Ultrasonic enhanced robot path planning method and device based on neural network

Abstract

The invention discloses an ultrasonic enhancement robot path planning method and device based on a neural network, and belongs to the technical field of robot path planning. The method comprises the following steps: outputting a thermodynamic diagram of a prediction path from a starting point to a termination point for an input map, the starting point and the termination point by using a pre-trained neural network; a training sample used for training the neural network comprises a sample map, a sample starting point and a sample ending point which are used as input, and further comprises a sample optimal path which is used as output; the sample optimal path is generated by using a path planning algorithm; converting the thermodynamic diagram into one-dimensional probability distribution; and sampling the one-dimensional probability distribution based on a first random number to obtain a current sampling point, and determining a currently selected path point by using an adaptive step length and the current sampling point until path point selection from the starting point to the ending point is completed to obtain a planned path. A large amount of invalid sampling can be avoided, and redundant nodes are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of robot path planning, and in particular to a method and device for ultrasonic enhanced robot path planning based on a neural network. Background Art

[0002] Ultrasonic enhanced robot path planning involves planning a collision-free path from a starting point to an end point. Generally speaking, the shorter the path length, the better the planning effect.

[0003] Related technologies use RRT* (Rapidly-exploring Random Tree Star) and its variants for path planning. However, due to the random sampling nature, they suffer from slow convergence, numerous redundant nodes, and difficulty navigating narrow passages, resulting in a low success rate in complex maps with multiple obstacles.

[0004] Therefore, there is an urgent need to provide an efficient path planning method. Summary of the Invention

[0005] The present invention provides a neural network-based ultrasonic enhanced robot path planning method and device. The technical solution is as follows:

[0006] In one aspect, a neural network-based ultrasound-enhanced robot path planning method is provided, the method comprising:

[0007] Using a pre-trained neural network, based on an input map, a starting point, and an ending point, a heat map of a predicted path from the starting point to the ending point is output; the training samples used to train the neural network include a sample map, a sample starting point, and a sample ending point as input, and also include a sample optimal path as output; the sample optimal path is generated using a path planning algorithm;

[0008] Convert the heat map into a one-dimensional probability distribution;

[0009] Sampling is performed on the one-dimensional probability distribution based on the first random number to obtain a current sampling point, and the currently selected path point is determined using the adaptive step size and the current sampling point until the path point selection from the starting point to the end point is completed to obtain a planned path.

[0010] On the other hand, a neural network-based ultrasound-enhanced robot path planning device is provided, the device comprising:

[0011] a heat map acquisition unit configured to output a heat map of a predicted path from the starting point to the ending point using a pre-trained neural network for an input map, a starting point, and an ending point; the training samples used to train the neural network include a sample map, a sample starting point, and a sample ending point as input, and also include a sample optimal path as output; the sample optimal path is generated using a path planning algorithm;

[0012] A conversion unit, used to convert the heat map into a one-dimensional probability distribution;

[0013] A planning unit is used to sample the one-dimensional probability distribution based on a first random number to obtain a current sampling point, and to determine a currently selected path point using an adaptive step size and the current sampling point until the path point selection from the starting point to the ending point is completed to obtain a planned path.

[0014] On the other hand, a computer device is provided, comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of the above-mentioned neural network-based ultrasonic enhanced robot path planning method.

[0015] On the other hand, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned neural network-based ultrasound enhanced robot path planning method are implemented.

[0016] On the other hand, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the steps of the neural network-based ultrasound-enhanced robot path planning method described above.

[0017] The technical solution provided by the present invention can at least bring the following beneficial effects:

[0018] For the complex ultrasonic enhanced robot path planning environment, the path is predicted by training a neural network to output a heat map of the predicted path. By converting the heat map into a one-dimensional probability distribution, the heat map is used to guide the sampling of path planning, thereby avoiding a large number of invalid sampling and reducing redundant nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 This is a flow chart of a neural network-based ultrasonic enhanced robot path planning method provided by one embodiment of the present invention;

[0021] Figure 2 is a corresponding schematic diagram of an input map and a heat map provided by an embodiment of the present invention;

[0022] Figure 3 This is a schematic diagram of a bridge test provided by an embodiment of the present invention;

[0023] Figure 4 This is a test result of narrow channel recognition in a two-dimensional complex environment provided by an embodiment of the present invention;

[0024] Figure 5 This is a test result of narrow channel recognition in a three-dimensional complex environment provided by an embodiment of the present invention;

[0025] Figure 6 This is an example of a two-dimensional complex map provided by an embodiment of the present invention;

[0026] Figure 7 This is an example of a three-dimensional complex map provided by an embodiment of the present invention;

[0027] Figure 8 This is a comparison chart of the test results of two-dimensional complex map path planning provided by one embodiment of the present invention;

[0028] Figure 9 This is a comparison chart of the test results of a three-dimensional complex map path planning provided by an embodiment of the present invention;

[0029] Figure 10 This is a structural diagram of a neural network-based ultrasonic enhanced robot path planning device provided by one embodiment of the present invention;

[0030] Figure 11 This is a hardware architecture diagram of a computer device provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0032] Please refer to Figure 1 , an embodiment of the present invention provides a neural network-based ultrasonic enhanced robot path planning method, the method comprising:

[0033] Step 100: Using a pre-trained neural network, based on an input map, a starting point, and an ending point, output a heat map of a predicted path from the starting point to the ending point; the training samples used to train the neural network include a sample map, a sample starting point, and a sample ending point as input, and also include a sample optimal path as output; the sample optimal path is generated using a path planning algorithm;

[0034] Step 102, converting the heat map into a one-dimensional probability distribution;

[0035] Step 104: Sampling the one-dimensional probability distribution based on the first random number to obtain the current sampling point, and using the adaptive step size and the current sampling point to determine the currently selected path point until the path point selection from the starting point to the end point is completed to obtain the planned path.

[0036] In an embodiment of the present invention, for a complex ultrasonic enhanced robot path planning environment, a neural network is trained to predict the path to output a heat map of the predicted path. The heat map is converted into a one-dimensional probability distribution and used to guide the sampling of the path planning, thereby avoiding a large number of invalid sampling and reducing redundant nodes.

[0037] Described below Figure 1 How to perform the steps shown.

[0038] First, for step 100, a pre-trained neural network is used to output a heat map of the predicted path from the starting point to the ending point based on the input map, starting point and ending point.

[0039] In one embodiment of the present invention, the neural network can adopt U-net, which includes an encoder, a decoder and a jump structure, and has better segmentation accuracy in image segmentation tasks. Furthermore, U-net can be integrated with DenseNet, and the convolution operation part in the encoder of the original U-net is replaced by the DenseBlock (dense connection block) in DenseNet121, and the dense connection mechanism is used to strengthen feature transfer. In addition, convolution is used to process coordinate information, and the coordinate information is interpolated onto feature maps of different resolutions to enhance the expression of spatial information. In addition, DenseNet (densely connected convolutional network) strengthens feature transfer by interconnecting all layers, preserves low-dimensional features, alleviates the gradient vanishing problem of the convolutional network, and has a regularization effect. It can reduce the risk of overfitting for tasks with smaller training sets and is more suitable for path planning prediction tasks.

[0040] In an embodiment of the present invention, the training method of the neural network may include:

[0041] Step A1: Obtain multiple training samples; the training samples may include a sample map as input and a sample optimal path as output; the sample map is marked with a sample starting point and a sample ending point, and the sample optimal path is generated by a path planning algorithm;

[0042] When obtaining training samples, multiple multi-obstacle maps containing starting points and ending points can be randomly generated, and a path planning algorithm, such as the A* algorithm, can be used to generate an optimal path. The generated optimal path can be used as a target mask to obtain a training data set.

[0043] A2: Train the neural network using multiple training samples.

[0044] During the training process, the loss function can be optimized to enable the neural network to learn how to generate an output segmentation map that matches the target mask based on the input map and the coordinate information of the starting and ending points, and to minimize the difference between the predicted segmentation map and the true mask through continuous learning.

[0045] After training is complete, when the neural network receives the new map and the coordinates of the starting and ending points as input, it will output a predicted path heat map, see Figure 2 This is a comparison diagram of a map and a heat map. The green point in the map is the starting point and the red point is the end point. The heat map includes the probability of the optimal path point, and different probabilities are presented in different colors. Figure 2 Set the blue to red value to increase the probability gradually.

[0046] Then, step 102 "converting the heat map into a one-dimensional probability distribution" and step 104 "sampling the one-dimensional probability distribution based on the first random number to obtain the current sampling point, and using the adaptive step size and the current sampling point to determine the currently selected path point until the path point selection from the starting point to the end point is completed to obtain the planned path" are explained at the same time.

[0047] During path planning, it is necessary to obtain planned path points through sampling. Traditionally, random sampling is performed on the input map, resulting in a large number of redundant sampling points. To reduce the number of redundant sampling points, an embodiment of the present invention utilizes a heat map of the predicted path output by a neural network for probability-guided node sampling. The core process is to convert the heat map into a one-dimensional probability distribution and then perform sampling based on a first random number.

[0048] In the embodiment of the present invention, the method of converting the heat map into a one-dimensional probability distribution is:

[0049] Flatten the heat map from a two-dimensional or three-dimensional matrix into a one-dimensional array, perform positive processing, and then normalize the normalized one-dimensional array to obtain an initial probability distribution. Normalization is to divide each weight by the total weight sum to obtain an initial probability distribution with a sum of 1. This initial probability distribution includes the probability and position index of each point; the position index is used to indicate the position of the point in the one-dimensional array of the initial probability distribution.

[0050] Furthermore, to facilitate sampling, the initial probability distribution can be formed into a cumulative probability array. Each item in the cumulative probability array represents the sum of all current and previous weights, so high-weight positions occupy a larger interval range. When the one-dimensional probability distribution is in the form of a one-dimensional cumulative probability array, a first random number is randomly generated in the interval [0, 1), and sampling is performed on the one-dimensional probability distribution based on the first random number. The sampling method may include:

[0051] Determine the cumulative probability greater than the first random number in the cumulative probability array, and use the position index of the minimum cumulative probability among the cumulative probabilities greater than the first random number as the selected target point, index the target point to the corresponding coordinate position in the heat map, and use the point at the coordinate position as the current sampling point.

[0052] For example, the cumulative probability array is (0.01, 0.03, 0.05, 0.09, …, 1). If the first random number is 0.04, the minimum cumulative probability greater than the first random number is 0.05, and the position index of 0.05 is used as the selected target point; since each number in the cumulative probability array has a corresponding coordinate position in the heat map, the target point can be indexed to the corresponding coordinate position in the heat map, and the point at the coordinate position is used as the current sampling point.

[0053] Furthermore, you can also set a bias for the neural network to choose whether to perform probability-guided sampling on the heat map or to use a random sampling strategy to perform random sampling on the map, so as to ensure that the sampling process does not completely rely on the predicted path, retain a certain degree of exploratory nature, and avoid falling into the local optimal situation.

[0054] Specifically, before sampling the one-dimensional probability distribution based on the first random number, the method further includes:

[0055] Generate a second random number, and determine whether the second random number is less than the bias set for the neural network. If so, perform sampling on the one-dimensional probability distribution based on the first random number; otherwise, perform sampling on the map using a random sampling strategy, so as to determine the currently selected path point using the adaptive step size and the current sampling point using the current sampling point obtained by random sampling.

[0056] The second random number is in the interval (0, 1), and the bias is a value in the interval (0, 1), such as 0.5, 0.2, or 0.8. The specific value of the bias can be set based on the design proportion of the random sampling strategy required in the application process, or based on specific experimental results.

[0057] Whether sampling the heat map for probability-guided sampling or random sampling on the map using a random sampling strategy, the current sampling point will be obtained. In one embodiment of the present invention, when determining the currently selected path point using the adaptive step size and the current sampling point, the following steps may be specifically included:

[0058] B1: Determine the neighbor path point closest to the current sampling point among the selected path points;

[0059] B2: Expand from the neighboring path points toward the current sampling point according to the adaptive step size calculation to obtain a new sampling point;

[0060] B3: Determine whether a collision occurs when connecting the neighbor path point with the new sampling point. If so, re-execute the sampling process; if not, determine the new sampling point as the currently selected path point.

[0061] In an embodiment of the present invention, in step B2, the neighbor path point and the current sampling point are connected, and the point corresponding to the increase of the adaptive step size from the neighbor path point is determined on the connecting line and its extension line, and the point is used as the new sampling point.

[0062] In step B3, if a neighboring path point collides with the new sampling point, it indicates that the new sampling point is incorrect and needs to be resampled to iteratively update to obtain a new sampling point. If the path point cannot be selected after the number of iterations reaches the set value, it indicates that the planning has failed.

[0063] In order to balance path planning efficiency and path smoothness, in the embodiment of the present invention, adaptive step size is used for expansion, so that the step size is reduced when the RRT* tree is expanded to narrow channels and near the end point, so that the path reduces collisions in narrow channels, speeds up the connection to the end point, and completes the path exploration.

[0064] Based on this, in one embodiment of the present invention, before the step of extending from the neighboring path points toward the current sampling point according to the adaptive step size calculation, the following steps may also be included:

[0065] The current adaptive step size is determined as follows:

[0066] C1: Pre-calibrate the narrow channel center using bridge test and density-based spatial clustering algorithm;

[0067] C2: Determine the current adaptive step size based on the distance between the neighbor path point and the nearest narrow channel center, and based on the distance between the neighbor path point and the end point.

[0068] In an embodiment of the present invention, in step C1, the Bridge Test and DBSCAN (density-based spatial clustering algorithm) are introduced to complete the calibration of the center of the narrow channel, and the positional relationship between the neighboring path points and the narrow channel is converted into the Euclidean distance between the path points and the center of the narrow channel, and the position of the neighboring path points is determined by a simple mathematical relationship.

[0069] Specifically, the method for calibrating the center of a narrow channel is to randomly sample two points in non-free space (obstacles). If the midpoint of the two points is in the free space, the area is determined to be a narrow channel. Figure 3 , the two randomly sampled points are P1 and P2, where the point is located in free space (no obstacles). Multiple points located in the narrow channel are obtained through multiple random samplings, and the density-based spatial clustering algorithm is used to identify clusters to obtain the center position of the narrow channel. Figure 4 and Figure 5 , respectively, are the test results of narrow channel recognition in two-dimensional and three-dimensional complex environments. The blue scattered points are the points that may be in the narrow channel obtained by bridge test sampling, and the red triangles are the centers of the narrow channels marked by DBSCAN.

[0070] In this embodiment of the present invention, in step C2, the adaptive step size takes into account two factors: first, the distance from the narrow passage is considered, and the area near the narrow passage is explored with a smaller step size to avoid excessive collisions; second, the step size is reduced when approaching the end point to reduce ineffective exploration near the end point area. In one implementation, the current adaptive step size can be determined using the following formula:

[0071]

[0072] StepSize = S × β × γ

[0073] Among them, StepSize is the adaptive step size, c is the distance between the neighbor path point and the center of the nearest narrow channel, λ controls the attenuation degree, = is the narrow channel influence factor, and β decreases as c decreases; d is the distance from the neighbor path point to the end point, dt is the target distance threshold, k controls the target distance influence degree, γ is the target distance influence factor, and γ decreases as d decreases; S is the set initial step size.

[0074] By setting the adaptive step size in the above way, the RRT* tree can reduce the step size near the narrow channel and the terminal point area, speeding up the efficiency of passing through the narrow channel and connecting the terminal points.

[0075] By using the embodiment of the present invention, all path points from the starting point to the ending point can be obtained, and these path points can form an initial path. One implementation method can directly use the initial path as the final planned path.

[0076] In another implementation, in order to further reduce redundant nodes, the initial path formed by all path points selected from the starting point to the end point can be optimized using a post-pruning strategy of the triangle inequality to delete redundant nodes and obtain a planned path.

[0077] Specifically, the algorithm traverses all path points, sequentially selecting the starting point and subsequent path points along the path. It then checks for collisions between the two points. If the straight line connecting the two points does not collide with obstacles, and its Euclidean distance is less than the sum of the lengths of the paths passing through all intermediate path points, then the triangle inequality theory is satisfied and the intermediate path points are deemed redundant and deleted. This method effectively simplifies and smoothes the path, significantly reducing the number of nodes in the pruned path and further improving the coherence of the robot's motion.

[0078] In order to further verify the embodiment of the present invention, experimental tests were carried out. Figure 6 、 Figure 7 The two-dimensional and three-dimensional complex maps shown in the figure are used as objects, and RRT, RRT-Connect, RRT*, Bi-RRT* and Informed-RRT* are selected as comparison algorithms to verify the performance of the path planning method AN-RRT* of the embodiment of the present invention. The longest path length L after 30 independent runs is max , average path length L mean , the shortest path L min The results are as follows Figure 8 、 Figure 9 As shown in the experimental results, in the path obtained by the AN-RRT* algorithm of the embodiment of the present invention, L max 、L mean and L min The values ​​of are all minimum, which verifies that the method provided by the embodiment of the present invention has powerful search capabilities and effective stability, which is also necessary for solving actual path search problems and can be used to efficiently solve path planning problems in complex environments.

[0079] Please refer to Figure 10 The embodiment of the present invention provides a neural network-based ultrasonic enhanced robot path planning device, the device comprising:

[0080] The heat map acquisition unit 1000 is configured to use a pre-trained neural network to output a heat map of a predicted path from the starting point to the ending point based on an input map, a starting point, and an ending point. The training samples used to train the neural network include a sample map, a sample starting point, and a sample ending point as input, and also include a sample optimal path as output. The sample optimal path is generated using a path planning algorithm.

[0081] A conversion unit 1002 is used to convert the heat map into a one-dimensional probability distribution;

[0082] The planning unit 1004 is used to sample the one-dimensional probability distribution based on the first random number to obtain the current sampling point, and use the adaptive step size and the current sampling point to determine the currently selected path point until the path point selection from the starting point to the end point is completed to obtain the planned path.

[0083] In one embodiment of the present invention, the device may further include a sampling determination unit for generating a second random number before sampling the one-dimensional probability distribution based on the first random number, and determining whether the second random number is less than the bias set for the neural network; if so, executing the sampling of the one-dimensional probability distribution based on the first random number; otherwise, sampling on the map using a random sampling strategy, and determining the currently selected path point using the adaptive step size and the current sampling point using the current sampling point obtained by random sampling.

[0084] In one embodiment of the present invention, the one-dimensional probability distribution is in the form of a one-dimensional cumulative probability array; the first random number is greater than or equal to 0 and less than 1;

[0085] When the planning unit executes the sampling of the one-dimensional probability distribution based on the first random number, it specifically includes: determining the cumulative probability in the cumulative probability array that is greater than the first random number, and taking the position index of the minimum cumulative probability among the cumulative probabilities greater than the first random number as the selected target point, indexing the target point to the corresponding coordinate position in the heat map, and taking the point at the coordinate position as the current sampling point.

[0086] In one embodiment of the present invention, when the planning unit executes the method of determining the currently selected path point using the adaptive step size and the current sampling point, the method specifically includes: determining a neighbor path point that is closest to the current sampling point among the already selected path points; expanding from the neighbor path point toward the current sampling point according to the adaptive step size calculation to obtain a new sampling point; determining whether a collision will occur when connecting the neighbor path point with the new sampling point, and if so, re-executing the sampling process; if not, determining the new sampling point as the currently selected path point.

[0087] In one embodiment of the present invention, the current adaptive step size is determined as follows:

[0088] The narrow channel center is pre-calibrated using bridge test and density-based spatial clustering algorithm;

[0089] The current adaptive step size is determined based on the distance between the neighbor path point and the nearest narrow channel center, and based on the distance between the neighbor path point and the end point.

[0090] In one embodiment of the present invention, the device may also include: an optimization unit for optimizing the initial path formed by all the path points selected from the starting point to the end point after completing the selection of the path points from the starting point to the end point and before obtaining the planned path, using the post-pruning strategy of the triangle inequality to delete redundant nodes and obtain the planned path.

[0091] It should be noted that the neural network-based ultrasound-enhanced robot path planning device provided in the above embodiment is merely an example of the division of the aforementioned functional modules. In actual applications, the aforementioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the neural network-based ultrasound-enhanced robot path planning device provided in the above embodiment and the neural network-based ultrasound-enhanced robot path planning method embodiment are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0092] The embodiment of the present application also provides a computer device, please refer to Figure 11 The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set, or instruction set, and the at least one instruction, at least one program, code set, or instruction set is loaded and executed by the processor to implement the neural network-based ultrasound enhancement robot path planning method provided by the above-mentioned method embodiments.

[0093] An embodiment of the present application also provides a computer-readable storage medium, on which is stored at least one instruction, at least one program, code set, or instruction set, and the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the neural network-based ultrasonic enhanced robot path planning method provided in the above-mentioned method embodiments.

[0094] An embodiment of the present application also provides a computer program product, which includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the neural network-based ultrasonic enhanced robot path planning method described in any of the above embodiments.

[0095] For the convenience of description, the above systems or devices are described as being divided into various modules or units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0096] Through the description of the above embodiments, it can be seen that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present application or certain parts of the embodiments.

[0097] Finally, it should be noted that, in this document, relational terms such as first, second, third, and fourth are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

[0098] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A neural network-based ultrasonic enhanced robot path planning method, characterized in that: The method comprises: Using a pre-trained neural network, based on an input map, a starting point, and an ending point, a heat map of a predicted path from the starting point to the ending point is output; the training samples used to train the neural network include a sample map, a sample starting point, and a sample ending point as input, and also include a sample optimal path as output; the sample optimal path is generated using a path planning algorithm; Convert the heat map into a one-dimensional probability distribution; Sampling is performed on the one-dimensional probability distribution based on the first random number to obtain a current sampling point, and the currently selected path point is determined using the adaptive step size and the current sampling point until the path point selection from the starting point to the end point is completed to obtain a planned path.

2. The method according to claim 1, characterized in that Before sampling the one-dimensional probability distribution based on the first random number, the method further includes: Generate a second random number, and determine whether the second random number is less than the bias set for the neural network. If so, perform sampling on the one-dimensional probability distribution based on the first random number; otherwise, perform sampling on the map using a random sampling strategy, so as to determine the currently selected path point using the adaptive step size and the current sampling point using the current sampling point obtained by random sampling.

3. The method according to claim 1 or 2, characterized in that The one-dimensional probability distribution is in the form of a one-dimensional cumulative probability array; the first random number is greater than or equal to 0 and less than 1; The sampling of the one-dimensional probability distribution based on the first random number includes: Determine the cumulative probability greater than the first random number in the cumulative probability array, and use the position index of the minimum cumulative probability among the cumulative probabilities greater than the first random number as the selected target point, index the target point to the corresponding coordinate position in the heat map, and use the point at the coordinate position as the current sampling point.

4. The method according to claim 1, wherein The method of determining the currently selected path point by using the adaptive step size and the current sampling point includes: Determine a neighbor path point closest to the current sampling point among the selected path points; Expanding from the neighboring path points toward the current sampling point according to the adaptive step size calculation to obtain a new sampling point; Determine whether a collision occurs when connecting the neighbor path point to the new sampling point. If so, re-execute the sampling process; if not, determine the new sampling point as the currently selected path point.

5. The method according to claim 4, characterized in that Before the step of extending from the neighboring path points toward the current sampling point according to the adaptive step size calculation, the method further includes: The current adaptive step size is determined as follows: The narrow channel center is pre-calibrated using bridge test and density-based spatial clustering algorithm; The current adaptive step size is determined based on the distance between the neighbor path point and the nearest narrow channel center, and based on the distance between the neighbor path point and the end point.

6. The method according to any one of claims 1 to 5, characterized in that: After completing the path point selection from the starting point to the end point, before obtaining the planned path, it also includes: For the initial path formed by all the path points selected from the starting point to the end point, the post-pruning strategy of the triangle inequality is used to perform optimization processing to delete redundant nodes and obtain the planned path.

7. A neural network-based ultrasonic enhanced robot path planning device, characterized in that: The device comprises: a heat map acquisition unit configured to output a heat map of a predicted path from the starting point to the ending point using a pre-trained neural network for an input map, a starting point, and an ending point; the training samples used to train the neural network include a sample map, a sample starting point, and a sample ending point as input, and also include a sample optimal path as output; the sample optimal path is generated using a path planning algorithm; A conversion unit, used to convert the heat map into a one-dimensional probability distribution; A planning unit is used to sample the one-dimensional probability distribution based on a first random number to obtain a current sampling point, and to determine a currently selected path point using an adaptive step size and the current sampling point until the path point selection from the starting point to the ending point is completed to obtain a planned path.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of any one of the methods described in claims 1-6.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises a computer program, wherein when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images