Path planning method and device based on signal sequential logic and nerve guidance sampling

Abstract

The invention discloses a path planning method and device based on signal sequential logic and nerve guidance sampling, and relates to the technical field of mobile robot path planning. The method comprises the steps of obtaining working environment information of the mobile robot and converting the working environment information into an environment point cloud data set; analyzing and splitting a to-be-executed signal sequential logic task into a plurality of sub-tasks; for each subtask, inputting the environment point cloud data and the dynamically updated starting and ending point masks into a pre-trained PointNet + + deep network, and outputting a guide point cloud data set; carrying out mixed sampling by adopting a fast exploration random tree star algorithm and generating a new node; selecting a father node for a new node by using a hybrid cost function fused with sequential logic robustness, and recursively updating the accumulated cost; if it is judged that the trajectory meets all subtask constraints in sequence, the optimal path meeting global signal sequential logic constraints is extracted, output and executed, and the path planning success rate and solving efficiency in the complex environment are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of mobile robot path planning technology, and in particular to a path planning method and apparatus based on signal timing logic and neural guided sampling. Background Technology

[0002] With the increasing application of robotics in intelligent warehousing and logistics, disaster relief, home services, security patrols, and autonomous driving, mobile robots face increasingly complex and diverse tasks. Traditional path planning problems typically focus only on how to move from a starting point to a destination while avoiding obstacles (i.e., "point-to-point" geometric planning). However, real-world tasks often involve more advanced and stringent spatiotemporal constraints. For example, in warehousing and logistics, a robot may need to first reach area A to pick up goods, then deliver them to area B within 5 minutes, while maintaining a distance of no more than 10 meters from the charging station throughout the process; in security patrols, a robot needs to visit monitoring points an unlimited number of times and immediately execute specific avoidance actions when an anomaly is detected. These tasks not only involve obstacle avoidance in geometric space but also implicitly involve strict temporal order and logical causal relationships.

[0003] To accurately describe and verify such complex tasks, formal methods have been introduced into the field of robot motion planning. Among them, Signal Temporal Logic (STL) has become the mainstream language for describing complex spatiotemporal tasks of continuous dynamic systems due to its powerful semantic expressiveness and quantitative robustness metrics. STL's robustness can quantitatively evaluate the degree to which a trajectory satisfies the specification (i.e., safety margin), with positive values ​​indicating satisfaction and negative values ​​indicating violation. Larger absolute values ​​indicate greater distance from or closer to the violation boundary, making it highly suitable for optimization-based control and planning.

[0004] In existing STL path planning methods, optimization-based methods such as Mixed Integer Linear Programming (MILP) can find the optimal solution, but their computational complexity increases exponentially with the task size, making it difficult to meet real-time requirements. In contrast, sampling-based algorithms such as Rapidly-exploring Random Tree (RRT) or Rapidly-exploring Random Tree star schema algorithm offer better solutions. RRT Due to its probabilistic completeness and adaptability to high-dimensional spaces, it has become the mainstream approach for solving STL programming problems. Existing STL-RRT... The algorithm guides the search tree to grow in a direction that satisfies the logic by incorporating STL robustness into the cost function. However, STL-RRT... The algorithm has significant shortcomings in practical applications. The timing and logical constraints in STL tasks essentially divide the high-dimensional state space into several discrete and narrow "feasible pipelines" or "narrow channels." Traditional STL-RRT... The algorithm employs a strategy of uniform random sampling throughout the free space, which is essentially a "blind search." Under STL constraints, the vast majority of random samples will fall into regions that violate temporal logic (i.e., invalid samples), causing the algorithm to require an extremely large number of samples to "happen" to find a feasible solution through the narrow channel. This results in excessively long planning time and extremely slow convergence speed, and may even lead to failure to find a solution within a finite time.

[0005] To address the problem of low sampling efficiency, learning-based guided sampling methods have emerged. For example, the Neural Rapidly-exploring Random Tree (Neural RRT) algorithm... Neural networks (such as U-Net) learn environmental features and predict regions with a high probability of containing the optimal path, thereby generating a non-uniform guide sampling distribution, significantly improving the efficiency of geometric planning. However, most existing neural guidance methods only focus on obstacle avoidance and shortest paths at the geometric level, without fully considering the complex STL temporal logic constraints. If these methods are applied directly, the guidance region generated by the neural network may only satisfy the geometric shortest path but violate the temporal requirements, leading to guidance failure.

[0006] In summary, existing sampling-based STL planning algorithms suffer from low sampling efficiency and slow convergence, while existing neural guided planning algorithms lack effective handling of temporal logic constraints. As a result, the low efficiency and success rate of path planning for mobile robots in complex and unstructured environments is a technical problem that urgently needs to be solved. Summary of the Invention

[0007] To address the issues of low sampling efficiency and slow convergence in existing technologies, and the lack of effective handling of temporal logic constraints in current neural-guided planning algorithms, which leads to low path planning efficiency and success rate for mobile robots in complex and unstructured environments, this invention provides a path planning method and apparatus based on signal temporal logic and neural-guided sampling. The technical solution is as follows: On the one hand, a path planning method based on signal timing logic and neural guided sampling is provided. This method is implemented by a path planning device based on signal timing logic and neural guided sampling, and includes: S1. Obtain the working environment information of the mobile robot in the intelligent warehousing and logistics distribution task, and model it as a two-dimensional map. The working environment information includes static environment features and dynamic state features. The static environment features include the global physical boundary, the position and geometry of static obstacles, and the dynamic state features include the robot's current pose state and the free space region after removing obstacles and considering the robot's safe expansion radius. Sampling is performed within the free space region to discretize the two-dimensional map into an environmental point cloud dataset. S2. Define the signal timing logic task to be executed. Input the environmental point cloud dataset into the pre-trained PointNet++ deep network and output the probability score of each point belonging to the optimal path neighborhood. Compare the probability score of each point belonging to the optimal path neighborhood with a pre-set threshold, filter out points that are greater than the pre-set threshold, and construct the guide point cloud dataset. S3. Based on the guided point cloud dataset, candidate nodes are randomly sampled and generated using the fast exploratory random tree star schema algorithm; all nodes in the tree are traversed based on the candidate nodes, and the nearest node is found; the nearest node is expanded into a new node according to the set expansion step size; collision detection is performed based on the new node, and if there is no collision, the parent node is selected by constructing the hybrid cost function of the path. S4. Add the new node to the tree, and recursively update the cost of all its child nodes by calculating the cumulative cost of the new node and its parent node. If it is determined that the trajectory to be executed satisfies the signal timing logic constraints of all subtasks in sequence, then extract and output the optimal path that satisfies the global signal timing logic constraints and execute it to complete the path planning.

[0008] On the other hand, a path planning device based on signal temporal logic and neural guided sampling is provided. This device is applied to a path planning method based on signal temporal logic and neural guided sampling. The device includes: The acquisition unit is used to acquire the working environment information of the mobile robot in the intelligent warehousing and logistics distribution task, and model it into a two-dimensional map. The working environment information includes static environment features and dynamic state features. The static environment features include the global physical boundary, the position and geometry of static obstacles, and the dynamic state features include the robot's current pose state and the free space region after removing obstacles and considering the robot's safe expansion radius. Sampling is performed within the free space region to discretize the two-dimensional map into an environmental point cloud dataset. The building unit is used to define the signal timing logic task to be executed. It inputs the environmental point cloud dataset into a pre-trained PointNet++ deep network and outputs the probability score of each point belonging to the optimal path neighborhood. It compares the probability score of each point belonging to the optimal path neighborhood with a pre-set threshold, filters out points that are greater than the pre-set threshold, and builds the guide point cloud dataset. The detection unit is used to randomly sample and generate candidate nodes based on the guided point cloud dataset using a fast exploratory random tree star schema algorithm; it traverses all nodes in the tree based on the candidate nodes and finds the node that is closest to the candidate node; it expands the nearest node into a new node according to the set expansion step size; it performs collision detection based on the new node, and if there is no collision, it selects the parent node by constructing the path using a hybrid cost function; The output and execution unit is used to add new nodes to the tree and recursively update the costs of all its child nodes by calculating the cumulative cost of the new node and its parent node. If it is determined that the trajectory to be executed satisfies the signal timing logic constraints of all subtasks in sequence, the optimal path that satisfies the global signal timing logic constraints is extracted, output, and executed to complete the path planning.

[0009] On the other hand, a path planning device based on signal timing logic and neural guided sampling is provided. The path planning device based on signal timing logic and neural guided sampling includes: a processor; a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, they implement any of the above-described path planning methods based on signal timing logic and neural guided sampling.

[0010] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, the at least one instruction being loaded and executed by a processor to implement any of the above-described path planning methods based on signal timing logic and neural guided sampling.

[0011] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: This invention significantly improves planning efficiency: Addressing the "narrow channel" problem in the state space caused by complex temporal logic tasks, this invention innovatively introduces a PointNet++ deep neural network for environmental feature extraction and sampling guidance. This network can directly predict the sampling probability distribution (guiding point cloud) that satisfies the temporal task from unstructured environmental point clouds, thereby achieving intelligent pruning of the invalid search space. In complex maze and sequence task scenarios, this invention significantly improves planning efficiency compared to traditional STL-RRT. The algorithm reduces planning time and significantly improves computational efficiency, enabling it to meet the real-time planning needs of mobile robots.

[0012] This invention improves planning success rate and path quality: By employing a hybrid sampling strategy, this invention leverages the prior knowledge of neural networks to quickly traverse narrow regions while preserving the probabilistic completeness of full-space sampling. This effectively solves the problem of traditional algorithms struggling to converge under complex constraints, thus improving planning success rate. Simultaneously, since the guiding point cloud is typically located on the optimal homotopy class, the generated path length is shortened.

[0013] This invention enhances the temporal safety (robustness) of paths: This invention redefines RRT. The cost function uses the temporal robustness integral of STL as the core optimization metric. This allows the planner to move beyond simply finding the geometrically shortest path and actively search for paths that are farther from obstacles, deeper into the core of the target area, and more in line with temporal logic constraints. This "robust-oriented" optimization mechanism gives the generated trajectory a higher safety margin, effectively addressing positioning errors and control noise in actual robot operation. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of a path planning method based on signal timing logic and neural guided sampling provided by an embodiment of the present invention; Figure 2 This is an overall flowchart of path planning based on signal timing logic and neural guided sampling provided by an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a PointNet++ neural guidance network provided in an embodiment of the present invention; Figure 4 This is a block diagram of a path planning device based on signal timing logic and neural guided sampling provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a path planning device based on signal timing logic and neural guided sampling provided in an embodiment of the present invention. Detailed Implementation

[0016] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0017] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0018] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0019] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0020] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0021] This invention provides a path planning method based on signal timing logic and neural guided sampling. This method can be implemented by a path planning device based on signal timing logic and neural guided sampling, which can be a terminal or a server. Figure 1 The flowchart shown is for a path planning method based on signal timing logic and neural guided sampling. The processing flow of this method may include the following steps: S1. Obtain the working environment information of the mobile robot in the intelligent warehousing and logistics distribution task, and model it as a two-dimensional map. The working environment information includes static environment features and dynamic state features. The static environment features include the global physical boundary, the position and geometry of static obstacles, and the dynamic state features include the robot's current pose state and the free space region after removing obstacles and considering the robot's safe expansion radius. Sampling is performed in the free space region to discretize the two-dimensional map into an environmental point cloud dataset.

[0022] In one feasible implementation, the environmental point cloud dataset is represented as ;in The number of point clouds is defined. Simultaneously, the signal timing logic task to be executed is defined. For example, a typical sequence task can be described as... This means that the robot needs to be ready in time. Pre-arrival area and in time to Reaching the area .

[0023] in, Figure 2This invention provides an overall flowchart of path planning based on signal timing logic and neural guided sampling. The overall process includes two stages: the first stage is to generate guided point cloud data by constructing a PointNet++ deep network; the second stage is an STL-constrained RRT. The planning loop is as follows. The first phase of implementation includes: acquiring the map environment and initial state, and setting the STL task specifications and target area; generating raw point cloud data based on the map environment, initial state, and the set STL task specifications and target area; and inputting the raw point cloud data into a pre-trained PointNet++ deep network to generate guiding point cloud data. Specifically, acquiring the map environment... Compared with the initial state and set STL task specifications. and target area Sampling in free space based on the map environment (e.g., sampling) Generate raw point cloud data (points); input the raw point cloud data into the pre-trained PointNet++ model. In the middle, high-probability areas are output, which are used to generate guide point cloud data. The second phase of implementation includes: iteratively generating the optimal path that satisfies the STL using a fast exploratory random tree star topology algorithm based on the guiding point cloud data. The specific steps are as follows: (1) Start the iterative loop, and set the maximum number of iterations to . (Right now ); (2) In each iteration, a random probability is generated. and compared with the set neural sampling probability threshold. (like Compare: If Then, the neural guidance strategy is executed, starting from the guidance point cloud. Mid-sampling to obtain random nodes Otherwise in free space Uniform sampling is performed inside. (Uniform / Target Bias); (3) Find the nearest node in the search tree. and according to step size (e.g. Expand to get new nodes ; (4) Perform collision detection ( If there is no collision, then within the set search radius (e.g., Find the set of all nearest neighbors within a given range. If a collision occurs, return to step (2) and continue iterating; (5) Select the optimal parent node from the nearest neighbor nodes. The selection criterion is to minimize the cumulative mixing cost. ; Among them, the cost increment .

[0024] (6) Add new nodes Add to search tree The cost of updating; (7) For other nearest neighbor nodes ( ) Perform a reconnection (rerouting) operation: if the cost of the new path is less than the cost of the old path ( If the topology is updated using STL cost, then the topology is updated using STL cost. (8) Determine whether the current new node has entered the target area and whether its trajectory meets the STL specification ( and If true, then proceed from the search tree. Extract and output the optimal path Otherwise, continue iterating.

[0025] S2. Define the signal timing logic task to be executed. Input the environmental point cloud dataset into the pre-trained PointNet++ deep network and output the probability score of each point belonging to the optimal path neighborhood. Compare the probability score of each point belonging to the optimal path neighborhood with a pre-set threshold, filter out points that are greater than the pre-set threshold, and construct the guide point cloud dataset.

[0026] The PointNet++ deep network was pre-trained through offline supervised learning. The training steps included: constructing a large-scale map dataset containing randomly distributed obstacles and start and end points; running an exact graph search algorithm to generate pixel-level global optimal paths as ground truth labels; sampling to generate point clouds in the free space of each environment, marking points within the radius of the optimal path as positive samples and points outside the radius as negative samples; and using a binary cross-entropy loss function and optimization algorithm to perform multiple rounds of forward and backward propagation iterative training until convergence to obtain the model weights, thus obtaining the trained PointNet++ deep network.

[0027] In order to enable the PointNet++ deep network to perceive task intent, this embodiment of the invention provides for each point... A feature vector was constructed. .in, Let be the geometric coordinates of the point. Indicate whether the point is an obstacle. Indicate whether this point is the initial state. Indicate whether the point belongs to the target area of ​​the current subtask.

[0028] In this context, the guiding point cloud data is presented geometrically as a "channel" connecting the starting point and the target area, indicating a high-probability feasible path area.

[0029] Optionally, the PointNet++ deep network includes: an encoder, a decoder, a fully connected layer, and a classifier; Among them, the PointNet++ deep network is used to learn the geometric topological features of the environment and predict the distribution of the optimal path.

[0030] The encoder consists of a multi-level set of abstraction layers, which is used to perform stepwise downsampling and feature extraction on the input environmental point cloud data with feature labels, and output the local geometric features and global topological features of the environment. Among them, the local geometric features of the environment can be the edges of obstacles; the global topological features can be channels or dead ends.

[0031] The decoder consists of a multi-level feature propagation layer, which is used to upsample deep features and map them back to the original point cloud resolution.

[0032] in, Figure 3 This is a schematic diagram of the structure of a PointNet++ neural guidance network provided in an embodiment of the present invention.

[0033] Optionally, the specific implementation process of S2 includes S21-S24: S21. Input the environmental point cloud dataset into a pre-trained PointNet++ deep network, and construct point cloud features using predefined map free space points and starting point neighborhood masks; Among them, the map free space points are the discretized coordinate points sampled in the free space after obstacle inflation processing of the original map; Among them, the starting point neighborhood mask is a context fusion feature, which is obtained by: calculating the Euclidean distance from the map free space point to the subtask starting point or target point; if the distance is less than the preset neighborhood radius, the point is assigned a mask value of 1, otherwise it is 0, and a one-hot vector is generated; the free space point coordinates are concatenated with the mask vector to construct a point cloud feature.

[0034] S22. Input the point cloud features into the encoder, perform step-by-step sampling and feature extraction through a multi-level ensemble abstraction layer, and output local and global features; S23. Input the local and global features into the decoder, upsample them through a multi-level feature propagation layer and map them back to the original point cloud resolution, and output the processed feature points. S24. Input the processed feature points into the fully connected layer and generate the probability score and path point prediction mask for each point belonging to the optimal path neighborhood through the classifier.

[0035] Specifically, an encoder-decoder network structure based on PointNet++ was constructed. The input is environmental point cloud data with features of obstacles, starting point and STL task target region. The output is the probability score of each spatial point belonging to the neighborhood of the optimal path, thereby generating a non-uniform guided sampling distribution and realizing effective pruning of the high-dimensional search space.

[0036] S3. Based on the guided point cloud dataset, a fast exploratory random tree star schema algorithm is used to randomly sample and generate candidate nodes; based on the candidate nodes, all nodes in the tree are traversed and the nearest node to the candidate node is found; the nearest node is expanded into a new node according to the set expansion step size; collision detection is performed on the new node, and if there is no collision, the parent node is selected by constructing the hybrid cost function of the path.

[0037] In one feasible implementation, the nearest neighbor node is determined by calculating the Euclidean distance between all nodes in the tree and the candidate node.

[0038] Optionally, S3 uses a fast exploratory random tree star schema algorithm to randomly sample and generate candidate nodes based on the guided point cloud dataset, including: Set neural sampling probability threshold Generate a random number that follows a uniform distribution in the range [0,1]. ; like Then, a point is randomly selected directly from the guide point cloud dataset as the sampling point. ; like Then, uniform random sampling is performed within the free space of the entire working environment to obtain sampling points. .

[0039] This includes the process of resampling nodes if a collision occurs.

[0040] To guide the search tree towards safer and more logically constrained regions, this embodiment of the invention redefines the hybrid cost function of the path. Specifically, it introduces a spatial robustness index based on signal temporal logic. Used to quantize state Distance violates STL task Safety margin at the boundary. For example, for obstacle avoidance tasks, a larger robustness value indicates a greater distance from the obstacle; for target approach tasks, a larger robustness value indicates greater depth.

[0041] Alternatively, the hybrid cost function of the path can be expressed by the following formula (1): (1) in, Indicates the parent node; Indicates a new node; Indicates Euclidean distance; Indicates the first weighting coefficient; This represents the second weighting coefficient; Represents the STL timing robustness integral on the path segment; The mixed cost function represents the path.

[0042] In this embodiment of the invention, the cost function for the constructed path tends to select paths that are shorter and more robust (i.e., more time-order-safe). This function not only penalizes path length but also rewards the path's time-order robustness, guiding the RRT. The search tree grows towards regions that satisfy temporal logic and offer higher security.

[0043] Alternatively, the STL timing robustness integral on the path segment is expressed by the following formula (2): (2) in, Indicates the parent node; Indicates a new node; Indicates the time step between two nodes; This indicates that the parent node satisfies the current signal timing logic task. Spatial robustness; This indicates that the new node satisfies the current signal timing logic task. Spatial robustness.

[0044] S4. Add the new node to the tree, and recursively update the cost of all its child nodes by calculating the cumulative cost of the new node and its parent node. If it is determined that the trajectory to be executed satisfies the signal timing logic constraints of all subtasks in sequence, then extract and output the optimal path that satisfies the global signal timing logic constraints and execute it to complete the path planning.

[0045] One feasible implementation involves a path-based hybrid cost function, achieved through a rapidly-exploring random tree star schema algorithm. RRT The process of selecting the parent node has been repeated.

[0046] Optionally, the specific implementation process of S4 includes S41-S43: S41. Within the neighborhood radius of the new sampling node, traverse all new sampling nodes and calculate the cumulative cost when the new sampling node is connected to each potential parent node. Select the node with the smallest cumulative cost as the parent node of the new sampling node. In one feasible implementation, the cumulative cost is expressed as ,in, This represents the accumulated historical cost of reaching the potential parent node from the root node (i.e., the initial state) along the tree topology. This represents the incremental cost of connecting the potential parent node to the newly sampled node; Indicates a potential parent node within the neighborhood; This indicates a new sampling node.

[0047] S42. Add the new sampled node to the search tree and find other neighboring nodes in its neighborhood; calculate the new cumulative cost generated when the new sampled node is connected to each neighboring node as its parent node; determine whether the new cumulative cost is less than the original cumulative cost of the neighboring node: if it is, disconnect the neighboring node from its original parent node, update the parent node of the neighboring node to the new sampled node, and recursively update the cost of the neighboring node and all its child nodes so that the entire search tree gradually converges to the optimal path that satisfies the time-series task; if it is not, keep the original topological connection relationship and cost of the neighboring node unchanged, and continue to traverse the next neighboring node in the neighborhood until all neighboring nodes in the neighborhood have been detected. The condition that the cumulative cost of a new sampling node connecting to each potential parent node is less than the cumulative cost of the parent node is expressed by the following formula (3): (3) in, This represents the cumulative cost when a new sampling node connects to each potential parent node; This represents the cumulative cost of the parent node.

[0048] Specifically, for STL tasks that include multiple sub-task series, this embodiment of the invention adopts a phased planning strategy.

[0049] S43. Check if a node exists in the search tree. Entering the target region specified by the current subtask and from the root node to the node trajectory The termination conditions are met. If the termination condition is met, the optimal path for the current stage is extracted, and the path planning is completed; where the termination condition is a signal timing logic constraint.

[0050] In one feasible implementation, if the termination condition is met, the optimal path for the current stage will be extracted. As the starting state for the next stage of subtasks, repeat steps S2 to S4 until all subtask sequences are completed.

[0051] In particular, this invention, for sequential tasks, employs segmented planning and termination determination based on region coverage to achieve efficient solution and logical connection of complex multi-stage tasks.

[0052] This invention employs a PointNet++ network to generate a guiding point cloud, which performs intelligent pruning on the high-dimensional search space, enabling a rapid-exploring random tree star schema algorithm. RRT It can focus on high-probability feasible areas, avoiding blind sampling in invalid areas and significantly shortening planning time in complex environments.

[0053] The embodiments of the present invention can effectively identify and guide the search tree through the "narrow passage" in the environment by adopting a neural guidance strategy, which solves the problem that traditional sampling algorithms are difficult to converge in narrow feasible regions and greatly improves the planning success rate.

[0054] This invention introduces the STL temporal robustness integral as part of the cost function, enabling the algorithm to actively search for paths that are farther from obstacles and more in line with temporal logic constraints. This generates a planned trajectory that is both efficient and has a high safety margin, effectively improving the robot's adaptability in dynamic and uncertain environments.

[0055] The embodiments of the present invention are not only applicable to two-dimensional planar navigation, but can also be extended to three-dimensional spatial path planning, and can be deployed on various mobile robot systems equipped with LiDAR and embedded computing platforms (such as NVIDIA Jetson series), and have strong engineering practical value.

[0056] This invention proposes a mobile robot path planning method that integrates signal temporal logic robustness with neural-guided sampling. Compared with existing technologies, it has the following significant advantages: (1) Significantly improved planning efficiency: Addressing the "narrow channel" problem in the state space caused by complex temporal logic tasks, this embodiment of the invention innovatively introduces a PointNet++ deep neural network for environmental feature extraction and sampling guidance. This network can directly predict the sampling probability distribution (guiding point cloud) that satisfies the temporal task from unstructured environmental point clouds, thereby achieving intelligent pruning of the invalid search space. In complex maze and sequence task scenarios, this embodiment of the invention significantly improves planning efficiency compared to traditional STL-RRT. The algorithm reduces planning time and significantly improves computational efficiency, enabling it to meet the real-time planning needs of mobile robots.

[0057] (2) Improved planning success rate and path quality: Through a hybrid sampling strategy, this embodiment of the invention utilizes the prior knowledge of neural networks to quickly traverse narrow regions while preserving the probabilistic completeness of full-space sampling, effectively solving the problem of traditional algorithms' difficulty in convergence under complex constraints and improving the planning success rate. Simultaneously, since the guiding point cloud is usually located on the optimal homotopy class, the generated path length is shortened.

[0058] (3) Enhanced path timing safety (robustness): The embodiments of this invention redefine RRT The cost function uses the temporal robustness integral of STL as the core optimization metric. This allows the planner to move beyond simply finding the geometrically shortest path and actively search for paths that are farther from obstacles, deeper into the core of the target area, and more in line with temporal logic constraints. This "robust-oriented" optimization mechanism gives the generated trajectory a higher safety margin, effectively addressing positioning errors and control noise in actual robot operation.

[0059] Figure 4 This is a block diagram of a path planning device based on signal timing logic and neural guided sampling, provided in an embodiment of the present invention. This device is used in a path planning method based on signal timing logic and neural guided sampling. (Refer to...) Figure 4 The device includes an acquisition unit 410, a construction unit 420, a detection unit 430, and an output and execution unit 440. Wherein: The acquisition unit 410 is used to acquire the working environment information of the mobile robot in the intelligent warehousing and logistics distribution task, and model it into a two-dimensional map. The working environment information includes static environment features and dynamic state features. The static environment features include global physical boundaries, static obstacle positions and geometric shapes. The dynamic state features include the robot's current pose state and the free space region after removing obstacles and considering the robot's safe expansion radius. Sampling is performed within the free space region to discretize the two-dimensional map into an environmental point cloud dataset. The building unit 420 is used to define the signal timing logic task to be executed. It inputs the environmental point cloud dataset into a pre-trained PointNet++ deep network and outputs the probability score of each point belonging to the optimal path neighborhood. It compares the probability score of each point belonging to the optimal path neighborhood with a pre-set threshold, filters out points that are greater than the pre-set threshold, and builds the guide point cloud dataset. The detection unit 430 is used to randomly sample and generate candidate nodes based on the guided point cloud dataset using a fast exploratory random tree star schema algorithm; traverse all nodes in the tree based on the candidate nodes and find the node that is closest to the candidate node; expand the nearest node into a new node according to the set expansion step size; perform collision detection based on the new node, and if there is no collision, select the parent node through the constructed hybrid cost function. The output and execution unit 440 is used to add new nodes to the tree and recursively update the cost of all its child nodes by calculating the cumulative cost of the new node and its parent node. If it is determined that the trajectory to be executed satisfies the signal timing logic constraints of all subtasks in sequence, the optimal path that satisfies the global signal timing logic constraints is extracted, output, and executed to complete the path planning.

[0060] Optionally, the PointNet++ based deep network includes: an encoder, a decoder, a fully connected layer, and a classifier; The encoder consists of a multi-level set of abstraction layers, which is used to perform stepwise downsampling and feature extraction on the input environmental point cloud data with feature labels, and output the local geometric features and global topological features of the environment. The decoder consists of a multi-level feature propagation layer, which is used to upsample deep features and map them back to the original point cloud resolution.

[0061] Optionally, the step of inputting the environmental point cloud dataset into a pre-trained PointNet++ deep network and outputting the probability score of each point belonging to the optimal path neighborhood includes: The environmental point cloud dataset is input into a pre-trained PointNet++ deep network, and point cloud features are constructed using predefined map free space points and starting point neighborhood masks. The point cloud features are input into the encoder, and the multi-level ensemble abstraction layer performs step-by-step sampling and feature extraction to output local and global features. Local and global features are input into the decoder, upsampled through a multi-level feature propagation layer and mapped back to the original point cloud resolution, and the processed feature points are output. The processed feature points are input into a fully connected layer, and a classifier is used to generate a probability score for each point belonging to the optimal path neighborhood and a path point prediction mask.

[0062] Optionally, the step of randomly sampling and generating candidate nodes based on the guided point cloud dataset using a fast exploratory random tree star schema algorithm includes: Set neural sampling probability threshold Generate a random number that follows a uniform distribution in the range [0,1]. ;like Then, a point is randomly selected directly from the guide point cloud dataset as the sampling point. ;like Then, uniform random sampling is performed within the free space of the entire working environment to obtain sampling points. .

[0063] Optionally, the hybrid cost function of the path is represented by the following formula (1): (1) in, Indicates the parent node; Indicates a new node; Indicates Euclidean distance; Indicates the first weighting coefficient; This represents the second weighting coefficient; Represents the STL timing robustness integral on the path segment; The mixed cost function represents the path.

[0064] Optionally, the STL timing robustness integral on the path segment is expressed by the following formula (2): (2) in, Indicates the parent node; Indicates a new node; Indicates the time step between two nodes; This indicates that the parent node satisfies the current signal timing logic task. Spatial robustness; This indicates that the new node satisfies the current signal timing logic task. Spatial robustness.

[0065] Optionally, the output and execution unit 440 is used for: Within the neighborhood radius of the new sampling node, traverse all new sampling nodes and calculate the cumulative cost when the new sampling node is connected to each potential parent node. Select the node with the smallest cumulative cost as the parent node of the new sampling node. Add the new sampled node to the search tree and find other neighboring nodes in its neighborhood; calculate the new cumulative cost generated when the new sampled node is connected to each neighboring node as its parent node; determine whether the new cumulative cost is less than the original cumulative cost of the neighboring node: if it is, disconnect the neighboring node from its original parent node, update the parent node of the neighboring node to the new sampled node, and recursively update the cost of the neighboring node and all its child nodes so that the entire search tree gradually converges to the optimal path that satisfies the time-series task; if it is not, keep the original topological connection relationship and cost of the neighboring node unchanged, and continue to traverse the next neighboring node in the neighborhood until all neighboring nodes in the neighborhood have been detected. Check if a node exists in the search tree. Entering the target region specified by the current subtask and from the root node to the node trajectory The termination conditions are met. If the termination condition is met, the optimal path for the current stage is extracted, and the path planning is completed; where the termination condition is a signal timing logic constraint.

[0066] This invention significantly improves planning efficiency: Addressing the "narrow channel" problem in the state space caused by complex temporal logic tasks, this invention innovatively introduces a PointNet++ deep neural network for environmental feature extraction and sampling guidance. This network can directly predict the sampling probability distribution (guiding point cloud) that satisfies the temporal task from unstructured environmental point clouds, thereby achieving intelligent pruning of the invalid search space. In complex maze and sequence task scenarios, this invention significantly improves planning efficiency compared to traditional STL-RRT. The algorithm reduces planning time and significantly improves computational efficiency, enabling it to meet the real-time planning needs of mobile robots.

[0067] This invention improves planning success rate and path quality: By employing a hybrid sampling strategy, this invention leverages the prior knowledge of neural networks to quickly traverse narrow regions while preserving the probabilistic completeness of full-space sampling. This effectively solves the problem of traditional algorithms struggling to converge under complex constraints, thus improving planning success rate. Simultaneously, since the guiding point cloud is typically located on the optimal homotopy class, the generated path length is shortened.

[0068] This invention enhances the temporal safety (robustness) of paths: This invention redefines RRT. The cost function uses the temporal robustness integral of STL as the core optimization metric. This allows the planner to move beyond simply finding the geometrically shortest path and actively search for paths that are farther from obstacles, deeper into the core of the target area, and more in line with temporal logic constraints. This "robust-oriented" optimization mechanism gives the generated trajectory a higher safety margin, effectively addressing positioning errors and control noise in actual robot operation.

[0069] Figure 5 This is a schematic diagram of the structure of a path planning device based on signal timing logic and neural guided sampling provided in an embodiment of the present invention, as shown below. Figure 5 As shown, the path planning device based on signal timing logic and neural guided sampling can include the above-mentioned... Figure 4 The illustrated path planning device is based on signal timing logic and neural guided sampling. Optionally, the path planning device 510 based on signal timing logic and neural guided sampling may include a first processor 2001.

[0070] Optionally, the path planning device 510 based on signal timing logic and neural guided sampling may also include a memory 2002 and a transceiver 2003.

[0071] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0072] The following is combined with Figure 5The various components of the path planning device 510 based on signal timing logic and neural guided sampling are described in detail below: The first processor 2001 is the control center of the path planning device 510 based on signal timing logic and neural guided sampling. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0073] Optionally, the first processor 2001 can execute various functions of the path planning device 510 based on signal timing logic and neural guided sampling by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0074] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 5 CPU0 and CPU1 are shown in the diagram.

[0075] In a specific implementation, as one example, the path planning device 510 based on signal timing logic and neural guided sampling may also include multiple processors, for example... Figure 5 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0076] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0077] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently, and may be connected via an interface circuit of the signal timing logic and neural guided sampling path planning device 510. Figure 5 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0078] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0079] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 5 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0080] Optionally, the transceiver 2003 can be integrated with the first processor 2001 or exist independently, and can be connected to the interface circuit of the path planning device 510 based on signal timing logic and neural guided sampling. Figure 5 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0081] It should be noted that, Figure 5 The structure of the path planning device 510 based on signal timing logic and neural guided sampling shown in the figure does not constitute a limitation on the router. Actual path planning devices based on signal timing logic and neural guided sampling may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0082] Furthermore, the technical effects of the path planning device 510 based on signal timing logic and neural guided sampling can be referred to the technical effects of the path planning method based on signal timing logic and neural guided sampling described in the above method embodiments, and will not be repeated here.

[0083] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or it may be any conventional processor, etc.

[0084] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0085] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0086] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0087] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0088] It should be understood that, in various embodiments of the present invention, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0089] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0090] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0091] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0092] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0093] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0094] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0095] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A path planning method based on signal timing logic and neural guided sampling, characterized in that, The method includes: S1. Obtain the working environment information of the mobile robot in the intelligent warehousing and logistics distribution task, and model it as a two-dimensional map. The working environment information includes static environment features and dynamic state features. The static environment features include the global physical boundary, the position and geometry of static obstacles, and the dynamic state features include the robot's current pose state and the free space region after removing obstacles and considering the robot's safe expansion radius. Sampling is performed within the free space region to discretize the two-dimensional map into an environmental point cloud dataset. S2. Define the signal timing logic task to be executed. Input the environmental point cloud dataset into the pre-trained PointNet++ deep network and output the probability score of each point belonging to the optimal path neighborhood. Compare the probability score of each point belonging to the optimal path neighborhood with a pre-set threshold, filter out points that are greater than the pre-set threshold, and construct the guide point cloud dataset. S3. Based on the guided point cloud dataset, candidate nodes are randomly sampled and generated using the fast exploratory random tree star schema algorithm; all nodes in the tree are traversed based on the candidate nodes, and the nearest node is found; the nearest node is expanded into a new node according to the set expansion step size; collision detection is performed based on the new node, and if there is no collision, the parent node is selected by constructing the hybrid cost function of the path. S4. Add the new node to the tree, and recursively update the cost of all its child nodes by calculating the cumulative cost of the new node and its parent node. If it is determined that the trajectory to be executed satisfies the signal timing logic constraints of all subtasks in sequence, then extract and output the optimal path that satisfies the global signal timing logic constraints and execute it to complete the path planning.

2. The path planning method based on signal timing logic and neural guided sampling according to claim 1, characterized in that, The PointNet++ based deep network includes: an encoder, a decoder, a fully connected layer, and a classifier; The encoder consists of a multi-level set of abstraction layers, which is used to perform step-by-step downsampling and feature extraction on the input environmental point cloud data with feature labels, and output the local geometric features and global topological features of the environment. The decoder consists of a multi-level feature propagation layer, which is used to upsample deep features and map them back to the original point cloud resolution.

3. The path planning method based on signal timing logic and neural guided sampling according to claim 2, characterized in that, S2 inputs the environmental point cloud dataset into a pre-trained PointNet++ deep network and outputs a probability score for each point belonging to the optimal path neighborhood, including: S21. Input the environmental point cloud dataset into a pre-trained PointNet++ deep network, and construct point cloud features using predefined map free space points and starting point neighborhood masks; S22. Input the point cloud features into the encoder, perform step-by-step sampling and feature extraction through a multi-level ensemble abstraction layer, and output local and global features; S23. Input the local and global features into the decoder, upsample them through a multi-level feature propagation layer and map them back to the original point cloud resolution, and output the processed feature points. S24. Input the processed feature points into the fully connected layer and generate the probability score and path point prediction mask for each point belonging to the optimal path neighborhood through the classifier.

4. The path planning method based on signal timing logic and neural guided sampling according to claim 1, characterized in that, The S3 dataset, based on the guided point cloud dataset, uses a fast exploratory random tree star schema algorithm to randomly sample and generate candidate nodes, including: Set neural sampling probability threshold Generate a random number that follows a uniform distribution in the range [0,1]. ;like Then, a point is randomly selected directly from the guide point cloud dataset as the sampling point. ;like Then, uniform random sampling is performed within the free space of the entire working environment to obtain sampling points. .

5. The path planning method based on signal timing logic and neural guided sampling according to claim 1, characterized in that, The hybrid cost function of the path is expressed by the following formula (1): （1） in, Indicates the parent node; Indicates a new node; Indicates Euclidean distance; Indicates the first weighting coefficient; This represents the second weighting coefficient; Represents the STL timing robustness integral on the path segment; This represents the path hybrid cost function (incremental cost) of the new node relative to its parent node.

6. The path planning method based on signal timing logic and neural guided sampling according to claim 5, characterized in that, The STL timing robustness integral on the path segment is expressed by the following formula (2): （2） in, Indicates the parent node; Indicates a new node; Indicates the time step between two nodes; This indicates that the parent node satisfies the current signal timing logic task. Spatial robustness; This indicates that the new node satisfies the current signal timing logic task. Spatial robustness.

7. The path planning method based on signal timing logic and neural guided sampling according to claim 1, characterized in that, S4 adds the new node to the tree and recursively updates the cost of all its child nodes by calculating the cumulative cost of the new node and its parent node; if it is determined that the trajectory to be executed satisfies the signal timing logic constraints of all subtasks in sequence, the optimal path that satisfies the global signal timing logic constraints is extracted, output, and executed to complete the path planning, including: S41. Within the neighborhood radius of the new sampling node, traverse all new sampling nodes and calculate the cumulative cost when the new sampling node is connected to each potential parent node. Select the node with the smallest cumulative cost as the parent node of the new sampling node. S42. Add the new sampled node to the search tree and find other neighboring nodes in its neighborhood; calculate the new cumulative cost generated when the new sampled node is connected to each neighboring node as its parent node; determine whether the new cumulative cost is less than the original cumulative cost of the neighboring node: if it is, disconnect the neighboring node from its original parent node, update the parent node of the neighboring node to the new sampled node, and recursively update the cost of the neighboring node and all its child nodes so that the entire search tree gradually converges to the optimal path that satisfies the time-series task; if it is not, keep the original topological connection relationship and cost of the neighboring node unchanged, and continue to traverse the next neighboring node in the neighborhood until all neighboring nodes in the neighborhood have been detected. S43. Check if a node exists in the search tree. Entering the target region specified by the current subtask and from the root node to the node trajectory The termination condition is met If the termination condition is met, the optimal path for the current stage is extracted, and the path planning is completed; where the termination condition is a signal timing logic constraint.

8. A path planning device based on signal timing logic and neural guided sampling, wherein the path planning device based on signal timing logic and neural guided sampling is used to implement the path planning method based on signal timing logic and neural guided sampling as described in any one of claims 1-7, characterized in that, The device includes: The acquisition unit is used to acquire the working environment information of the mobile robot in the intelligent warehousing and logistics distribution task. The working environment information includes static environment features and dynamic state features. The static environment features include the global physical boundary, the position and geometry of static obstacles, and the dynamic state features include the robot's current pose state and the free space region after removing obstacles and considering the robot's safe expansion radius. Sampling is performed within the free space region to discretize the two-dimensional map into an environmental point cloud dataset. The building unit is used to define the signal timing logic task to be executed. It inputs the environmental point cloud dataset into a pre-trained PointNet++ deep network and outputs the probability score of each point belonging to the optimal path neighborhood. The probability score of each point belonging to the optimal path neighborhood is compared with a pre-set threshold, and points that are greater than the pre-set threshold are selected to build the guide point cloud dataset. The detection unit is used to randomly sample and generate candidate nodes based on the guided point cloud dataset using a fast exploratory random tree star schema algorithm; it traverses all nodes in the tree based on the candidate nodes and finds the node that is closest to the candidate node; it expands the nearest node into a new node according to the set expansion step size; it performs collision detection based on the new node, and if there is no collision, it selects the parent node by constructing the path using a hybrid cost function; The output and execution unit is used to add new nodes to the tree and recursively update the costs of all its child nodes by calculating the cumulative cost of the new node and its parent node. If it is determined that the trajectory to be executed satisfies the signal timing logic constraints of all subtasks in sequence, the optimal path that satisfies the global signal timing logic constraints is extracted, output, and executed to complete the path planning.

9. A path planning device based on signal timing logic and neural guided sampling, characterized in that, The path planning device based on signal timing logic and neural guided sampling includes: processor; A memory storing computer-readable instructions that, when executed by the processor, implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 7.

Images