Intelligent agent path planning method based on improved RRT and related equipment

Abstract

The invention provides an agent path planning method based on improved RRT and related equipment, and the method comprises the steps: 1, obtaining a current position and a target position of an agent, and constructing a search tree; 2, in the expansion process, sampling is carried out based on a dynamic weight target bias sampling strategy; extending a distance of a dynamic step length from the extension starting point to the extension direction based on a dynamic step length adjustment strategy to obtain a new node, and performing kinematics constraint detection and collision detection on the new node; 3, obtaining an initial path through path backtracking; step 4, when a dynamic obstacle is detected, re-planning a path section conflicting with the dynamic obstacle in the initial path based on an incremental re-planning strategy to obtain a final path for indicating the intelligent agent to move; otherwise, taking the initial path as a final path. According to the invention, the accuracy of path planning can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent agent path planning, and specifically relates to an intelligent agent path planning method based on improved RRT and related equipment. Background Art

[0002] Path planning technology uses spatial modeling and optimization algorithms to determine the optimal trajectory of a mobile object from a starting point to a destination within a specific environment. As a core technology for robotic navigation, autonomous driving, and intelligent logistics systems, this technology significantly improves equipment efficiency and ensures task integrity by constructing optimal paths that meet constraints. The application of path planning technology is particularly critical in modern warehousing and logistics. For example, the widespread use of AGVs (Automated Guided Vehicles) in warehousing and logistics fully demonstrates the importance of path planning technology. AGVs can autonomously transport goods from receiving areas to storage areas and then to shipping areas. Through tight integration with warehouse management systems (WMS), AGVs can automatically plan optimal paths based on order requirements and cargo location, achieving efficient and accurate cargo handling. The current path planning technology architecture primarily consists of two hierarchical layers: Global Path Planning focuses on establishing a theoretically optimal global navigation benchmark for mobile robots within a fully known environmental model, using prior map data and multi-objective optimization criteria (such as shortest path, minimum energy consumption, etc.). This layer utilizes an offline computing model, and its algorithmic architecture can be categorized into three main types: deterministic algorithms based on graph search (such as A* and Dijkstra), stochastic planning algorithms based on probability sampling (such as the RRT series), and machine learning-based environmental understanding and decision-making algorithms. Local path planning, a real-time navigation strategy in dynamic environments, acquires local spatial information through an online environmental perception system, enabling mobile robots to instantly avoid obstacles and correct their trajectories. This reactive planning approach does not rely on a complete prior map, but instead makes online decisions based on sensor data streams, demonstrating enhanced environmental adaptability and dynamic response. The synergy of these two planning techniques forms a complete autonomous navigation system: global planning provides a strategic-level path benchmark, while local planning implements tactical-level motion control.

[0003] Global planning constructs a complete path topology by modeling the prior environment. This allows the system to establish a feasible solution space that satisfies kinematic constraints and mission objectives at the initial stage, thereby avoiding the risk of path inaccessibility that may arise from local planning in an unknown environment. Traditional graph search algorithms are static path planning algorithms. Given a consistent model and environment, they can often only plan a single path, leading to path conflicts when planning for multiple devices simultaneously. In recent years, sampling methods such as Monte Carlo Tree Search (MCTS), Rapid Random Tree Search (RRT), and Probabilistic Roadmap (PRM) have been widely used in path planning. Sampling-based random search technology is a key method in global path planning algorithms. By randomly sampling within the planning space, a large number of possible paths can be quickly acquired, accelerating the path planning process while meeting the path planning requirements for multiple devices.

[0004] Currently, scholars at home and abroad have conducted extensive research based on the RRT algorithm. For example, some researchers have proposed an improved dual-tree obstacle edge search artificial potential field RRT* algorithm (IBPF-RRT*), which accelerates path search speed through the obstacle edge search artificial potential field strategy (ESAPF), and combines a bidirectional pruning strategy to significantly reduce the number of iterations and optimize path quality. Some researchers have proposed an improved RRT-Connect that integrates a dynamic step size strategy and sampling optimization method, improving the RRT algorithm's ability to pass through dense obstacle areas. Other researchers have integrated RRTConnect with the dynamic window method (DWA) for single-target point navigation tasks of indoor mobile robots, enabling autonomous path planning and obstacle avoidance in complex environments.

[0005] However, the path planning method based on traditional RRT has the defects of high randomness, poor path adaptability and insufficient dynamic response in dynamic environments, resulting in low path planning accuracy. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide an intelligent agent path planning method and related equipment based on improved RRT to improve the accuracy of path planning.

[0007] In a first aspect, the present invention provides an agent path planning method based on an improved RRT, the method comprising the following steps:

[0008] Step 1: Get the current position and target position of the agent, and build a search tree with the current position as the root node;

[0009] Step 2: During the expansion process, sampling is performed based on the dynamic weight target bias sampling strategy to obtain a sampling point. The sampling point is used as the expansion direction, and the search tree node closest to the sampling point is used as the expansion starting point. Based on the dynamic step size adjustment strategy, a dynamic step size is extended from the expansion starting point to the expansion direction to obtain a new node. The new node is subjected to kinematic constraint detection and collision detection. If the detection fails, resampling is performed. Otherwise, the new node is added to the search tree and connected to the expansion starting point.

[0010] Step 3: When the distance between the new node and the target location is less than the preset distance threshold, the initial path is obtained by path backtracking; otherwise, the process returns to step 2 to continue the expansion.

[0011] Step 4: When a dynamic obstacle is detected, the path segment in the initial path that conflicts with the dynamic obstacle is replanned based on the incremental replanning strategy to obtain the final path used to instruct the movement of the intelligent agent; otherwise, the initial path is used as the final path.

[0012] Optionally, in step 2, sampling is performed based on a dynamic weight target bias sampling strategy to obtain sampling points, including:

[0013] By calculating the formula

[0014] N sample =w1N goal +w2N rand

[0015]

[0016] stw1+w2=1

[0017] Get the position N of the sampling point sample ; Among them, w1, w2 represent the target weight and random weight respectively, N goal Indicates the target position, N rand Represents the position of a randomly generated random point, w min represents the minimum random sampling weight, is the initial random sampling weight, α represents the weight growth coefficient, and t represents the current expansion times of the search tree, t=1,...,T max , T max Indicates the preset maximum number of expansions, T threshold Indicates the segmentation point, w max Represents the maximum target weight.

[0018] Optionally, the dynamic step size adjustment strategy in step 2 includes:

[0019] The dynamic step size is designed based on the target position distance, obstacle distance and search process to balance the global search efficiency and local path accuracy.

[0020] Optionally, the expression for the dynamic step size is:

[0021] d=min(d max ,α′·d goal +β·d obstacle +γ·S P )

[0022] d goal =||N current -N goal ||

[0023] d obstacle =min ob∈O ‖N current -ob‖

[0024]

[0025] Among them, d represents the dynamic step length of the current node extending in the expansion direction, d max Indicates the maximum allowed step size, d goal Represents the target point distance factor, which is used to measure the Euclidean distance between the current node and the target position, d obstacle Represents the obstacle distance factor, which is used to evaluate the shortest distance between the current node and the nearest obstacle, S P Represents the search process factor, which is used to quantify the search process. α′, β, and γ represent weight coefficients. N current Represents the location of the current node, ob represents the nearest obstacle and its location, and O represents the set of obstacles in the environment where the agent is located.

[0026] Optionally, kinematic constraint detection and collision detection are performed on the new node in step 2, including:

[0027] The line between the new node and the starting point of the expansion is used as the current path;

[0028] If any obstacle in the obstacle set is on the current path, the collision detection result is a collision and the detection fails; otherwise, a kinematic model is constructed according to the geometric characteristics of the intelligent body, and kinematic constraints are constructed based on the kinematic model. If the motion state of the intelligent body in the current path does not conform to the kinematic constraints, the kinematic constraint detection result is non-compliance and the detection fails; otherwise,

[0029] The test passed.

[0030] Optionally, the intelligent body includes at least a vehicle, the geometric characteristics include at least a vehicle body length, a turning radius and a steering angle, and the kinematic constraints include at least a curvature constraint.

[0031] The kinematic constraints are:

[0032] k current≤k max

[0033]

[0034] Among them, k current represents the current curvature, k max represents the maximum curvature, represents the deflection angle between the current node and the target position, D represents the straight-line distance between the current node and the target position, δ max Indicates the maximum steering angle, L indicates the vehicle body length, R min Indicates the minimum turning radius.

[0035] Optionally, the incremental replanning strategy in step 4 includes:

[0036] Determine whether the trajectory of the dynamic obstacle intersects the initial path;

[0037] If there is an intersection, the path segment where the intersection is located in the initial path is pruned from the search tree, and the subtree disconnected from the search tree is pruned to ensure the connectivity between the search tree and the starting point, and the path from the disconnected point to the target location is re-extended on the pruned search tree.

[0038] In a second aspect, the present invention provides an intelligent agent path planning device based on an improved RRT, comprising:

[0039] The search tree construction module is used to obtain the current position and target position of the agent and construct a search tree with the current position as the root node;

[0040] The initial path planning module is used to sample based on the dynamic weight target bias sampling strategy during the expansion process to obtain sampling points. The sampling points are used as the expansion direction, and the search tree node closest to the sampling points is used as the expansion starting point. Based on the dynamic step adjustment strategy, a new node is obtained by extending a dynamic step length from the expansion starting point to the expansion direction. The new node is subjected to kinematic constraint detection and collision detection. If the detection fails, the node is resampled. Otherwise, the new node is added to the search tree and connected to the expansion starting point.

[0041] The initial path acquisition module is used to obtain the initial path by backtracking when the distance between the new node and the target location is less than a preset distance threshold; otherwise, it returns to execute the initial path planning module to continue expansion;

[0042] The initial path optimization module, when a dynamic obstacle is detected, replans the path segment in the initial path that conflicts with the dynamic obstacle based on the incremental replanning strategy to obtain the final path used to instruct the intelligent agent to move; otherwise, the initial path is used as the final path.

[0043] In a third aspect, the present invention provides a terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method when executing the computer program.

[0044] In a fourth aspect, the present invention discloses a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0045] The beneficial effects of the present invention are:

[0046] The intelligent agent path planning method based on improved RRT provided by the present invention performs sampling based on a dynamic weight target bias sampling strategy to obtain sampling points, takes the sampling points as the expansion direction, and takes the search tree node closest to the sampling points as the expansion starting point to extend and obtain new nodes. This method can overcome the defects of low environmental exploration and easy falling into local optimal solutions caused by traditional methods that perform sampling in the direction of the target point with a fixed probability. The target direction weight is adaptively adjusted through dynamic weights to balance path exploration and search performance, which is beneficial to improving the accuracy of path planning. The node extension distance is adjusted based on a dynamic step size adjustment strategy, which can overcome the defects of traditional methods that perform expansion with a fixed step size, resulting in an uneven and rough path, balance the global search efficiency and local path accuracy, and is beneficial to improving the accuracy of path planning. Kinematic constraint detection and collision detection are performed on new nodes to ensure that the new nodes conform to the actual environment and the kinematic characteristics of the intelligent agent itself, which is beneficial to improving the accuracy of path planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flowchart of an intelligent agent path planning method based on improved RRT in one embodiment of the present application;

[0048] Figure 2 This is a schematic diagram of a dynamic weight target bias sampling strategy in one embodiment of the present application;

[0049] Figure 3 This is a schematic diagram of kinematic constraints in one embodiment of the present application;

[0050] Figure 4 This is a structural diagram of an intelligent agent path planning device based on improved RRT in one embodiment of the present application;

[0051] Figure 5 This is a schematic diagram of the structure of a terminal device in one embodiment of the present application. DETAILED DESCRIPTION

[0052] In response to the problem of low accuracy of existing path planning methods based on traditional RRT, the present invention discloses an intelligent agent path planning method and related equipment based on improved RRT. The method performs sampling based on a dynamic weight target bias sampling strategy to obtain sampling points, uses the sampling points as the expansion direction, and uses the search tree node closest to the sampling points as the expansion starting point to extend and obtain new nodes. This method can overcome the defects of low environmental exploration and easy falling into local optimal solutions caused by traditional methods that perform sampling in the direction of the target point with a fixed probability. The target direction weight is adaptively adjusted through dynamic weights to balance path exploration and search performance, which is beneficial to improving the accuracy of path planning. The node extension distance is adjusted based on a dynamic step size adjustment strategy, which can overcome the defects of traditional methods that perform expansion with a fixed step size, resulting in an uneven and rough path, balance the global search efficiency and local path accuracy, and is beneficial to improving the accuracy of path planning. Kinematic constraint detection and collision detection are performed on new nodes to ensure that the new nodes conform to the actual environment and the kinematic characteristics of the intelligent agent itself, which is beneficial to improving the accuracy of path planning.

[0053] The following is an explanation of the intelligent agent path planning method based on improved RRT disclosed in the present invention.

[0054] like Figure 1 As shown, the intelligent agent path planning method based on the improved RRT includes the following steps:

[0055] Step 1: Get the current position and target position of the agent, and build a search tree with the current position as the root node.

[0056] It should be understood that the RRT (Rapidly-exploring Random Tree) algorithm is a path planning algorithm that constructs a tree through random sampling, and is mainly used to find a path from the starting point to the target point in a high-dimensional space. The basic idea is to explore the reachable area in the space by rapidly expanding the tree structure, so as to find an effective path from the starting point to the target point. In the initialization phase of the RRT algorithm, a search tree is constructed starting from the current position. At this time, the search tree has only one root node. The root node represents the current position of the agent and represents the starting point of the path planning. The target position is the end point of the path planning. Through expansion, the search tree will gradually extend the path from the root node to the target position. The path includes multiple nodes and directed edges connecting multiple nodes. The node represents the position of the agent, and the directed edge represents the direction of the agent's movement.

[0057] In the embodiments of the present invention, different numbers of agents correspond to different numbers of search trees, with the number of agents corresponding to the number of search trees, and each search tree corresponding to a planned path. In single-agent path planning, there is only one search tree, which determines a planned path. In multi-agent path planning, there are multiple search trees, each corresponding to a planned path for a single agent.

[0058] Step 2: During the expansion process, sampling is performed based on the dynamic weight target bias sampling strategy to obtain sampling points. The sampling points are used as the expansion direction, and the search tree node closest to the sampling points is used as the expansion starting point. Based on the dynamic step adjustment strategy, a dynamic step length is extended from the expansion starting point to the expansion direction to obtain a new node. The new node is subjected to kinematic constraint detection and collision detection. If the detection fails, resampling is performed. Otherwise, the new node is added to the search tree and connected to the expansion starting point.

[0059] On the one hand, the original RRT algorithm's sampling strategy is random sampling within the sampling space. While this sampling strategy can discover more paths, it significantly reduces efficiency. To address this issue, researchers in related fields have introduced target point direction information, making it more likely that the sampling points will be oriented toward the target point, thereby accelerating the path planning process and improving efficiency. Traditional approaches include target-guided sampling, where each sampling step selects the direction toward the target point with a certain probability and randomly samples with another probability to ensure the global nature of the search. While this approach can significantly accelerate path planning efficiency, it provides a low level of exploration of the environment and is prone to falling into local optimal solutions.

[0060] To this end, the present invention proposes a sampling strategy based on dynamic weight target bias, which balances the exploration and convergence performance by adaptively adjusting the target direction weight.

[0061] Specifically, by calculating the formula

[0062] N sample =w1N goal +w2N rand

[0063]

[0064] stw1+w2=1

[0065] Get the position N of the sampling point sanple ; Among them, w1, w2 represent the target weight and random weight respectively, N goal Indicates the target position, N rand Represents the position of a randomly generated random point, w minrepresents the minimum random sampling weight, that is, the initial random sampling weight, α represents the weight growth coefficient, and t represents the current expansion times of the search tree, t=1,...,T max , T max Indicates the preset maximum number of expansions, T threshold Indicates the segmentation point, w max Represents the maximum target weight.

[0066] For ease of understanding, the following Figure 2 The dynamic weight target bias sampling strategy disclosed in the present invention is described.

[0067] like Figure 2 As shown, in the embodiment of the present invention, combined with the current exploration process The target weight is adjusted by segmentation strategy. In the initial stage (t≤T threshold ), the present invention ensures that the sampling point is oriented toward the target position by using a smaller target weight while increasing the degree of environmental exploration as much as possible to avoid falling into a local optimal solution. threshold ), at this time, rich environmental information has been explored. In order to improve the efficiency of path planning, the sampling point is ensured to be oriented towards the target position with the maximum target weight. This reflects the balance between exploration and convergence performance of the present invention, which is conducive to improving the accuracy of path planning. start Represents the root node.

[0068] On the other hand, in traditional RRT algorithms, path planning usually uses a fixed step size for node expansion. Although this strategy is simple and easy to implement, it has the problem of path roughness in practical applications. The local details of the path are difficult to accurately characterize, which may lead to an uneven path or obstacle avoidance failure.

[0069] In order to balance the global search efficiency and local path accuracy, in an embodiment of the present invention, a new node is obtained by extending a dynamic step length from the expansion starting point to the expansion direction based on a dynamic step length adjustment strategy. Specifically, the dynamic step length is designed in combination with the target position distance, obstacle distance and search process to balance the global search efficiency and local path accuracy.

[0070] In a feasible embodiment, the expression of the dynamic step size is:

[0071] d=min(d max ,α′·d goal +β·d obstacle +γ·S P )

[0072] d goal =||N current -N goal ||

[0073] dobstacle =min ob∈O ‖N current -ob‖

[0074]

[0075] Among them, d represents the dynamic step length of the current node extending in the expansion direction, d max Indicates the maximum step length allowed within the map range, d goal Represents the target point distance factor, which is used to measure the Euclidean distance between the current node and the target position, d obstacle Represents the obstacle distance factor, which is used to evaluate the shortest distance between the current node and the nearest obstacle, S P Represents the search process factor, which is used to quantify the search process. α′, β, and γ represent weight coefficients. N current Represents the location of the current node, ob represents the nearest obstacle and its location, and O represents the set of obstacles in the environment where the agent is located.

[0076] It should be noted that the target point distance factor is designed to guide the algorithm to use a larger step size when the current node or current position is far from the target location to accelerate the search process; while gradually reducing the step size as it approaches the target location to improve path accuracy. The obstacle distance factor is designed to use a larger step size when far from obstacles to improve search efficiency, and a smaller step size when approaching obstacles to avoid collisions and improve path safety. The search progress factor is designed to use a larger step size in the early stages of the search to quickly explore the unknown space, and gradually reduce the step size in the later stages of the search to optimize path accuracy. These measures help achieve a balance between global search efficiency and local path accuracy.

[0077] On the other hand, in traditional RRT algorithms, state expansion primarily relies on extending edges from the current node to the new sampling point. This fails to consider the agent's actual kinematic constraints, leading to situations where the agent's kinematic characteristics prevent it from moving along the planned path. To address this, the present invention performs both collision detection and kinematic constraint detection on new nodes.

[0078] Specifically, kinematic constraint detection and collision detection are performed on the new node, including steps I to II:

[0079] In step I, the line between the new node and the expansion starting point is used as the current path.

[0080] In step II, if any obstacle in the obstacle set is on the current path, the collision detection result is a collision and the detection fails. Otherwise, a kinematic model is constructed according to the geometric characteristics of the intelligent body, and kinematic constraints are constructed based on the kinematic model. If the motion state of the intelligent body in the current path does not comply with the kinematic constraints, the kinematic constraint detection result is non-compliance and the detection fails. Otherwise, the detection passes.

[0081] In a feasible embodiment, the intelligent agent is a vehicle, the geometric characteristics include at least the body length, turning radius and steering angle, and the kinematic constraints include at least the curvature constraint, specifically Figure 3 In this embodiment, the kinematic constraints are:

[0082] k current ≤k max

[0083]

[0084] Among them, k current represents the current curvature, k max represents the maximum curvature, represents the deflection angle between the current node and the target position, D represents the straight-line distance between the current node and the target position, δ max Indicates the maximum steering angle, L indicates the vehicle body length, R min Indicates the minimum turning radius.

[0085] In another feasible embodiment, the intelligent agent is a drone, and its corresponding kinematic constraints include attitude constraints, specifically roll, pitch, and yaw constraints. Based on the same principle, during the expansion process, the maximum roll, pitch, and yaw angles are determined by constructing a kinematic model of the drone and used as the drone's kinematic constraints.

[0086] Step 3: If the distance between the new node and the target location is less than the preset distance threshold, the initial path is retrieved by backtracking. Otherwise, the process returns to step 2 to continue the expansion.

[0087] When the distance between the new node and the target location is less than a preset distance threshold, it indicates that the new node is close to the target location. The new node can be used as the target node, and the complete path can be obtained by backtracking from the target node to the starting node. It should be understood that the backtracking process is usually implemented by tracking the parent node of each node.

[0088] In one possible embodiment, the path may refer to A * The algorithm or the implementation of the Dijkstra algorithm is common knowledge to those skilled in the art and will not be described in detail here.

[0089] In another feasible embodiment, the path may be backtracked according to the shortest distance or the shortest time to obtain the initial path according to actual needs.

[0090] Step 4: When a dynamic obstacle is detected, the path segments in the initial path that conflict with the dynamic obstacle are replanned based on the incremental replanning strategy to obtain the final path used to instruct the agent to move. Otherwise, the initial path is used as the final path.

[0091] Specifically, the incremental replanning strategy includes:

[0092] Determine whether the trajectory of the dynamic obstacle intersects the initial path. The existence of an intersection indicates that the dynamic obstacle may collide with the agent, requiring obstacle avoidance.

[0093] If there is an intersection, the path segment where the intersection is located in the initial path is pruned from the search tree, and the subtree disconnected from the search tree is pruned to ensure the connectivity between the search tree and the starting point, and the path from the disconnected point to the target location is re-extended on the pruned search tree.

[0094] If there is no intersection, it means that the dynamic obstacle will not collide with the agent and there is no need to replan the path.

[0095] It is worth mentioning that the incremental replanning strategy disclosed in the present invention, on the one hand, can respond to dynamically appearing obstacles and adjust the path in a timely manner, which is conducive to improving the accuracy of path planning; on the other hand, it only makes local adjustments to the part of the path affected by dynamic obstacles, thereby avoiding the complete regeneration of the entire path, which is conducive to improving the efficiency of path planning.

[0096] In order to verify the effectiveness of the intelligent agent path planning method based on improved RRT provided by the present invention, in another embodiment of the present invention, the traditional RRT algorithm (the sampling strategy uses random sampling, without kinematic constraints and dynamic planning mechanism), the traditional target-guided RRT algorithm (the RRT algorithm that only adds the target direction probability-guided sampling strategy), the dynamic RRT algorithm (the dynamic RRT algorithm that adopts incremental replanning) and the present method are compared and verified under the same conditions.

[0097] Since the algorithm is random, all algorithms are run 100 times on the same map, and the average value of the comparison indicators is taken. The comparison indicators include path length (the cumulative value of the Euclidean distance of the planned path), planning time (the time taken to complete the first planning, unit: seconds), re-planning time (the time taken to complete the planning after dynamic obstacles trigger re-planning), path curvature (smoothness is evaluated by the mean, maximum and standard deviation of the curvature of discrete points), and planning failure rate (the proportion of experiments that cannot reach the end point in 100 experiments). The calculation formula of curvature is: Represents the first-order derivative of the path in the x, y direction, representing the speed, It represents the second-order derivative of the path in the x and y directions, characterizing the acceleration, and can be calculated using the finite difference method on discrete paths.

[0098] The path planning environment is set to a 1000x900 rectangular area with nine randomly distributed static obstacles. After planning is complete, a fixed-radius obstacle is inserted at a randomly selected node on the planned path to truncate the original path. The algorithm parameters are set based on the optimization results of previous pre-experimental experiments: 5000 iterations, a basic step size of 10 (balancing search efficiency and path resolution), an initial target guidance weight of 0.1, and kinematic constraints based on a standard passenger car model: a minimum turning radius of 6m, a maximum steering angle of 33°, and a vehicle length of 4m.

[0099] The hardware and software configuration of the simulation experiment environment is as follows: operating system Windows 11, CPU Intel(R) Core(TM) i7-14650HX, main frequency 2.20GHz, memory 32GB, programming language Python.

[0100] The comparison results are shown in Table 1.

[0101] Table 1

[0102]

[0103] As can be seen from Table 1, the path length: This method uses dynamic weight target bias sampling to guide the search direction, reducing invalid expansion nodes, and is 4.8% shorter than the traditional RRT;

[0104] Planning time: This method uses adaptive adjustment of target bias weights to significantly reduce redundant calculations caused by random sampling;

[0105] Replanning time: This method uses incremental replanning to prune only the affected path segments, avoiding global reconstruction of the search tree, and improving efficiency by 99.7%;

[0106] Path curvature: This method uses kinematic constraint filtering to force the path to meet vehicle steering constraints, reducing the curvature standard deviation by 31.3%;

[0107] Failure rate: This method greatly improves the probability of finding a feasible path through efficient sampling and replanning strategies, reducing the blindness of the search.

[0108] In summary, the intelligent agent path planning method based on improved RRT provided by the present invention performs sampling based on a dynamic weight target bias sampling strategy to obtain sampling points, takes the sampling points as the expansion direction, and takes the search tree node closest to the sampling points as the expansion starting point to extend and obtain new nodes. This method can overcome the defects of low degree of environmental exploration and easy falling into local optimal solutions caused by the traditional method of sampling in the direction of the target point with a fixed probability. The target direction weight is adaptively adjusted through dynamic weights to balance the path exploration and search performance, which is beneficial to improving the accuracy of path planning; the node extension distance is adjusted based on the dynamic step size adjustment strategy, which can overcome the defects of the traditional method of expanding with a fixed step size, resulting in an uneven and rough path, balance the global search efficiency and local path accuracy, and is beneficial to improving the accuracy of path planning; kinematic constraint detection and collision detection are performed on the new nodes to ensure that the new nodes conform to the actual environment and the kinematic characteristics of the intelligent agent itself, which is beneficial to improving the accuracy of path planning.

[0109] The following describes an intelligent agent path planning device based on improved RRT provided by the present invention.

[0110] like Figure 4 As shown, the intelligent agent path planning device 400 includes:

[0111] Search tree construction module 401 is used to obtain the current position and target position of the agent and construct a search tree with the current position as the root node;

[0112] Initial path planning module 402 is used to perform sampling based on a dynamic weight target bias sampling strategy during the expansion process to obtain a sampling point. Using the sampling point as the expansion direction and the search tree node closest to the sampling point as the expansion starting point, a new node is obtained by extending the distance of a dynamic step length from the expansion starting point in the expansion direction based on a dynamic step size adjustment strategy. The new node is subjected to kinematic constraint and collision checks. If the checks fail, the new node is resampled. Otherwise, the new node is added to the search tree and connected to the expansion starting point.

[0113] The initial path acquisition module 403 is used to obtain the initial path by backtracking when the distance between the new node and the target location is less than a preset distance threshold; otherwise, it returns to the initial path planning module 402 to continue the expansion;

[0114] The initial path optimization module 404, when a dynamic obstacle is detected, replans the path segment in the initial path that conflicts with the dynamic obstacle based on the incremental replanning strategy to obtain the final path used to instruct the movement of the intelligent agent; otherwise, the initial path is used as the final path.

[0115] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiments of the present application. Their specific functions and technical effects can be found in the method embodiment section and will not be described in detail here. Those skilled in the art will clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the above-mentioned method embodiment and will not be described in detail here.

[0116] like Figure 5 As shown, an embodiment of the present invention provides a terminal device, such as Figure 5 As shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 5 Only one processor is shown in the figure), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 implements the steps of any of the above method embodiments when executing the computer program D102.

[0117] Specifically, when the processor D100 executes the computer program D102, it executes step 1 to obtain the current position and target position of the intelligent agent, and constructs a search tree with the current position as the root node; step 2, during the expansion process, sampling is performed based on the dynamic weight target bias sampling strategy to obtain a sampling point, the sampling point is used as the expansion direction, and the search tree node closest to the sampling point is used as the expansion starting point. Based on the dynamic step adjustment strategy, a dynamic step length is extended from the expansion starting point to the expansion direction to obtain a new node, and kinematic constraint detection and collision detection are performed on the new node. If the detection fails, the sampling is resampled; otherwise, the new node is added to the search tree and connected to the expansion starting point; step 3, when the distance between the new node and the target position is less than a preset distance threshold, the initial path is obtained by path backtracking; otherwise, return to execute step 2 to continue expansion; step 4, when a dynamic obstacle is detected, the path segment in the initial path that conflicts with the dynamic obstacle is replanned based on the incremental replanning strategy to obtain the final path for instructing the movement of the intelligent agent; otherwise, the initial path is used as the final path. Among them, sampling is performed based on the dynamic weight target bias sampling strategy to obtain sampling points. The sampling points are used as the expansion direction, and the search tree node closest to the sampling points is used as the expansion starting point to extend the new nodes. This can overcome the defects of the traditional method of sampling in the direction of the target point with a fixed probability, which causes a low degree of environmental exploration and is prone to falling into local optimal solutions. The target direction weight is adaptively adjusted by dynamic weight to balance the path exploration and search performance, which is beneficial to improving the accuracy of path planning; the node extension distance is adjusted based on the dynamic step size adjustment strategy, which can overcome the defects of the traditional method of expanding with a fixed step size, which leads to an uneven and rough path, and balance the global search efficiency and local path accuracy, which is beneficial to improving the accuracy of path planning; kinematic constraint detection and collision detection are performed on the new nodes to ensure that the new nodes conform to the actual environment and the kinematic characteristics of the intelligent body itself, which is beneficial to improving the accuracy of path planning.

[0118] The processor D100 may be a central processing unit (CPU), or may be another general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0119] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may also be an external storage device of the terminal device D10, such as a plug-in hard disk, a smart memory card (SMC, SmartMedia Card), a secure digital (SD, Secure Digital) card, a flash card, etc. equipped on the terminal device D10. Furthermore, the memory D101 may also include both an internal storage unit of the terminal device D10 and an external storage device. The memory D101 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program. The memory D101 may also be used to temporarily store data that has been output or is to be output.

[0120] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned various method embodiments can be implemented.

[0121] An embodiment of the present application provides a computer program product. When the computer program product is run on a terminal device, the terminal device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0122] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0123] The one or more embodiments of this application are intended to encompass all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of this application should be included in the scope of protection of this application.

Claims

1. An agent path planning method based on improved RRT, characterized in that: include: Step 1: Get the current position and target position of the agent, and build a search tree with the current position as the root node; Step 2: During the expansion process, sampling is performed based on the dynamic weight target bias sampling strategy to obtain a sampling point. The sampling point is used as the expansion direction, and the search tree node closest to the sampling point is used as the expansion starting point. Based on the dynamic step size adjustment strategy, a new node is obtained by extending the distance of a dynamic step size from the expansion starting point to the expansion direction. The new node is subjected to kinematic constraint detection and collision detection. If the detection fails, resampling is performed. Otherwise, the new node is added to the search tree and connected to the expansion starting point. Step 3: When the distance between the new node and the target location is less than a preset distance threshold, an initial path is obtained by path backtracking; Otherwise, return to step 2 to continue expanding; Step 4: when a dynamic obstacle is detected, replanning the path segment in the initial path that conflicts with the dynamic obstacle based on an incremental replanning strategy to obtain a final path for instructing the movement of the agent; Otherwise, the initial path is used as the final path.

2. The agent path planning method according to claim 1, characterized in that: In step 2, sampling is performed based on the dynamic weight target bias sampling strategy to obtain sampling points, including: By calculating the formula N sample =w1N goal +w2N rand stw1+w2=1 Get the position N of the sampling point sample ; Among them, w1, w2 represent the target weight and random weight respectively, N goal Indicates the target position, N rand Represents the position of the randomly generated random point, w min represents the minimum random sampling weight, is the initial random sampling weight, α represents the weight growth coefficient, and t represents the current expansion times of the search tree, t=1,...,T max , T max Indicates the preset maximum number of expansions, T threshold Indicates the segmentation point, w max Represents the maximum target weight.

3. The agent path planning method according to claim 2, characterized in that: The dynamic step size adjustment strategy in step 2 includes: The dynamic step size is designed based on the target position distance, obstacle distance and search process to balance the global search efficiency and local path accuracy.

4. The agent path planning method according to claim 3, characterized in that: The expression of the dynamic step size is: d=min(d max ,α′·d goal +β·d obstacle +γ·S P ) d goal =||N current -N goal || the obstacle =min ob∈O ‖N current -whether‖ Among them, d represents the dynamic step length of the current node extending in the expansion direction, d max Indicates the maximum allowed step size, d goal Represents the target point distance factor, which is used to measure the Euclidean distance between the current node and the target position, d obstacle Represents the obstacle distance factor, which is used to evaluate the shortest distance between the current node and the nearest obstacle, S P Represents the search process factor, which is used to quantify the search process. α′, β, and γ represent weight coefficients. N current Represents the location of the current node, ob represents the nearest obstacle and its location, and O represents the set of obstacles in the environment where the agent is located.

5. The agent path planning method according to claim 4, characterized in that: In step 2, kinematic constraint detection and collision detection are performed on the new node, including: Taking the line between the new node and the expansion starting point as the current path; If any obstacle in the obstacle set is on the current path, the collision detection result is a collision, and the detection fails; otherwise, a kinematic model is constructed according to the geometric characteristics of the intelligent body, and a kinematic constraint is constructed based on the kinematic model. If the motion state of the intelligent body in the current path does not conform to the kinematic constraint, the kinematic constraint detection result is non-compliance, and the detection fails; otherwise, The test passed.

6. The agent path planning method according to claim 5, characterized in that: The intelligent agent includes at least a vehicle, the geometric characteristics include at least a vehicle body length, a turning radius and a steering angle, and the kinematic constraints include at least a curvature constraint; The kinematic constraints are: k current ≤k max Among them, k current represents the current curvature, k max represents the maximum curvature, represents the deflection angle between the current node and the target position, D represents the straight-line distance between the current node and the target position, δ max Indicates the maximum steering angle, L indicates the vehicle body length, R min Indicates the minimum turning radius.

7. The agent path planning method according to claim 6, characterized in that: The incremental replanning strategy in step 4 includes: Determining whether the trajectory of the dynamic obstacle intersects the initial path; If there is an intersection, the path segment where the intersection is located in the initial path is pruned from the search tree by pruning, and the subtree disconnected from the search tree is pruned at the same time to ensure the connectivity between the search tree and the starting point, and re-expand the path from the disconnected point to the target location on the pruned search tree.

8. An intelligent agent path planning device based on improved RRT, characterized in that: include: A search tree construction module is used to obtain the current position and target position of the agent and construct a search tree with the current position as the root node; An initial path planning module is configured to perform sampling based on a dynamic weight target bias sampling strategy during the expansion process to obtain a sampling point, use the sampling point as the expansion direction, and use the search tree node closest to the sampling point as the expansion starting point. Based on a dynamic step size adjustment strategy, the module extends a distance of a dynamic step size from the expansion starting point toward the expansion direction to obtain a new node. The module performs kinematic constraint and collision checks on the new node. If the checks fail, the module performs resampling. Otherwise, the module adds the new node to the search tree and connects it to the expansion starting point. An initial path acquisition module, configured to acquire an initial path by path backtracking when the distance between the new node and the target location is less than a preset distance threshold; Otherwise, return to execute the initial path planning module to continue expansion; an initial path optimization module, which, when a dynamic obstacle is detected, replans the path segment of the initial path that conflicts with the dynamic obstacle based on an incremental replanning strategy to obtain a final path for instructing the movement of the agent; Otherwise, the initial path is used as the final path.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images