of path planning iterations. s To improve the success rate of obstacle avoidance. 2) Path target value 2.1) The target path length is: Where: L is the total length; L i m is the Euclidean distance between adjacent waypoints; m is the number of waypoints. 2.2) The target for path smoothness is: θ i =atan2(y i -y i-1 ,x i -x i-1 ), Dth i =|θ i+1 -θ i |; Where: Θ represents the cumulative heading change; θ i Let Δθ be the heading angle of the i-th segment. i The difference in heading between adjacent segments; 2.3) A piecewise risk function is used for driving safety: Wherein: H i For the risk at point i; h CiO h is the distance from the vehicle's center of gravity to the nearest obstacle's center of gravity. min ,h max To achieve the minimum safety and maximum radius of influence; 3) Total path time The total path time starts from the starting point p. s (x s ,y s After dynamic obstacle avoidance, the target point p is reached. D (x D ,y D For the total time, this paper often uses Euclidean distance to calculate the distance between adjacent points and sum them: T i =T i -T i-1 , Among them, T i-1 This indicates that the intelligent vehicle has reached the previous node p. i-1 =(x i-1 ,y i-1 At that moment, T i This indicates that the intelligent vehicle has reached the current node p. i =(x i ,y i At that moment, T i L represents the total time it takes for the intelligent vehicle to reach the current node after passing through the previous section, and L represents the time from the starting point p. s (x s ,y s ) to target point p D (x D ,y D Total time required; 4) Minimum distance from obstacles The calculation formula is as follows: D i,min =mind(p i ,O j ), Among them, D i,min Let d(p) be the minimum distance between the i-th point and the obstacle. i O j Let be the distance matrix between the i-th point and each obstacle.

7. A local obstacle avoidance path planning system for intelligent vehicles based on an improved artificial potential field, characterized in that, The system is used to implement the method according to any one of claims 1-6, comprising: The perception and mapping module is used to obtain the grid environment containing static and / or dynamic obstacles, as well as the current position and attitude information of the intelligent vehicle; The global path management module is used to provide the global path and output the "next key point" as the current gravity source point and its arrival determination threshold; The gravity calculation module is used to obtain the gravity vector by taking the negative gradient of the current key point based on the modified gravitational potential field function; the repulsion calculation module is used to obtain the radial repulsion component and the tangential repulsion component by taking the negative gradient of each obstacle within the detection radius based on the repulsion potential energy function containing the "distance adjustment factor to the target point". The resultant force and trajectory update module is used to sum the gravitational and repulsive force vectors to obtain the total resultant force, and update the position of the intelligent vehicle, record the path point sequence and the change in heading angle accordingly; The dynamic obstacle prediction module is used to update the position of dynamic obstacles in real time based on their speed and calculate the distance matrix between them and the intelligent vehicle to refresh the repulsive force. Return to the global path strategy module, which is used to construct two attraction points on the global path and generate corresponding attraction potential fields based on the heading decision angle when there are no obstacles within the detection radius, so as to limit the return heading angle and guide the fast loop.

8. The system according to claim 7, characterized in that, The repulsion calculation module superimposes obstacle repulsion effects from the detection boundary to enhance safety redundancy; the global path management module automatically switches to the next key point on the global path until the endpoint when it detects that the current key point has been reached; the global path return strategy module uses the gravitational superposition of two attraction points to make the resultant force direction deviate towards the global path and satisfy the vehicle's maximum turning angle constraint; the resultant force and trajectory update module is configured to support iteration termination conditions, including reaching the endpoint or reaching the maximum number of iterations.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the method of any one of claims 1–7.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the method of any one of claims 1–7.

Images