MDC-AFI adaptive feedback fusion intelligent path planning method and system based on multi-dimensional cooperative enhancement

Abstract

The invention discloses an MDC-AFI adaptive feedback fusion intelligent path planning method and system based on multi-dimensional cooperative enhancement, and belongs to the technical field of robot path planning, and the method comprises the steps: obtaining the initial state information of an intelligent robot; based on the initial state information, calling a dynamic adaptive global path planner to search a global path, performing iterative calculation to obtain global path nodes, and generating a global optimization path; decomposing the global optimization path to obtain a plurality of local target points, taking the local target points as initial reference input of a local path planner, adaptively adjusting parameters of the local path planner through an MDC-AFI framework, and iteratively generating a dynamically optimized local path; and when the degree that the local path deviates from the global optimization path exceeds a preset tolerance range, judging whether the global path needs to be optimized again, and feeding back the latest global path data to the local path planner for feedback fusion until the intelligent robot reaches the target point.

Description

Technical Field

[0001] The present invention belongs to the technical field of robot path planning, and particularly relates to an intelligent path planning method and system based on multi-dimensional collaborative enhancement MDC-AFI adaptive feedback fusion. Background Art

[0002] In the field of modern robot autonomous navigation and path planning, improving the efficiency and reliability of path planning is the key for intelligent robots to achieve efficient operation in complex dynamic environments. In application scenarios such as logistics warehousing and driverless, intelligent robots need to simultaneously consider global path planning and local path obstacle avoidance to cope with the challenges of dynamic obstacle distribution and real-time environmental changes. Traditional path planning methods mainly include graph theory-based path search algorithms and sampling-based probabilistic path planning algorithms. The former, such as the A* algorithm, is widely used for its simple and efficient characteristics, but its real-time performance and flexibility in dealing with dynamic environments are poor. The latter, such as the RRT algorithm, searches for paths in high-dimensional spaces through random sampling and is suitable for solving complex problems, but there are deficiencies in the smoothness and global optimality of the planned paths. The feature fusion ability between the global and local path planning modules is the key to improving path planning performance.

[0003] To improve the adaptability of path planning algorithms to complex environments, researchers have proposed dynamic path optimization models, cost map fusion algorithms, and reinforcement learning-driven path planning methods to optimize paths by integrating environmental perception, path evaluation, and dynamic characteristics. However, there are significant deficiencies in the feature fusion between the global and local path planning modules of the existing technologies. For example, the decomposition and reconstruction processes of global paths and local paths lack comprehensive consideration of real-time environmental information, resulting in weak adaptability to dynamic environments. Therefore, there is an urgent need for an intelligent path planning method in the existing technologies that can achieve efficient dynamic fusion of global and local paths in complex dynamic environments while taking into account path smoothness and dynamic constraints. Summary of the Invention

[0004] The present invention aims to solve the deficiencies of the existing technologies and provides the following solutions:

[0005] An intelligent path planning method based on multi-dimensional collaborative enhancement MDC-AFI adaptive feedback fusion, comprising the following steps:

[0006] S1. Obtain the initial state information of the intelligent robot, where the initial state information includes: initial coordinates, attitude information, obstacle distribution layer information, and target point coordinates;

[0007] S2. Based on the initial state information, call a dynamic adaptive global path planner to search for a global path, iteratively calculate to obtain global path nodes, and generate a globally optimized path;

[0008] S3. Decompose the global optimized path to obtain several local target points and use them as the initial reference inputs of the local path planner. Adaptively adjust the parameters of the local path planner through the multi-dimensional feasibility dynamic evaluator in the MDC-AFI framework and the distribution information of dynamic obstacles in the local environment. Iteratively generate a dynamically optimized local path according to the dynamic constraints of the intelligent robot and the real-time environment information.

[0009] S4. When the deviation degree of the local path from the global optimized path exceeds the preset tolerance range, the multi-dimensional feasibility dynamic evaluator will judge whether it is necessary to re-optimize the global path and feedback the latest global path data to the local path planner for feedback fusion.

[0010] S5. Repeat S1 to S4 until the intelligent robot reaches the target point and completes the target navigation task.

[0011] Preferably, in S1, the method for obtaining the attitude information includes:

[0012] The formula for calculating the quaternion using the accelerometer and magnetometer data in the IMU:

[0013]

[0014] where θ represents the rotation angle around the unit vector u = [u X , u Y , u Z , w represents the real part of the quaternion, and x, y, z represent the imaginary parts of the quaternion, respectively representing the components of rotation around the X, Y, and Z axes;

[0015] Substitute the initial attitude into the quaternion formula to obtain the initial quaternion q 0 = [w 0 , x 0 , y 0 , z 0 :

[0016] Calculate the angular increment Δq through the gyroscope angular velocity, update the quaternion, and obtain the updated quaternion:

[0017]

[0018] where t 0 represents the initial moment, and Δt represents the update time step;

[0019] Solve the attitude angle using the updated quaternion q' = [w', x', y', z'] to obtain the attitude information:

[0020]

[0021] Pitch = sin -1 (2(w′y′ - x′z′)), |Pitch| ≤ 90°,

[0022]

[0023] where Roll represents the roll angle of rotation about the X-axis, Pitch represents the pitch angle of rotation about the Y-axis, and Yaw represents the yaw angle of rotation about the Z-axis;

[0024] The rotation matrix R is calculated through the quaternion q = [w, x, y, z]:

[0025]

[0026] The corresponding rotation matrix R is calculated based on the attitude information of each sensor ε ;

[0027] Map the local coordinate systems of each sensor to the global coordinate system to provide a unified reference for subsequent algorithms and ensure the correct transfer and processing of data between different coordinate systems:

[0028]

[0029] where ε represents the index of each sensor in the sensor group;

[0030] A reversible transformation, that is, converting the global coordinates to local coordinates:

[0031]

[0032] where represents the local coordinates of the sensor with index ε.

[0033] Preferably, the method for optimizing the global path includes:

[0034] Introduce a dynamic adaptive cost heuristic function h′(n), combine multi-dimensional information such as the distribution density of obstacles, the velocity direction of dynamic obstacles, and the Euclidean distance to the target point, and optimize the path search efficiency;

[0035] Adopt a path smoothing algorithm to optimize the objective function to process the initial global path, reduce the path polyline, and improve the smoothness and controllability of the path;

[0036] Use hierarchical grid modeling, set higher weights for high-complexity regions and lower weights for low-complexity regions to improve the globality and efficiency of path search.

[0037] Preferably, the method for optimizing the path search efficiency includes:

[0038] Calculate the time cost h time (n), that is, the shortest time estimate from node n to the target point G:

[0039]

[0040] where, ||n - g|| represents the Euclidean distance from node n to the target point G, and v represents the linear velocity of the intelligent robot;

[0041] Calculate the spatial distance cost h distance (n), that is, the Euclidean distance from node n to the target point G:

[0042]

[0043] where, (x n , y n ) represents the coordinates of node n, and (x g , y g ) represents the coordinates of the target point G;

[0044] Calculate the dynamic environment cost h dynamic (n), that is, the impact of dynamic obstacles on path planning:

[0045]

[0046] where, k represents the number of dynamic obstacles, d i represents the distance from node n to the i-th dynamic obstacle, and θ i represents the angle between the direction from node n to the target point G and the direction of the velocity of the dynamic obstacle;

[0047] Based on the time cost h time (n), the spatial distance cost h distance (n) and the dynamic environment cost h dynamic (n), construct a dynamic adaptive cost heuristic function h′(n):

[0048] h'(n) = αh time (n) + βh distance (n) + γh dynamic (n),

[0049] where, α, β, γ respectively represent the weight parameters of h time (n), h distance (n), h dynamin (n), and α + β + γ = 1;

[0050] Express the dynamic adaptive cost heuristic function h′(n) as the product of a weight vector and a cost vector:

[0051] h′(n) = W·H(n),

[0052] where W = [α, β, γ] represents the weight vector, and H(n) = [h time (n), h distance (n), h dynamic (n)] T represents the cost vector;

[0053] Define the dynamic environment parameter vector and dynamically and adaptively adjust the weight vector W according to the environment parameter vector E(t):

[0054] W = W 0 + K·E(t),

[0055]

[0056] E(t) = [ρ obs , d obs , c env ,

[0057]

[0058] where W 0 represents the initial weight vector, K represents the gain matrix, ρ obs represents the dynamic obstacle density, d obs represents the closest distance between the node and the dynamic obstacle, c env represents the openness of the environment, represents the dynamic density function, represents the area of the environmental area p = [x, y] T represents the two-dimensional coordinates of the dynamic obstacle, represents the two-dimensional real space, i represents the index of the dynamic obstacle set, P node represents the coordinates of the node position to be evaluated, v i represents the linear velocity vector of the dynamic obstacle with index i at time t, P i (t) represents the coordinates of the dynamic obstacle with index i at time t, represents the obstacle set, ||·|| 2 represents the Euclidean norm, ρ max represents the maximum possible obstacle density within the unit area.

[0059] Preferably, the method for generating the dynamically optimized local path includes:

[0060] Taking the global path optimized by the improved dynamic adaptive cost heuristic function h′(n) as a reference, decompose the path into local target points;

[0061] Call the local dynamic obstacle avoidance planner in a complex environment, obtain the local path according to the distribution environment of dynamic obstacles and the motion state of the intelligent robot, use the multi-dimensional constraint matrix potential field optimization algorithm MDC-MPFO to evaluate the impact of local path planning, and make dynamic obstacle avoidance adjustments in time according to the evaluation results;

[0062] While avoiding obstacles dynamically, the robot is combined with the dynamic characteristics of the intelligent robot to constrain path planning and calculate the curvature of the discrete points κ. i The smoothness of the path is evaluated to ensure the smoothness and dynamic feasibility of the path.

[0063] Preferably, the method for obtaining the local path includes:

[0064] Matrix path point coordinate set P and obstacle coordinate set O:

[0065]

[0066] Where p represents the number of local path points to be evaluated, and m represents the number of obstacles;

[0067] Calculate the path point P i =(x i ,y i ) to the target point G = (x g ,y g )'s gravitational field strength F att，i :

[0068]

[0069] Among them, d goal is the gravitational range of the target point, k att is the gravitational field gain coefficient;

[0070] Based on the gravitational field strength F att，i Establish the gravitational field matrix F att :

[0071]

[0072] Among them, F att，xn represents the x-axis component of the gravitational field strength at each path point to be evaluated, F att，yn Represents the y-axis component of the gravitational field strength at each path point to be evaluated;

[0073] Calculate the path point P i =(x i ,y i ) to the nearest obstacle O j =(x oj ,y oj ) of the repulsive field strength F rep，i :

[0074]

[0075] Among them, d safe represents the safety distance, and k rep represents the repulsive force field gain coefficient, and d obs,i represents the path point P i =(x i , y i ) to the nearest distance of the obstacle O j =(x oj , y oj );

[0076] Based on the gravitational field strength F rep，i establish a repulsive force field matrix F rep :

[0077]

[0078] Among them, F rep，xp represents the x-axis component of the repulsive force field strength from each path point to be evaluated to the nearest obstacle, and F rep，yp represents the y-axis component of the repulsive force field strength from each path point to be evaluated to the nearest obstacle;

[0079] According to the superposition theorem, based on the gravitational field matrix F att and the repulsive force field matrix F rep construct a resultant force field F total :

[0080]

[0081] Update the position of the path point of the local path planning based on the resultant force field to generate the dynamically optimized local path:

[0082]

[0083] Among them, P i，new represents the updated coordinates of the local path point, and P i，current represents the coordinates of the local path point before update, represents the path update step size, and F total，i represents the resultant force field strength at each local path point.

[0084] Preferably, the method for evaluating the influence of local path planning by using the multi-dimensional constraint matrix potential field optimization algorithm MDC-MPFO includes: path feasibility evaluation and path cost value evaluation;

[0085] The method for the path feasibility evaluation includes:

[0086] If The local path planning is unreliable and needs to be replanned;

[0087] If The local path passes the feasibility assessment and drives the motion module of the intelligent robot to the local target point;

[0088] The method for evaluating the path cost value includes:

[0089] Take the sum of the magnitudes of the resultant force fields F total at each path point as J force :

[0090]

[0091] where, ‖F total，i ‖ represents the magnitude of the resultant force field F total and p represents the number of local path points to be evaluated;

[0092] Calculate the first-order smoothness:

[0093]

[0094] where, J smooth，1st represents the first-order smoothness value, P i represents the i-th local path point, and P i-1 represents the (i - 1)-th local path point;

[0095] Calculate the second-order smoothness:

[0096]

[0097] where, J smooth，2nd represents the second-order smoothness value, and P i+1 represents the (i + 1)-th local path point;

[0098] Perform weighted fusion to calculate the path planning quality evaluation cost J evaluate :

[0099] J evaluate = J force + λ 1 · J smooth，1st + λ 2 · J smooth，2nd ,

[0100] where, λ 1 represents the first-order smoothness penalty coefficient, and λ 2 represents the second-order smoothness penalty coefficient.

[0101] Preferably, the method for constrained path planning in combination with the dynamic characteristics of the intelligent robot includes:

[0102] According to the dynamic model of the intelligent robot, the local path needs to satisfy three-point constraints, including:

[0103] Velocity-angular velocity joint vector form constraint:

[0104]

[0105] where Ψ represents the velocity-angular velocity joint vector, X lim represents the velocity-angular velocity joint constraint, v represents the linear velocity of the intelligent robot, ω represents the angular velocity, v min represents the minimum linear velocity, v max represents the maximum linear velocity, ω min represents the minimum angular velocity, ω max represents the maximum angular velocity;

[0106] Linear-angular acceleration vector form constraint:

[0107]

[0108] where, represents the linear-angular acceleration joint vector, V lim represents the linear-angular acceleration joint constraint, a represents the linear acceleration, β represents the angular acceleration, a min represents the minimum linear acceleration, a max represents the maximum linear acceleration, β min represents the minimum angular acceleration, β max represents the maximum angular acceleration;

[0109] Turning radius vector form constraint:

[0110]

[0111] where R(Ψ) represents the non-linear function of the turning radius, R min represents the minimum turning radius of the robot model.

[0112] Preferably, the discrete point curvature κ i is used to evaluate the path smoothness, where the calculation method of the discrete point curvature κ i includes:

[0113] Select three consecutive local path points and

[0114] Calculate the area S of the curvature triangle by the determinant method ki :

[0115]

[0116] Calculate the point-to-point distance between three points:

[0117]

[0118] Among them, d i-1,i represents the distance between and i,i+1 represents the distance between and i-1,i+1 represents the distance between and;

[0119] Take the maximum pairwise distance among the three points as the total chord length L:

[0120] L = max{d i-1，i , d i，i+1 , d i-1，i+1};

[0121] Substitute and solve for the discrete point curvature κ i :

[0122]

[0123] When the local path points are dense, simplify the discrete point curvature κ i to:

[0124]

[0125] Preferably, the method for performing feedback fusion includes:

[0126] Based on the local target points of the global path decomposition optimized by the dynamic adaptive cost heuristic function h′(n), update the initial position of the intelligent robot after each local target point is completed;

[0127] When the local path deviates from the global path by more than a set threshold, the multi-dimensional feasibility dynamic evaluation module under the MDC-AFI framework determines whether to trigger the global path planner to re-optimize the global path, and dynamically updates the data fed back from the global path planner to the local path planning module;

[0128] Set the path switching priority. When the local path is severely affected by dynamic obstacles, use the global path points as the priority reference points to ensure the global optimality and safety of the path planning and prevent the robot from falling into local optimality.

[0129] Preferably, the method for determining whether to trigger the global path planner to re-optimize the global path includes:

[0130] Calculate the Euclidean distance and point projection distance between the local path points and the global path points;

[0131] Euclidean distance δ euclidean is as follows:

[0132]

[0133] wherein, is the point coordinate of the local path, is the point coordinate of the global path;

[0134] Point projection distance δ projection is as follows:

[0135]

[0136] wherein, and represent two endpoints of a certain section in the global path;

[0137] According to the complexity of the planned scenario, combining the Euclidean distance δ euclidean and the point projection distance δ euclidean , weighted fusion is performed according to the weight parameters σ and ψ to obtain comprehensive deviation data:

[0138] δ fused-linear = σδ euclidean + ψδ projection

[0139]

[0140] wherein, σ + ψ = 1, δ fused-linear represents the linear fusion deviation data, δ fused-nonliner represents the non - linear fusion deviation data, σ represents the fusion weight of the Euclidean distance, and ψ represents the fusion weight of the point projection distance;

[0141] When it is detected that the fusion deviation data δ fused-linear and δ fused-nonliner are greater than the deviation threshold δ max , the multi - dimensional feasibility dynamic evaluator in the MDC - AFI framework triggers the global path planning to re - optimize the global path, fits the feedback data of the local path planner to obtain a more reliable global path, and then uses the obtained global path as the input of the local path planner for continuous iterative optimization.

[0142] The present invention also provides an MDC - AFI adaptive feedback fusion intelligent path planning system based on multi - dimensional collaborative enhancement. The system applies the method described in any one of the above, and is characterized in that it includes: an initial information acquisition module, a global path optimization module, a local path optimization module, and a feedback fusion module;

[0143] The initial information acquisition module is used to acquire the initial state information of the intelligent robot, and the initial state information includes: initial coordinates, attitude information, obstacle distribution layer information, and target point coordinates;

[0144] Based on the initial state information, the global path optimization module calls a dynamic adaptive global path planner to search for a global path, iteratively calculates global path nodes, and generates a globally optimized path;

[0145] The local path optimization module is used to decompose the globally optimized path into several local target points and use them as the initial reference input of the local path planner. Through the multi-dimensional feasibility dynamic evaluator under the MDC-AFI framework and the distribution information of dynamic obstacles in the local environment, it adaptively adjusts the parameters of the local path planner. According to the dynamic constraints of the intelligent robot and the real-time environment information, it iteratively generates a dynamically optimized local path;

[0146] When the degree of deviation of the local path from the globally optimized path exceeds the preset tolerance range, the feedback fusion module determines whether it is necessary to re-optimize the global path through the multi-dimensional feasibility dynamic evaluator, and feeds the latest global path data back to the local path planner for feedback fusion until the intelligent robot reaches the target point and completes the target navigation task.

[0147] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0148] (1) To solve the problems of low path planning efficiency and poor safety in complex dynamic environments, the present invention uses multi-dimensional collaborative enhancement technology and an adaptive feedback mechanism to adjust the global and local modules of path planning in real time, and can effectively generate efficient and safe paths in complex dynamic environments. Especially in areas with dense obstacles, it still has high path feasibility and planning efficiency, realizing efficient and high-performance path planning;

[0149] (2) Aiming at the problem that the path dynamic adjustment in areas with dense obstacles falls into local optima, the present invention proposes a multi-dimensional feature fusion strategy to perceive and process dynamic environment change parameters in real time, and can comprehensively optimize the dynamic constraints of multi-objective paths, ensuring the dynamic balance between global and local goals;

[0150] (3) To overcome the problems of insufficient path continuity and stability caused by the limitations of the cost function in complex dynamic environments, the present invention introduces a dynamic adaptive cost heuristic function to achieve efficient feedback adjustment and reconstruction between global and local path planning, and can perform high-precision correction on path deviations in dynamic environments, improving the smoothness and robustness of the path;

[0151] (4) The present invention proposes a multi-dimensional constraint matrix potential field optimization algorithm to solve the problem that the path planner fails to plan due to low accuracy in identifying obstacles or lag in path adjustment, and can still effectively identify and avoid obstacles in dense obstacle areas, significantly enhancing the obstacle avoidance performance of autonomous navigation. BRIEF DESCRIPTION OF THE DRAWINGS

[0152] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0153] Figure 1 It is a schematic flowchart of the method according to an embodiment of the present invention;

[0154] Figure 2 It is a block diagram of the method according to an embodiment of the present invention;

[0155] Figure 3 It is a comparison diagram of resolving the initial data collected from the IMU using quaternions or rotation matrices and directly integrating the gyroscope data according to an embodiment of the present invention. Among them, a is the variation of the roll angle with time in a conventional scenario, b is the variation of the roll angle with time in a Gimbal Lock scenario, c is the variation of the pitch angle with time in a conventional scenario, d is the variation of the pitch angle with time in a Gimbal Lock scenario, e is the variation of the yaw angle with time in a conventional scenario, and f is the variation of the yaw angle with time in a Gimbal Lock scenario;

[0156] Figure 4 It is a schematic diagram of the multi-dimensional constraint matrix potential field optimization algorithm evaluating and adjusting local path planning according to an embodiment of the present invention. Among them, a is the overall effect of global path planning and optimization of the transportation task, b is the overall effect of dynamically avoiding dynamic obstacles on the left side of the cargo hold during local path tracking, c is the overall effect of dynamically avoiding dynamic obstacles on the right side of the cargo hold during local path tracking, and d is the overall effect of local path tracking of the global optimized path and dynamic obstacle avoidance;

[0157] Figure 5 It is a comparison diagram of calculating the path steering change amount by introducing the Kalman filter algorithm according to an embodiment of the present invention and other algorithms (such as the weighted average method and the simple average method);

[0158] Figure 6 It is a comparison diagram of evaluating path smoothness using discrete point curvature and the smoothness analysis method based on the integral of the second derivative according to an embodiment of the present invention. Among them, a is a comparison diagram of the effects of three filtering algorithms for obtaining the path steering angle, and b is a comparison diagram of the errors of three filtering algorithms for obtaining the path steering angle. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0159] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0160] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0161] Embodiment 1

[0162] In this embodiment, as Figure 1 、 Figure 2 shown, a multi-dimensional collaborative enhanced MDC-AFI adaptive feedback fusion intelligent path planning method includes the following steps:

[0163] S1. Obtain the initial state information of the intelligent robot, where the initial state information includes: initial coordinates, attitude information, obstacle distribution layer information, and target point coordinates.

[0164] In S1, the method for obtaining the attitude information includes: the formula for calculating the quaternion using the accelerometer and magnetometer data in the IMU:

[0165]

[0166] where θ represents the rotation angle around the unit vector u = [u X ,u Y ,u Z , w represents the real part of the quaternion, and x, y, z represent the imaginary parts of the quaternion, respectively representing the components of rotation around the X, Y, and Z axes; substituting the initial attitude into the quaternion formula, the initial quaternion q 0 = [w 0 ,x 0 ,y 0 ,z 0 is obtained;

[0167] Calculate the angular increment Δq through the gyroscope angular velocity, update the quaternion, and obtain the updated quaternion:

[0168]

[0169] where t 0 represents the initial time, and Δt represents the update time step;

[0170] Solve the attitude angle using the updated quaternion q′ = [w′, x′, y′, z′] to obtain the attitude information:

[0171]

[0172] Pitch = sin -1 (2(w′y′ - x′z′)), |Pitch| ≤ 90°,

[0173]

[0174] where Roll represents the roll angle of rotation about the X-axis, Pitch represents the pitch angle of rotation about the Y-axis, and Yaw represents the yaw angle of rotation about the Z-axis;

[0175] Calculate the rotation matrix R through the quaternion q = [w, x, y, z]:

[0176]

[0177] Calculate the corresponding rotation matrix R based on the attitude information of each sensor ε ;

[0178] Map the local coordinate systems of each sensor to the global coordinate system to provide a unified benchmark for subsequent algorithms and ensure the correct transfer and processing of data between different coordinate systems:

[0179]

[0180] where ε represents the index of each sensor in the sensor group;

[0181] A reversible transformation, that is, converting the global coordinates to local coordinates:

[0182]

[0183] where represents the local coordinates of the sensor with index ε.

[0184] S2. Based on the initial state information, call the dynamic adaptive global path planner to perform iterative calculations to optimize the global path, obtain the global path nodes, and generate the globally optimized path.

[0185] The methods for optimizing the global path include:

[0186] Introduce the dynamic adaptive cost heuristic function h′(n), combine multi-dimensional information such as the distribution density of obstacles, the velocity direction of dynamic obstacles, and the Euclidean distance to the target point, and optimize the path search efficiency.

[0187] Methods for optimizing path search efficiency include: calculating the time cost h time (n), which is the estimated shortest time from node n to the target point G:

[0188]

[0189] where ||n - g|| represents the Euclidean distance from node n to the target point G, and v represents the linear velocity of the intelligent robot;

[0190] Calculating the spatial distance cost h distance (n), which is the Euclidean distance from node n to the target point G:

[0191]

[0192] where (x n , y n ) represents the coordinates of node n, and (x g , y g ) represents the coordinates of the target point G;

[0193] Calculating the dynamic environment cost h dynamic (n), which is the impact of dynamic obstacles on path planning:

[0194]

[0195] where k represents the number of dynamic obstacles, d i represents the distance from node n to the i-th dynamic obstacle, and θ i represents the angle between the direction from node n to the target point G and the direction of the velocity of the dynamic obstacle;

[0196] Based on the time cost h time (n), the spatial distance cost h distance (n), and the dynamic environment cost h dynamic (n), construct a dynamic adaptive cost heuristic function h′(n):

[0197] h′(n) = αh time (n) + βh distance (n) + γh dynamic (n),

[0198] where α, β, and γ respectively represent the weight parameters of h time (n), h distance (n), h dynamic (n), and α + β + γ = 1;

[0199] Express the dynamic adaptive cost heuristic function h′(n) as the product of a weight vector and a cost vector:

[0200] h′(n) = W·H(n),

[0201] where W = [α, β, γ] represents the weight vector, and h(n) = [h time (n), h distance (n), h dynamic (n)] T represents the cost vector;

[0202] Define the dynamic environment parameter vector and dynamically and adaptively adjust the weight vector W according to the environment parameter vector E:

[0203] W = W 0 + K·E(t),

[0204]

[0205] E(t) = [ρ obs , d obs , c env ,

[0206]

[0207] where W 0 represents the initial weight vector, K represents the gain matrix, ρ obs represents the dynamic obstacle density, d obs represents the closest distance between the node and the dynamic obstacle, c env represents the openness of the environment, represents the dynamic density function, represents the area of the environmental area p = [x, y] T represents the two-dimensional coordinates of the dynamic obstacle, represents the two-dimensional real number space, i represents the index of the dynamic obstacle set, P node represents the coordinate of the node position to be evaluated, v i represents the linear velocity vector of the dynamic obstacle with index i at time t, P i (t) represents the coordinate of the dynamic obstacle with index i at time t, represents the obstacle set, ||·|| 2 represents the Euclidean norm, ρ max represents the maximum possible obstacle density within the unit area.

[0208] Adopt a path smoothing algorithm (such as Bezier curve fitting, cubic spline interpolation, etc.) to process the initial global path for the objective function, reduce the path broken lines, and improve the smoothness and controllability of the path.

[0209] By using hierarchical grid modeling, a higher weight is set for high-complexity regions and a lower weight is set for low-complexity regions, improving the globality and efficiency of path search.

[0210] S3. Decompose the globally optimized path to obtain several local target points, and use these local target points as the initial reference input for the local path planner. Through the multi-dimensional feasibility dynamic evaluator under the MDC-AFI framework and the distribution information of dynamic obstacles in the local environment, adaptively adjust the parameters of the local path planner, and iteratively generate a dynamically optimized local path according to the dynamic constraints of the intelligent robot and the real-time environmental information.

[0211] The method for generating a dynamically optimized local path includes:

[0212] Taking the globally optimized path optimized by the improved dynamic adaptive cost heuristic function h′(n) as a reference, decompose the path into local target points.

[0213] Call the local dynamic obstacle avoidance planner in a complex environment to obtain a local path according to the distribution environment of dynamic obstacles and the motion state of the intelligent robot. Use the multi-dimensional constraint matrix potential field optimization algorithm MDC-MPFO to evaluate the impact of local path planning, and make dynamic obstacle avoidance adjustments in a timely manner according to the evaluation results.

[0214] The method for obtaining a local path includes: matrixizing the path point coordinate set P and the obstacle coordinate set O:

[0215]

[0216] where p represents the number of local path points to be evaluated, and m represents the number of obstacles;

[0217] Calculate the gravitational field strength F i =(x i , y i ) of the path point P to the target point G=(x g , y g ): att，i :

[0218]

[0219] where d goal is the gravitational range of the target point, and k att is the gravitational field gain coefficient;

[0220] Based on the gravitational field strength F att，i establish the gravitational field matrix F att :

[0221]

[0222] where Fatt，xn Represents the x-axis component of the gravitational field strength at each path point to be evaluated, F att，yn Represents the y-axis component of the gravitational field strength at each path point to be evaluated;

[0223] Calculate the repulsive force field strength F of the path point P i =(x i , y i ) to the nearest obstacle O j =(x oj , y oj ): rep，i :

[0224]

[0225] Where d safe Represents the safety distance, k rep Represents the repulsive force field gain coefficient, d obs,i Represents the path point P i =(x i , y i ) to the obstacle O j =(x oj , y oj )'s nearest distance;

[0226] Based on the gravitational field strength F rep，i Establish a repulsive force field matrix F rep :

[0227]

[0228] Where F rep，xp Represents the x-axis component of the repulsive force field strength from each path point to be evaluated to the nearest obstacle, F rep，yp Represents the y-axis component of the repulsive force field strength from each path point to be evaluated to the nearest obstacle;

[0229] According to the superposition theorem, based on the gravitational field matrix F att And the repulsive force field matrix F rep Construct the resultant force field F total :

[0230]

[0231] Based on the resultant force field, update the position of the path point of the local path planning to generate the dynamically optimized local path:

[0232]

[0233] Where P i，new Represents the updated coordinates of the local path point, P i，current Represents the coordinates of the local path point before update, represents the path update step size, F total，i represents the resultant force field intensity at each local path point

[0234] A method for evaluating the impact of local path planning using the multi-dimensional constraint matrix potential field optimization algorithm MDC-MPFO includes: path feasibility evaluation and path cost value evaluation;

[0235] If the local path planning is unreliable, re-planning is required;

[0236] If the local path passes the feasibility evaluation, drive the motion module of the intelligent robot to the local target point;

[0237] The method for evaluating the path cost value includes:

[0238] Take the sum of the magnitudes of the resultant force field F total at each path point as J force :

[0239]

[0240] where ||F total，i || represents the magnitude of the resultant force field F total and p represents the number of local path points to be evaluated;

[0241] Calculate the first-order smoothness:

[0242]

[0243] where J smooth，1st represents the first-order smoothness value, P i represents the i-th local path point, P i-1 represents the (i - 1)-th local path point;

[0244] Calculate the second-order smoothness:

[0245]

[0246] where J smooth，2nd represents the second-order smoothness value, P i+1 represents the (i + 1)-th local path point;

[0247] Perform weighted fusion to calculate the path planning quality evaluation cost J evaluate :

[0248] J evaluate = J force + λ 1 · J smooth，1st + λ 2 · J sniith，2nd ,

[0249] Among them, λ 1 represents the first-order smoothness penalty coefficient, and λ 2 represents the second-order smoothness penalty coefficient.

[0250] While dynamically avoiding obstacles, combining the dynamic characteristics of the intelligent robot, constraining path planning, and evaluating the smoothness of the path through the discrete point curvature κ i to ensure the smoothness and dynamic feasibility of the path.

[0251] The method of constraining path planning by combining the dynamic characteristics of the intelligent robot includes: According to the dynamic model of the intelligent robot, the local path needs to satisfy three-point constraints, including:

[0252] Velocity-angular velocity joint vector form constraint:

[0253]

[0254] Among them, Ψ represents the velocity-angular velocity joint vector, and X lim represents the velocity-angular velocity joint constraint, v represents the linear velocity of the intelligent robot, ω represents the angular velocity, v min represents the minimum linear velocity, v max represents the maximum linear velocity, ω min represents the minimum angular velocity, ω max represents the maximum angular velocity;

[0255] Linear-angular acceleration vector form constraint:

[0256]

[0257] Among them, represents the linear-angular acceleration joint vector, V lim represents the linear-angular acceleration joint constraint, a represents the linear acceleration, β represents the angular acceleration, a min represents the minimum linear acceleration, a max represents the maximum linear acceleration, β min represents the minimum angular acceleration, β max represents the maximum angular acceleration;

[0258] Turning radius vector form constraint:

[0259]

[0260] Among them, R(Ψ) represents the non-linear function of the turning radius, and R min represents the minimum turning radius of the robot model.

[0261] Evaluating the path smoothness using the discrete point curvature κ i where the discrete point curvature κi The calculation method includes:

[0262] Select three consecutive local path points and

[0263] Calculate the area S of the curvature triangle by the determinant method κi :

[0264]

[0265] Calculate the point-to-point distance between three points:

[0266]

[0267] where d i-1,i represents and the distance between, d i,i+1 represents and the distance between, d i-1,i+1 represents and the distance between;

[0268] Take the maximum point-to-point distance among the three points as the total chord length L:

[0269] L = max{d i-1，i , d i，i+1 , d i-1，i+1};

[0270] Substitute and solve for the discrete point curvature κ i :

[0271]

[0272] When the local path points are dense, simplify the discrete point curvature κ i to:

[0273]

[0274] S4. When the degree of deviation of the local path from the global optimization path exceeds the preset tolerance range, the multi-dimensional feasibility dynamic evaluator will determine whether it is necessary to re-optimize the global path and feedback the latest global path data to the local path planner for feedback fusion.

[0275] The method for feedback fusion includes:

[0276] The local target points obtained by decomposing the global path optimized based on the dynamic adaptive cost heuristic function h′(n). After each local target point is completed, update the initial position of the intelligent robot.

[0277] When the local path deviates from the global path by more than a set threshold, the multi-dimensional feasibility dynamic evaluation module under the MDC-AFI framework determines whether to trigger the global path planner to re-optimize the global path and dynamically updates the data fed back from the global path planner to the local path planning module.

[0278] The method for determining whether to trigger the global path planner to re-optimize the global path includes: calculating the Euclidean distance and the point projection distance between the local path point and the global path point;

[0279] The Euclidean distance δ euclidean is:

[0280]

[0281] where, is the point coordinate of the local path, is the point coordinate of the global path;

[0282] The point projection distance δ projection is:

[0283]

[0284] where, and represent the two endpoints of a certain section in the global path;

[0285] According to the complexity of the planning scenario, combining the Euclidean distance δ euclidean and the point projection distance δ euclidean , a comprehensive deviation data is obtained by weighted fusion with weight parameters σ and ψ:

[0286] δ fused-linear = σδ euclidean + ψδ projection ,

[0287]

[0288] where, σ + ψ = 1, δ fused-linear represents the linearly fused deviation data, δ fused-nonliner represents the non-linearly fused deviation data, σ represents the fusion weight of the Euclidean distance, and ψ represents the fusion weight of the point projection distance; calculate the path turning change amount Δθ local and Δθ global to dynamically and adaptively adjust σ and ψ: The vector form of the local path point state is including the position and direction angle information of the point; The control input vector is including the linear velocity, angular velocity, and acceleration information; Establish a state transition equation to describe how the local path point transfers from the (j - 1)th moment to the jth moment:

[0289]

[0290] Among them, A represents the state transition matrix, which is used to describe the influence of steering changes on the direction angle; B represents the control input matrix, which describes the influence of the input on the system state; v j represents the speed, and Δt represents the time step; w j represents the process noise, which follows the Gaussian distribution w j ~N(0, Q), where Q represents the process noise covariance, and θ j represents the direction angle at time j; an observation equation is established to describe how to calculate the state from the observation point:

[0291]

[0292] Among them, represents the state observation value at time j, and H represents the observation matrix: H = [0 0 1], which represents the angular information of the observed path point; l j represents the observation noise, which follows the Gaussian distribution l j ~N(0, γ), where γ is the observation noise covariance. Predict the direction angle at the next moment according to the state transition equation:

[0293]

[0294] Among them, represents the predicted direction angle, and θ j-1 represents the direction angle at time j - 1; predict the error covariance:

[0295]

[0296] Among them, P j-1 represents the error covariance at time j - 1; calculate the Kalman gain K j and update the direction angle using the Kalman filter:

[0297]

[0298] Among them, θ j represents the direction angle at time j; update the error covariance:

[0299] P j =(I - K j H)P j|j-1 ,

[0300] Among them, P j represents the updated error covariance, I represents the identity matrix, and P j|j-1 represents the predicted error covariance; repeat the above process to calculate the path steering change amount Δθ local and Δθglobal :

[0301]

[0302] Among them, represents the local direction angle at time j, represents the local direction angle at time j-1, represents the global direction angle at time k, represents the global direction angle at time j-1. Dynamically and adaptively adjust the fusion weights σ and ψ according to the environmental characteristics; when the local path changes violently, that is, when the turning change amount Δθ of the local path loacal is large, increase the weight ψ of the point projection distance δ projection ; when the local path is relatively smooth, that is, when the turning change amount Δθ of the local path global is small, increase the weight σ of the Euclidean distance δ euclidean . Dynamically adjust the weights σ and ψ:

[0303]

[0304] ψ = 1 - σ,

[0305] Fuse the deviation data δ fused-linear or δ fused-nonliner changes dynamically with the weights σ and ψ:

[0306] δ fused-(non)linear = f(σ, ψ, δ projection , δ euclidean ),

[0307] where f is a function of the variables σ, ψ, δ projection , δ euclidean .

[0308] If δ fused-(non)linear > δ max : The multi-dimensional feasibility dynamic evaluation module in the MDC-AFI framework triggers the global path planner to re-optimize the global path, fits the feedback data of the local path planner to obtain a more reliable global path, and then uses it as the input of the local path planner, and continuously iterates and optimizes.

[0309] Set the path switching priority. When the local path is severely affected by dynamic obstacles, use the global path points as the priority reference points to ensure the global optimality and safety of the path planning and prevent the robot from falling into local optimality.

[0310] S5. Repeat S1~S4. Through the multi-dimensional collaborative enhanced path feedback mechanism and the adaptive fusion dynamic optimization strategy, iteratively generate the real-time updated local path and global path until the autonomous mobile robot reaches the target shelf and completes the goods picking and transportation tasks.

[0311] As Figure 3 shown, the direct integration method may introduce cumulative errors in numerical calculations, especially during long-term integration processes. Quaternions, due to their mathematical properties (the normality of unit quaternions), can maintain high numerical stability and are not prone to drift problems. As Figure 3 shown in the left three figures from top to bottom in, the roll angle, pitch angle, and yaw angle of direct integration (solid line) and quaternion solution (dashed line) under normal scenarios change with time; when using direct integration, if Euler angles are used to represent rotation, it is easy to encounter the Gimbal Lock problem, that is, in certain specific postures, two rotation axes will coincide, resulting in the loss of a degree of freedom and the inability to perform correct rotation representation. Quaternions avoid this problem because they do not use Euler angles but represent rotation in the form of four-dimensional vectors and can always represent any three-dimensional rotation. As Figure 3 shown in the right three figures from top to bottom in, the roll angle, pitch angle, and yaw angle of direct integration (solid line) and quaternion solution (dashed line) under the Gimbal Lock scenario change with time.

[0312] As Figure 4 shown, when an autonomous mobile robot performs goods picking and transportation tasks between different warehouses, it evaluates and adjusts the local path planning through the multi-dimensional constraint matrix potential field optimization algorithm to effectively avoid dynamic obstacles. The four sub-figures sequentially show the global path planning and optimization of the transportation task, the dynamic avoidance of the dynamic obstacle on the left side of the warehouse during local path tracking, the dynamic avoidance of the dynamic obstacle on the right side of the warehouse during local path tracking, and the overall effect of global optimization of the local path tracking and dynamic obstacle avoidance in a clockwise direction.

[0313] As Figure 5 shown, introducing the Kalman filter algorithm to calculate the path steering change amount can more accurately fit the angle data transmitted by the autonomous mobile robot, and the error level is at a lower level compared to the other two algorithms, that is, the error lines in the figure are all below the other two algorithms.

[0314] As Figure 6 shown, compared with the smoothness analysis method based on the integral of the second derivative, using the discrete point curvature to evaluate the path smoothness makes the evaluation level fluctuate near the mean value of smoothness, with a smaller error range and the evaluation level closer to the theoretical value.

[0315] Embodiment 2

[0316] In this embodiment, an MDC-AFI adaptive feedback fusion intelligent path planning system based on multi-dimensional collaboration enhancement includes: an initial information acquisition module, a global path optimization module, a local path optimization module, and a feedback fusion module.

[0317] The initial information acquisition module is used to acquire the initial state information of the intelligent robot. The initial state information includes: initial coordinates, attitude information, obstacle distribution layer information, and target point coordinates.

[0318] Based on the initial state information, the global path optimization module calls the dynamic adaptive global path planner to perform iterative calculations to optimize the global path, obtain global path nodes, and generate a globally optimized path.

[0319] The local path optimization module is used to decompose the globally optimized path to obtain several local target points, and use these several local target points as the initial reference input of the local path planner. Through the multi-dimensional feasibility dynamic evaluator in the MDC-AFI framework and the distribution information of dynamic obstacles in the local environment, the parameters of the local path planner are adaptively adjusted. According to the dynamic constraints of the intelligent robot and the real-time environment information, a dynamically optimized local path is iteratively generated.

[0320] When the degree of deviation of the local path from the globally optimized path exceeds the preset tolerance range, the feedback fusion module determines whether it is necessary to re-optimize the global path through the multi-dimensional feasibility dynamic evaluator, and feeds back the latest global path data to the local path planner for feedback fusion until the intelligent robot reaches the target point and completes the target navigation task.

[0321] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A MDC-AFI adaptive feedback fusion intelligent path planning method based on multi-dimensional collaborative enhancement, characterized in that: The steps include: S1. Obtaining the initial state information of the intelligent robot, the initial state information includes: initial coordinates, posture information, obstacle distribution layer information and target point coordinates; S2. Based on the initial state information, a dynamic adaptive global path planner is called to search for a global path, and global path nodes are obtained by iterative calculation to generate a global optimization path; S3. Decomposing the global optimization path to obtain a number of local target points and using them as initial reference inputs for the local path planner, adaptively adjusting the parameters of the local path planner through the multi-dimensional feasibility dynamic evaluator under the MDC-AFI framework and the distribution information of dynamic obstacles in the local environment, and iteratively generating a dynamically optimized local path according to the dynamic constraints of the intelligent robot and the real-time environmental information; S4. When the local path deviates from the global optimization path to an extent exceeding a preset tolerance range, the multi-dimensional feasibility dynamic evaluator will determine whether the global path needs to be re-optimized, and feed back the latest global path data to the local path planner for feedback fusion; S5. Repeat S1 to S4 until the intelligent robot reaches the target point and completes the target navigation task.

2. According to claim 1, a MDC-AFI adaptive feedback fusion intelligent path planning method based on multi-dimensional collaborative enhancement is characterized in that: In S1, the method for obtaining the posture information includes: The formula for calculating quaternion using accelerometer and magnetometer data in IMU: Where θ represents the rotation around the unit vector u=[u X ,u Y ,u Z ], w represents the real part of the quaternion, x, y, z represent the imaginary part of the quaternion, and represent the components of rotation around the X, Y, and Z axes respectively; Substitute the initial posture into the formula of the value quaternion to obtain the initial quaternion q0 = [w0, x0, y0, z0]; The angular increment Δq is calculated by the gyroscope angular velocity, and the quaternion is updated to obtain the updated quaternion: Among them, t0 represents the initial time, Δt represents the update time step; The attitude angle is calculated using the updated quaternion q′=[w′, x′, y′, z′] to obtain the attitude information: Pitch=sin -1 (2(w′y'-x'z′)),|Pitch|≤90°, Among them, Roll represents the roll angle around the X-axis, Pitch represents the pitch angle around the Y-axis, and Yaw represents the heading angle around the Z-axis. The rotation matrix R is calculated by quaternion q = [w, x, y, z]: According to the posture information of each sensor, the corresponding rotation matrix R is calculated ε ; The local coordinate system of each sensor Mapping to the global coordinate system Provide a unified benchmark for subsequent algorithms to ensure that data is correctly transferred and processed between different coordinate systems: Among them, ε represents the index of each sensor in the sensor group; Reversible transformation, that is, converting global coordinates to local coordinates: in, represents the local coordinates of the sensor with index ε.

3. According to claim 1, a MDC-AFI adaptive feedback fusion intelligent path planning method based on multi-dimensional collaborative enhancement is characterized in that: The method for optimizing the global path comprises: The dynamic adaptive cost heuristic function h′(n) is introduced to optimize the path search efficiency by combining multi-dimensional information such as obstacle distribution density, dynamic obstacle speed direction and Euclidean distance between the target point. The path smoothing algorithm is used to optimize the objective function to process the initial global path, reduce path folds, and improve the smoothness and controllability of the path; By utilizing hierarchical grid modeling, higher weights are set for high-complexity areas and lower weights are set for low-complexity areas, thus improving the globality and efficiency of path search.

4. According to claim 3, the MDC-AFI adaptive feedback fusion intelligent path planning method based on multi-dimensional collaborative enhancement is characterized in that: The method for optimizing path search efficiency comprises: Computation time cost h time (n), which is the shortest time estimate from node n to target point G: Among them, ||ng|| represents the Euclidean distance from node n to target point G, and v represents the linear velocity of the intelligent robot; Calculate the spatial distance cost h distance (n), that is, the Euclidean distance from node n to the target point G: Among them, (x n ,y n ) represents the coordinates of node n, (x g ,y g ) represents the coordinates of the target point G; Calculate the dynamic environment cost h dynamic (n), that is, the impact of dynamic obstacles on path planning: Where k represents the number of dynamic obstacles, d i represents the distance from node n to the ith dynamic obstacle, θ i Represents the angle between the direction from node n to target point G and the speed direction of the dynamic obstacle; Based on the time cost h time (n), the spatial distance cost h distance (n) and the dynamic environment cost h dynamic (n) Construct a dynamic adaptive cost heuristic function h′(n): h′(n)=αh time (n)+βh distance (n)+γh dynamic (n), Among them, α, β, and γ represent h time (n), h distance (n), h dynamin (n) weight parameter, and α+β+γ=1; The dynamic adaptive cost heuristic function h′(n) is expressed as the product of the weight vector and the cost vector: h′(n)=W·H(n), Where W = [α, β, γ] represents the weight vector, H(n) = [h time (n), h distance (n), h dynamic (n)] T represents the cost vector; Define a dynamic environment parameter vector, and dynamically and adaptively adjust the weight vector W according to the environment parameter vector E(t): W=W0+K·E(t), E(t)=[ρ obs ,d obs ,c env ], Among them, W0 represents the initial weight vector, K represents the gain matrix, ρ obs represents the dynamic obstacle density, d obs represents the shortest distance between the node and the dynamic obstacle, c env Indicates the openness of the environment. represents the dynamic density function, Indicates environmental area The area of ​​​​the T represents the two-dimensional coordinates of the dynamic obstacle, represents a two-dimensional real space, i represents the index of the dynamic obstacle set, P node Indicates the node position coordinates to be evaluated, v i represents the linear velocity vector of the dynamic obstacle with index i at time t, P i (t) represents the coordinates of the dynamic obstacle with index i at time t, represents the obstacle set, ||·||2 represents the Euclidean norm, ρ max Indicates the maximum obstacle density that can be achieved in a unit area.

5. According to claim 3, the MDC-AFI adaptive feedback fusion intelligent path planning method based on multi-dimensional collaborative enhancement is characterized in that: The method for generating the dynamically optimized local path includes: Taking the global path optimized by the improved dynamic adaptive cost heuristic function h′(n) as a reference, the path is decomposed into local target points; Call the local dynamic obstacle avoidance planner in a complex environment, obtain the local path according to the distribution environment of dynamic obstacles and the motion state of the intelligent robot, use the multi-dimensional constraint matrix potential field optimization algorithm MDC-MPFO to evaluate the impact of local path planning, and make dynamic obstacle avoidance adjustments in time according to the evaluation results; While avoiding obstacles dynamically, the robot is combined with the dynamic characteristics of the intelligent robot to constrain path planning and calculate the curvature of the discrete points κ. i The smoothness of the path is evaluated to ensure the smoothness and dynamic feasibility of the path.

6. According to claim 5, a method for MDC-AFI adaptive feedback fusion intelligent path planning based on multi-dimensional collaborative enhancement is characterized in that: The method of obtaining the local path includes: Matrix path point coordinate set P and obstacle coordinate set O: Where p represents the number of local path points to be evaluated, and m represents the number of obstacles; Calculate the path point P i =(x i ,y i ) to the target point G = (x g ,y g )'s gravitational field strength F att，i : Among them, d goal is the gravitational range of the target point, k att is the gravitational field gain coefficient; Based on the gravitational field strength F att，i Establish the gravitational field matrix F att : Among them, F att，xn represents the x-axis component of the gravitational field strength at each path point to be evaluated, F att，yn Represents the y-axis component of the gravitational field strength at each path point to be evaluated; Calculate the path point P i =(x i ,y i ) to the nearest obstacle O j =(x oj ,y oj ) of the repulsive field strength F rep，i : Among them, d safe Indicates the safety distance, k rep represents the repulsive field gain coefficient, d obs,i Represents the path point P i =(x i ,y i ) to obstacle O j =(x oj ,y oj )’s closest distance; Based on the gravitational field strength F rep，i Establish the repulsive field matrix F rep : Among them, F rep，xp Represents the x-axis component of the repulsive field strength from each path point to be evaluated to the nearest obstacle, F rep，yp Represents the y-axis component of the repulsive field strength from each path point to be evaluated to the nearest obstacle; According to the superposition theorem, based on the gravitational field matrix F att and the repulsive field matrix F rep Constructing the force field F total : Based on the resultant force field, the path point positions of the local path planning are updated to generate the dynamically optimized local path: Among them, P i，new represents the updated coordinates of the local path point, P i，current Represents the coordinates of the local path point before updating, represents the path update step length, F total，i Represents the resultant force field strength at each local path point.

7. The MDC-AFI adaptive feedback fusion intelligent path planning method based on multi-dimensional collaborative enhancement according to claim 6 is characterized in that: The method of using the multi-dimensional constraint matrix potential field optimization algorithm MDC-MPFO to evaluate the impact of local path planning includes: path feasibility evaluation and path cost value evaluation; The method for evaluating the path feasibility includes: like The local path planning is unreliable and needs to be replanned; like The local path passes the feasibility assessment and drives the intelligent robot motion module to go to the local target point; The method for evaluating the path cost includes: Take the resultant force field F at each path point total The sum of the modulus lengths is J force : Among them, ||F total，i || represents the resultant force field F total The modulus of p represents the number of local path points to be evaluated; Compute first-order smoothness: Among them, J smooth，1st represents the first-order smoothness value, P i represents the i-th local path point, P i-1 represents the i-1th local path point; Compute second-order smoothness: Among them, J smooth，2nd represents the second-order smoothness value, P i+1 represents the i+1th local path point; Weighted fusion, calculate the path planning quality evaluation cost J evaluate : J evaluate =J force +λ1·J smooth，1st +λ2·J smooth，2nd , Among them, λ1 represents the first-order smoothness penalty coefficient, and λ2 represents the second-order smoothness penalty coefficient.

8. The MDC-AFI adaptive feedback fusion intelligent path planning method based on multi-dimensional collaborative enhancement according to claim 7 is characterized in that: The methods for constrained path planning combined with the dynamic characteristics of the intelligent robot include: According to the dynamic model of the intelligent robot, the local path needs to satisfy three constraints, including: Constraints in the form of joint vectors of velocity and angular velocity: Where Ψ represents the velocity and angular velocity joint vector, X lim represents the velocity-angular velocity joint constraint, v represents the linear velocity of the intelligent robot, ω represents the angular velocity, and v min Indicates the minimum linear velocity, v max represents the maximum linear velocity, ω min represents the minimum angular velocity, ω max represents the maximum angular velocity; Linear angular acceleration vector form constraint: in, Represents the linear angular acceleration joint vector, V lim represents the linear and angular acceleration joint constraint, a represents the linear acceleration, β represents the angular acceleration, a min represents the minimum linear acceleration, a max represents the maximum linear acceleration, β min represents the minimum angular acceleration, β max represents the maximum angular acceleration; Turning radius vector form constraint: Where R(Ψ) represents the nonlinear function of the turning radius, R min Indicates the minimum turning radius of the robot model.

9. The MDC-AFI adaptive feedback fusion intelligent path planning method based on multi-dimensional collaborative enhancement according to claim 8 is characterized in that: Using the discrete point curvature κ i The path smoothness is evaluated, where the discrete point curvature κ i The calculation methods include: Select three consecutive local path points and Calculate the area S of the curvature triangle by the determinant method κi : Compute the pairwise distance between three points: Among them, d i-1,i express and The distance between i,i+1 express and The distance between i-1,i+1 express and The distance between Take the maximum point-to-point distance between the three points as the total chord length L: L=max{d i-1，i ,d i，i+1 ,d i-1，i+1 }; Substitute into the solution for the discrete point curvature κ i : When the local path points are dense, the discrete point curvature κ i Simplified to:

10. The MDC-AFI adaptive feedback fusion intelligent path planning method based on multi-dimensional collaborative enhancement according to claim 3 is characterized in that: Methods for performing feedback fusion include: Based on the local target points of the global path decomposition optimized by the dynamic adaptive cost heuristic function h′(n), the initial position of the intelligent robot is updated after each local target point is completed; When the local path deviates from the global path by more than a set threshold, the multi-dimensional feasibility dynamic evaluation module under the MDC-AFI framework determines whether to trigger the global path planner to re-optimize the global path, and dynamically updates the data fed back by the global path planner to the local path planning module; Set the path switching priority. When the local path is seriously affected by dynamic obstacles, the global path point is used as the priority reference point to ensure the global optimality and safety of path planning and prevent the robot from falling into the local optimum.

11. The MDC-AFI adaptive feedback fusion intelligent path planning method based on multi-dimensional collaborative enhancement according to claim 10 is characterized in that: Methods for determining whether to trigger the global path planner to re-optimize the global path include: Calculate the Euclidean distance and point projection distance between local path points and global path points; Euclidean distance δ euclidean for: in, are the point coordinates of the local path, is the point coordinate of the global path; Point projection distance δ projection for: in, and Indicates that a segment in the global path is the two endpoints of that segment; According to the complexity of the planning scenario, combined with the Euclidean distance δ euclidean and point projection distance δ euclidean , weighted fusion according to weight parameters σ, ψ is used to obtain comprehensive deviation data: d fused-linear =sd euclidenan +ψd projection , Among them, σ+ψ=1, δ fused-linear represents the linear fusion deviation data, δ fused-nonliner represents nonlinear fusion bias data, σ represents the fusion weight of Euclidean distance, and ψ represents the fusion weight of point projection distance; When the fusion deviation data δ is detected fused-linear and δ fused-nonliner Greater than the deviation threshold δ max When the multi-dimensional feasibility dynamic evaluator under the MDC-AFI framework triggers the global path planning to re-optimize the global path, fit the feedback data of the local path planner to obtain a more reliable global path, and then use the obtained global path as the input of the local path planner for continuous iterative optimization.

12. An MDC-AFI adaptive feedback fusion intelligent path planning system based on multi-dimensional collaborative enhancement, the system applying the method described in any one of claims 1 to 11, characterized in that: include: Initial information acquisition module, global path optimization module, local path optimization module and feedback fusion module; The initial information acquisition module is used to acquire the initial state information of the intelligent robot, and the initial state information includes: initial coordinates, posture information, obstacle distribution layer information and target point coordinates; The global path optimization module calls the dynamic adaptive global path planner to search the global path based on the initial state information, iteratively calculates the global path nodes, and generates the global optimization path; The local path optimization module is used to decompose the global optimization path to obtain a number of local target points and use them as the initial reference input of the local path planner, and adaptively adjust the parameters of the local path planner through the multi-dimensional feasibility dynamic evaluator under the MDC-AFI framework and the distribution information of dynamic obstacles in the local environment, and iteratively generate a dynamically optimized local path according to the dynamic constraints of the intelligent robot and the real-time environmental information; When the local path deviates from the global optimization path beyond a preset tolerance range, the multi-dimensional feasibility dynamic evaluator will determine whether the global path needs to be re-optimized and feed back the latest global path data to the local path planner for feedback fusion until the intelligent robot reaches the target point and completes the target navigation task.

Images