A robot motion planning method and device inspired by spatial relation memory

Abstract

The present invention relates to the field of robot motion planning technology, and in particular to a robot motion planning method and device inspired by spatial relationship memory. The method comprises: constructing a first random geometric graph in the robot's configuration space using a selective sampling strategy through a K-NN algorithm based on motion starting position information, motion target position information, and map information; inputting the first random geometric graph and obstacle position information into an intelligent edge selector for edge screening based on a preset confidence threshold to obtain a second random geometric graph; calculating heuristic values ​​based on the second random geometric graph using a memory perception predictor to obtain node heuristic values; and searching for an optimal path using a differentiable module based on the node heuristic values ​​and the second random geometric graph to obtain a second optimal path. The present invention is an efficient and accurate robot motion planning method inspired by the spatial relationship memory of the human brain.

Description

Technical Field

[0001] The present invention relates to the technical field of robot motion planning, and in particular to a robot motion planning method and device inspired by spatial relationship memory. Background Art

[0002] Motion planning is a key technology in robotics systems, with significant application prospects in various fields, including transportation and healthcare. The goal of motion planning is to find a collision-free path connecting a starting point and a destination in a decision space, ideally minimizing the cost of that path. However, real-world decision spaces often contain a large number of continuous variables and high-dimensional, unstructured spaces, which greatly complicates motion planning. Therefore, developing fast and reliable motion planning methods in high-dimensional continuous spaces is crucial for advancing their practical applications.

[0003] Classic motion planning solutions can be roughly divided into two categories: search-based methods and sampling-based methods. Search-based methods conceptualize the planning problem as a graph search problem. Sampling-based methods, including fast exploration random trees and probabilistic roadmaps, construct exploration trees connecting the starting and goal points by randomly sampling in the configuration space. However, classical methods are often inefficient when dealing with high-dimensional planning problems. In high-dimensional continuous environments, learning-based planning methods are a key means to improve motion planning performance. Existing planning methods still face the following challenges when facing high-dimensional continuous environments.

[0004] Traditional planning methods construct random geometric graphs based on a uniform sampling strategy. This means that the entire space is explored indiscriminately, failing to fully utilize the structural information inherent in the actual problem. This can lead to unnecessary node exploration or the presence of inappropriate edges and nodes, making them ineffective for solving new problems and unable to generalize effectively. When faced with fast or close obstacles, robots must make quick and accurate decisions to avoid them. Traditional planning methods often focus on node position information and handle each task independently, failing to fully leverage previous experience and established data models.

[0005] The human brain possesses a natural advantage in decision-making and reasoning, with spatial relational memory being crucial in this process. Studies have shown that the human brain can transform decision-making problems into a graph structure corresponding to sensory observations. Relational memory integrates sensory observations with their structural relationships, thereby facilitating sensory reasoning. On the one hand, humans can make complex inferences from sparse observations and make rapid decisions by integrating prior experience with current environmental information. This ability is associated with the hippocampus-entorhinal cortex system. On the other hand, spatial memory is crucial for human survival. Humans use "cognitive maps" to structure their understanding of the world, a process closely associated with the hippocampus. Therefore, to address these two challenges, the present invention designs a brain-inspired planner based on a graph neural network (GNN) to improve planning efficiency and scalability. This planner, called Brainy Motion Planning (BrainyMP), is designed.

[0006] In recent years, learning-based planning methods have become the most promising approaches for improving the performance of high-dimensional continuous motion planning. By leveraging the powerful learning and representation capabilities of neural networks, these planning methods allow robots to learn patterns in high-dimensional spaces and the behavioral trajectories of expert planners from data. Consequently, strategies in the planning process, such as sampling mechanisms, can be optimized to achieve excellent performance in high-dimensional continuous spaces. Learning-based methods treat motion planning as a sequential decision-making problem, which can often be solved through reinforcement learning. Previous work has proposed various neural network-based motion planners, including those based on convolutional neural networks, recurrent neural networks, and graph neural networks.

[0007] GNN-based planners, such as GNN-Explorer and GraphMP, have demonstrated remarkable capabilities in handling high-dimensional continuous planning tasks. By accessing random geometric graphs formed by uniform sampling, these planners are able to capture the geometric patterns of the environment, enabling more efficient exploration without encoding the entire workspace. Although GNN-based planners perform well in high-dimensional continuous tasks, they still have some limitations: when constructing the initial random geometric graph (RGG), the process of uniformly sampling nodes ignores the topology of the environment, which may lead to unnecessary node exploration; in the process of forming edges in the RGG, there may be edges that collide with obstacles, thereby reducing the reliability of planning; when extracting key graph patterns from the model, existing methods mainly encode the positional relationships of nodes, while ignoring the structural relationships between nodes, which hinders the improvement of the model's expressive power.

[0008] In the existing technology, there is a lack of an efficient and accurate robot motion planning method inspired by the spatial memory of the human brain. Summary of the Invention

[0009] To address the technical issues of low node exploration efficiency and poor reliability of motion route planning in existing technologies, the present invention provides a robot motion planning method and device inspired by spatial relationship memory. The technical solution is as follows:

[0010] In one aspect, a robot motion planning method inspired by spatial relationship memory is provided, the method being implemented by a robot motion planning device, the method comprising:

[0011] Obtain movement starting position information, movement target position information, obstacle position information and map information;

[0012] Based on the selective sampling strategy, according to the movement starting position information, the movement target position information and the map information, a first random geometric graph is constructed in the robot's configuration space by using a K-NN algorithm;

[0013] Based on a preset confidence threshold, the first random geometric graph and the obstacle position information are input into the smart edge selector for edge screening to obtain a second random geometric graph;

[0014] performing heuristic value calculation using the memory-aware predictor according to the second random geometric graph to obtain a node heuristic value;

[0015] Based on differentiable The module searches for an optimal path using a shortcut path exploration method according to the node heuristic value and the second random geometric graph to obtain a second optimal path.

[0016] In another aspect, a robot motion planning device inspired by spatial relational memory is provided, wherein the device is applied to a robot motion planning method inspired by spatial relational memory, and the device comprises:

[0017] An information acquisition module is used to obtain movement starting position information, movement target position information, obstacle position information and map information;

[0018] A first random geometric graph construction module is configured to construct a first random geometric graph in the robot's configuration space using a K-NN algorithm based on a selective sampling strategy and according to the motion starting position information, the motion target position information, and the map information;

[0019] a second random geometric graph construction module, configured to input the first random geometric graph and the obstacle position information into the smart edge selector for edge screening based on a preset confidence threshold, thereby obtaining a second random geometric graph;

[0020] a heuristic value calculation module, configured to calculate a heuristic value based on the second random geometric graph by using the memory-aware predictor to obtain a node heuristic value;

[0021] Optimal path acquisition module for differentiable The module searches for an optimal path using a shortcut path exploration method according to the node heuristic value and the second random geometric graph to obtain a second optimal path.

[0022] On the other hand, a robot motion planning device is provided, which includes: a processor; a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, any one of the above-mentioned robot motion planning methods inspired by spatial relationship memory is implemented.

[0023] On the other hand, a computer-readable storage medium is provided, wherein the storage medium stores at least one instruction, and the at least one instruction is loaded and executed by a processor to implement any one of the above-mentioned robot motion planning methods inspired by spatial relationship memory.

[0024] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0025] The present invention proposes a robot motion planning method inspired by spatial relational memory. In the human brain's decision-making and relational reasoning processes, the brain mechanisms related to spatial relational memory show obvious advantages. Based on this, a new brain-like motion planning based on graph neural networks is designed to improve the performance of robot planning in high-dimensional continuous environments. When constructing a random geometric graph, a new random geometric graph generation method is used to make the sampled nodes closer to the optimal solution, thereby improving the search efficiency; through the intelligent edge selector, the edges with the highest confidence in the random geometric graph are retained, avoiding the negative impact of unreasonable edges on motion planning; through the memory perception predictor, the brain mechanism of integrating sensory observation information with relational structure for sensory reasoning is imitated in the brain's relational memory, and the introduction of subgraph structure into node features improves the graph representation ability of the graph neural network and enhances the overall learning effect of the model. The present invention is an efficient and accurate robot motion planning method inspired by the spatial relational memory of the human brain. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1This is a flow chart of a robot motion planning method inspired by spatial relationship memory provided by an embodiment of the present invention;

[0028] Figure 2 This is a schematic diagram of the structure of a smart edge selector provided by an embodiment of the present invention;

[0029] Figure 3 1 is a schematic diagram of the structure of a memory perception predictor provided by an embodiment of the present invention;

[0030] Figure 4 This is a block diagram of a robot motion planning device inspired by spatial relationship memory provided by an embodiment of the present invention;

[0031] Figure 5 It is a structural schematic diagram of a robot motion planning device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0033] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0034] In the embodiments of the present invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same. The terms "of," "corresponding," and "corresponding" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same.

[0035] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0036] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0037] The embodiment of the present invention provides a robot motion planning method inspired by spatial relationship memory, which can be implemented by a robot motion planning device, which can be a terminal or a server. Figure 1The flowchart of the robot motion planning method inspired by spatial relationship memory is shown in FIG. The processing flow of the method may include the following steps:

[0038] S1. Obtain movement starting position information, movement target position information, obstacle position information and map information.

[0039] In one feasible implementation, in a high-dimensional continuous robot motion scenario, the robot's ROS system publishes information to obtain the starting and target location information. The robot's onboard laser sensor scans the scene to obtain obstacle location information and current map information.

[0040] S2. Based on the selective sampling strategy, according to the motion starting position information, the motion target position information and the map information, a first random geometric graph is constructed in the robot's configuration space by using the K-NN algorithm.

[0041] Optionally, based on a selective sampling strategy, according to the movement starting position information, the movement target position information and the map information, a first random geometric graph is constructed in the robot's configuration space by a K-NN algorithm, including:

[0042] Map the robot's position based on the motion starting position information, motion target position information, and map information to obtain the robot's positioning information;

[0043] Based on the robot's configuration space, a node set is constructed using a selective sampling strategy according to the movement starting position information, map information and robot positioning information;

[0044] According to the node set, use the K-NN algorithm to search for neighboring nodes and obtain the neighboring node set;

[0045] Edge connections are performed based on the node set and the neighboring node set to obtain a first random geometric graph.

[0046] In a feasible implementation, the robot positioning information of the robot is determined based on the accurate spatial position description of obstacles in the robot's environment and matching it with a preset map.

[0047] During the planning process, this invention addresses the problem of random sampling ignoring environmental information when constructing an RGG. The robot generates a batch of nodes by sampling the area between the starting point and the goal with probability α, while also uniformly sampling across the entire configuration space with probability (1-α). This method can flexibly switch between biased and uniform sampling by adjusting the hyperparameter α, balancing efficiency and exploration while maintaining probabilistic completeness and reducing the exploration of redundant nodes.

[0048] The present invention uses the K-Nearest Neighbors (K-NN) algorithm to search for neighboring nodes and construct a first random geometric graph. This first random geometric graph defines feasible paths for the robot in the configuration space. This graph allows for efficient collision-free planning from the initial configuration to the target configuration.

[0049] S3. Based on a preset confidence threshold, the first random geometric graph and the obstacle position information are input into the smart edge selector for edge screening to obtain a second random geometric graph.

[0050] Among them, the smart edge selector includes the encoding layer, transformer layer and hidden layer;

[0051] The encoding layer is built based on the attention mechanism;

[0052] The hidden layer is constructed according to the multi-layer perceptron model structure.

[0053] In a feasible implementation, the present invention adopts the architecture of the smart edge selector as follows Figure 2 As shown in the figure, the encoding layer includes three encoders that respectively extract features of the nodes and edges of the first random geometric graph and the positions of obstacles; the transformer layer includes two transformer modules; and the hidden layer consists of three multi-layer perceptrons.

[0054] In the training learning phase, the smart edge selector uses the training data for each planning problem, which consists of obstacles, starting vertices, and the target vertex The k-NN graph is constructed by uniformly sampling nodes in the configuration space. ,in . Through binary cross entropy (BCE) loss To train, we minimize the confidence difference between the estimated value and the true label. Dijkstra algorithm is used as an expert planner to provide the true label for the training problem. .

[0055] Optionally, based on a preset confidence threshold, the first random geometric graph and the obstacle position information are input into a smart edge selector for edge screening to obtain a second random geometric graph, including:

[0056] Obtaining node information and edge information according to the first random geometric graph;

[0057] Input the node information, edge information and obstacle position information into the encoding layer for feature extraction to obtain the first node feature, the first edge feature and the obstacle position feature;

[0058] According to the first node feature, the first edge feature and the obstacle position feature, encoding processing is performed through the transformer layer to obtain the node-obstacle feature and the edge-obstacle feature;

[0059] Input the node-obstacle features and edge-obstacle features into the hidden layer for confidence calculation to obtain the edge confidence;

[0060] The edges of the first random geometric graph are screened according to the edge confidence and a preset confidence threshold to obtain a second random geometric graph.

[0061] In a feasible implementation, the first random geometric graph RGG As the input of the encoder, vertices and edges are encoded into the latent space. Each vertex and each edge The embedding representation of and ,in and are two different two-layer MLPs. In iterative obstacle coding The obstacle encoding process in the t-th iteration can be expressed as follows (1):

[0062] (1)

[0063] in, and are the outputs of vertex and edge after iterative t-1 obstacle encoding, Represents layer normalization, B is the output of the obstacle after passing through the encoder, is the MLP layer. are the outputs of vertices and edges after t iterations of obstacle encoding, respectively, where 、 for 、 The intermediate result output by the transformer layer during the iterative update process. is a typical attention function, where Represents key, query, and value respectively.

[0064] RGG Each edge The confidence of can be evaluated using obstacle information. Specifically, the obstacle encoder is passed through T iterations. , the information of RGG and obstacles is encoded into each edge The concatenated embedding vector of .side The confidence of Calculate, expressed as . Retain the edges whose confidence is higher than the pre-set threshold and obtain the second random geometric graph .

[0065] S4. Calculate the heuristic value of the node by using the memory-aware predictor according to the second random geometric graph to obtain the heuristic value of the node.

[0066] The memory-aware predictor consists of a subgraph extraction layer, a graph encoding layer, a linear layer, a pooling layer, and an output layer.

[0067] In a feasible implementation, the memory-aware predictor proposed in the present invention is constructed based on a graph neural network, such as Figure 3 As shown in Figure 2, the subgraph extraction layer extracts the Ego and Cut subgraphs, as well as vertex and edge features from the second random geometric graph. The graph encoding layer includes three encoders, one for encoding the Ego subgraph, the other for encoding the vertex and edge features of the second most taboo geometric graph. The output layer is constructed using three multilayer perceptrons.

[0068] Memory perceptron uses differentiable Modules, memory-aware perceptual predictors can be jointly trained in an end-to-end manner. During the learning phase, the training data consists of graph search problem instances. Composition, of which are all nodes and edges in the input RGG, and are the starting vertex and the target vertex respectively, is a binary vector that marks the vertices in the optimal path as 1. The training loss is defined as follows (2):

[0069] (2)

[0070] in, It is differentiable A vector of closed lists of module estimates, is the binary vector provided by the expert planner for the i-th training problem.

[0071] The training loss guides the path closer to the optimal path by penalizing the over-explored vertices in the path. Therefore, by minimizing the training function, the differentiable The optimal path can be searched efficiently.

[0072] Optionally, performing heuristic value calculation using a memory-aware predictor according to the second random geometric graph to obtain a node heuristic value includes:

[0073] Input the subgraph extraction layer according to the second random geometric graph to perform subgraph extraction, and obtain the Ego subgraph and the Cut subgraph;

[0074] Based on the preset random walk process, the probability distribution is calculated according to the Ego subgraph and the Cut subgraph to obtain the random walk probability of the Ego subgraph and the random walk probability of the Cut subgraph;

[0075] Input the second random geometric graph into the graph encoding layer for feature extraction to obtain second node features and second edge features;

[0076] Input the random walk probability of the Ego subgraph and the random walk probability of the Cut subgraph into the graph encoding layer for encoding processing to obtain the first hidden representation vector feature and the second hidden representation vector feature;

[0077] Splicing the first hidden representation vector feature and the second hidden representation vector feature to the vertex vector through a linear layer to obtain a first spliced ​​vertex vector;

[0078] Input the first spliced ​​vertex vector into the pooling layer for aggregation iteration to obtain the second spliced ​​vertex vector;

[0079] The second spliced ​​vertex vector and the second edge feature vector are input into the output layer to calculate the heuristic value, so as to obtain the node heuristic value of the second random geometric graph.

[0080] In a feasible implementation, the present invention adopts two subgraph extraction strategies: Ego subgraph and Cut subgraph. Defines a k-hop EgoNetwork with node v as the root, including the neighbor nodes within k-hop of node v ,in, Represents the set of nodes within k-hop distance from the root node v. Cut subgraphs are obtained from the original graph by continuously and selectively removing edges.

[0081] Extracted subgraph The i-th vertex in The random walk return probability encoding of is defined as follows (3):

[0082] (3)

[0083] in, Represents the subgraph middle root vertex The return probability of the random walk of s steps is given by . The return probability of the random walk of Ego subgraph and Cut subgraph is expressed as and Subsequently, two linear layers are applied to encode the random walk return probabilities of the Ego subgraph and the Cut subgraph into hidden representation vectors, which are and .

[0084] Given and the target vertex , then the vertex Features and the edge Features The definition of is as follows:

[0085] (4)

[0086] in, and are two different two-layer MLPs.

[0087] The Ego subgraph and the Cut subgraph are mapped to the same The same latent space. Then, the hidden representations of the two types of subgraphs are concatenated with the vertex features to obtain the subgraph information injection features, which are and , defined as follows (5):

[0088] (5)

[0089] Graph neural networks are constructed by learning from the neighbors of each vertex Aggregate local information, iteratively update Ego subgraph and Cut subgraph information to update node features. The update method of Ego message transmission channel is as follows (6):

[0090] (6)

[0091] in, and There are three different two-layer MLPs. The Cut messaging channel is the same as the Ego channel.

[0092] After L iterations, the information injection features of Ego subgraph and Cut subgraph are expressed as and Finally, the complete graph representation is obtained through the pooling operation as follows (7)

[0093] (7)

[0094] Among them, the POOL operation is a global pooling function used to aggregate information of all vertices.

[0095] vertex The perceptual prediction H can be expressed as , It is a three-layer MLP.

[0096] S5, based on differentiability The module uses a shortcut path exploration method to search for the optimal path based on the node heuristic value and the second random geometric graph to obtain the second optimal path.

[0097] Optionally, based on differentiable Module, based on the node heuristic value and the second random geometric graph, uses the shortcut path exploration method to search for the optimal path and obtain the second optimal path, including

[0098] Input the node heuristic value into the differentiable A module, which performs an optimal path search in the second random geometric graph to obtain a first optimal path;

[0099] According to the first optimal path, redundant nodes are deleted through a shortcut path exploration method to obtain a second optimal path.

[0100] In one possible implementation, the heuristic value is used as a differentiable The input of the module is used to search for the optimal path in the second random geometric graph RGG'.

[0101] Differentiable The module's expansion process for new nodes is as follows (8):

[0102] (8)

[0103] in Indicates the selected node, the initial open list represents all nodes of a random geometric graph, represents the total accumulated path cost from the starting point to the current node, and Indicates the estimated cost of the current node from the end point, which is replaced by the heuristic value H. As the search progresses, the selected node is gradually moved from Delete and Stored in the closed list C.

[0104] In the differentiable A After the module completes the path search, the path quality still needs to be improved because some detours may exist. Therefore, the present invention designs a shortcut path exploration to identify potential shorter edges. Once these shortcuts are identified, they are added to the exploration tree to eliminate unnecessary detours.

[0105] The search window length is preset to 2. Assume that the current window contains nodes v i-1 , v i , vi+1 , judge node v i-1 and v i+1 Is there a shortcut between v i-1 and v i+1 If the edges between them do not intersect with obstacles, there are shortcuts within the search window. Once these shortcuts are identified, they are added to the exploration tree to eliminate unnecessary detours. By moving the search window, iteratively checking whether there are shortcuts within the window to improve the path quality.

[0106] In one feasible implementation, the present invention uses five key indicators to evaluate performance, namely, the success rate of collision-free pathfinding (SR), the number of collision checks (CK), the planning time (PT), the path cost (CO), and the penalized path cost (PO).

[0107] The performance comparison of the model proposed in this invention and the most advanced method on the above 6 datasets is shown in Table 1 (Planning performance comparison table on maze planning dataset) and Table 2 (Planning performance comparison on robotic arm operation dataset).

[0108] Table 1

[0109]

[0110] Table 2

[0111]

[0112] The present invention has demonstrated remarkable performance in all tested simulation environments, achieving a success rate close to 100%. In particular, in high-dimensional environments, the present invention significantly reduces the number of collision checks and path costs. For some methods with lower success rates, they can only successfully find paths in simpler example problems with lower path costs. For this reason, the present invention introduces a penalty term in failed paths to balance the fairness of path cost comparison. For this metric, the present invention achieves the best results in all simulation environments. Experimental results show that the present invention is robust and stable for decision-making tasks in various complex environments.

[0113] The present invention proposes a robot motion planning method inspired by spatial relational memory. In the human brain's decision-making and relational reasoning processes, the brain mechanisms related to spatial relational memory show obvious advantages. Based on this, a new brain-like motion planning based on graph neural networks is designed to improve the performance of robot planning in high-dimensional continuous environments. When constructing a random geometric graph, a new random geometric graph generation method is used to make the sampled nodes closer to the optimal solution, thereby improving the search efficiency; through the intelligent edge selector, the edges with the highest confidence in the random geometric graph are retained, avoiding the negative impact of unreasonable edges on motion planning; through the memory perception predictor, the brain mechanism of integrating sensory observation information with relational structure for sensory reasoning is imitated in the brain's relational memory, and the introduction of subgraph structure into node features improves the graph representation ability of the graph neural network and enhances the overall learning effect of the model. The present invention is an efficient and accurate robot motion planning method inspired by the spatial relational memory of the human brain.

[0114] Figure 4 This is a block diagram of a robot motion planning device inspired by spatial relationship memory according to an exemplary embodiment. The device is used in a robot motion planning method inspired by spatial relationship memory. Figure 4 The device includes an information acquisition module 410, a first random geometric graph construction module 420, a second random geometric graph construction module 430, a heuristic value calculation module 440, and an optimal path acquisition module 450.

[0115] Information acquisition module 410, used to obtain movement starting position information, obstacle position information and map information;

[0116] A first random geometric graph construction module 420 is configured to construct a first random geometric graph in the robot's configuration space using a K-NN algorithm based on a selective sampling strategy and according to the motion start position information, the motion target position information, and the map information;

[0117] A second random geometric graph construction module 430 is configured to input the first random geometric graph and obstacle location information into a smart edge selector for edge screening based on a preset confidence threshold to obtain a second random geometric graph;

[0118] a heuristic value calculation module 440 for performing heuristic value calculation based on the second random geometric graph using a memory-aware predictor to obtain a node heuristic value;

[0119] The optimal path acquisition module 450 is used to obtain the optimal path based on the differentiable The module uses a shortcut path exploration method to search for the optimal path based on the graph node heuristic values ​​and the second random geometric graph to obtain the second optimal path.

[0120] Optionally, the first random geometric graph construction module 420 is further configured to:

[0121] Map the robot's position based on the motion starting position information, motion target position information, and map information to obtain the robot's positioning information;

[0122] Based on the robot's configuration space, a node set is constructed using a selective sampling strategy according to the movement starting position information, map information and robot positioning information;

[0123] According to the node set, use the K-NN algorithm to search for neighboring nodes and obtain the neighboring node set;

[0124] Edge connections are performed based on the node set and the neighboring node set to obtain a first random geometric graph.

[0125] Among them, the smart edge selector includes the encoding layer, transformer layer and hidden layer;

[0126] The encoding layer is built based on the attention mechanism;

[0127] The hidden layer is constructed according to the multi-layer perceptron model structure.

[0128] Optionally, the second random geometric graph construction module 430 is further configured to:

[0129] Obtaining node information and edge information according to the first random geometric graph;

[0130] Input the node information, edge information and obstacle position information into the encoding layer for feature extraction to obtain the first node feature, the first edge feature and the obstacle position feature;

[0131] According to the first node feature, the first edge feature and the obstacle position feature, encoding processing is performed through the transformer layer to obtain the node-obstacle feature and the edge-obstacle feature;

[0132] Input the node-obstacle features and edge-obstacle features into the hidden layer for confidence calculation to obtain the edge confidence;

[0133] The edges of the first random geometric graph are screened according to the edge confidence and a preset confidence threshold to obtain a second random geometric graph.

[0134] The memory-aware predictor consists of a subgraph extraction layer, a graph encoding layer, a linear layer, a pooling layer, and an output layer.

[0135] Optionally, the heuristic value calculation module 440 is further configured to:

[0136] Input the subgraph extraction layer according to the second random geometric graph to perform subgraph extraction, and obtain the Ego subgraph and the Cut subgraph;

[0137] Based on the preset random walk process, the probability distribution is calculated according to the Ego subgraph and the Cut subgraph to obtain the random walk probability of the Ego subgraph and the random walk probability of the Cut subgraph;

[0138] Input the second random geometric graph into the graph encoding layer for feature extraction to obtain second node features and second edge features;

[0139] Input the random walk probability of the Ego subgraph and the random walk probability of the Cut subgraph into the graph encoding layer for encoding processing to obtain the first hidden representation vector feature and the second hidden representation vector feature;

[0140] Splicing the first hidden representation vector feature and the second hidden representation vector feature to the vertex vector through a linear layer to obtain a first spliced ​​vertex vector;

[0141] Input the first spliced ​​vertex vector into the pooling layer for aggregation iteration to obtain the second spliced ​​vertex vector;

[0142] The second spliced ​​vertex vector and the second edge feature vector are input into the output layer to calculate the heuristic value, so as to obtain the node heuristic value of the second random geometric graph.

[0143] Optionally, the optimal path acquisition module 450 is further configured to

[0144] Input the graph node heuristic value into the differentiable A module, which performs an optimal path search in the second random geometric graph to obtain a first optimal path;

[0145] According to the first optimal path, redundant nodes are deleted through a shortcut path exploration method to obtain a second optimal path.

[0146] The present invention proposes a robot motion planning method inspired by spatial relational memory. In the human brain's decision-making and relational reasoning processes, the brain mechanisms related to spatial relational memory show obvious advantages. Based on this, a new brain-like motion planning based on graph neural networks is designed to improve the performance of robot planning in high-dimensional continuous environments. When constructing a random geometric graph, a new random geometric graph generation method is used to make the sampled nodes closer to the optimal solution, thereby improving the search efficiency; through the intelligent edge selector, the edges with the highest confidence in the random geometric graph are retained, avoiding the negative impact of unreasonable edges on motion planning; through the memory perception predictor, the brain mechanism of integrating sensory observation information with relational structure for sensory reasoning is imitated in the brain's relational memory, and the introduction of subgraph structure into node features improves the graph representation ability of the graph neural network and enhances the overall learning effect of the model. The present invention is an efficient and accurate robot motion planning method inspired by the spatial relational memory of the human brain.

[0147] Figure 5 FIG. 1 is a schematic diagram of a robot motion planning device provided by an embodiment of the present invention. Figure 5 As shown, the robot motion planning device may include the above Figure 4 Optionally, the robot motion planning device 510 may include a first processor 2001 .

[0148] Optionally, the robot motion planning device 510 may further include a memory 2002 and a transceiver 2003 .

[0149] The first processor 2001, the memory 2002 and the transceiver 2003 may be connected via a communication bus.

[0150] The following combination Figure 5 The components of the robot motion planning device 510 are described in detail:

[0151] The first processor 2001 is the control center of the robot motion planning device 510 and 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), or 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).

[0152] Optionally, the first processor 2001 may execute various functions of the robot motion planning device 510 by running or executing a software program stored in the memory 2002 and calling data stored in the memory 2002 .

[0153] In a specific implementation, as an embodiment, the first processor 2001 may include one or more CPUs, such as Figure 5 CPU0 and CPU1 are shown in FIG.

[0154] In a specific implementation, as an embodiment, the robot motion planning device 510 may also include multiple processors, such as Figure 5 1 and 2. The first processor 2001 and the second processor 2004 are shown in FIG. Each of these processors can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). A processor herein can refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0155] The memory 2002 is used to store the software program for executing the solution of the present invention, and is controlled by the first processor 2001 for execution. The specific implementation method can refer to the above method embodiment and will not be repeated here.

[0156] Alternatively, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, a random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, an optical disc storage (including a compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and capable of being accessed by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or exist independently and accessed through the interface circuit ( Figure 5 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0157] The transceiver 2003 is used to communicate with a network device or a terminal device.

[0158] Optionally, the transceiver 2003 may include a receiver and a transmitter ( Figure 5 The receiver is used to implement a receiving function, and the transmitter is used to implement a sending function.

[0159] Optionally, the transceiver 2003 may be integrated with the first processor 2001, or may exist independently and communicate with the first processor 2001 through the interface circuit ( Figure 5 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0160] It should be noted that Figure 5 The structure of the robot motion planning device 510 shown in the figure does not constitute a limitation on the router. The actual knowledge structure recognition device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0161] In addition, the technical effects of the robot motion planning device 510 can refer to the technical effects of the robot motion planning method inspired by spatial relationship memory described in the above method embodiment, and will not be repeated here.

[0162] It should be understood that the first processor 2001 in the embodiment of the present invention may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor, or the processor may be any conventional processor, etc.

[0163] It should also be understood that the memory in the embodiments of the present invention may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may 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 may be random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0164] The above embodiments can be implemented in whole or in part via software, hardware (e.g., circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are fully or partially performed. 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 means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0165] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0166] In this disclosure, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, "at least one of a, b, or c" can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0167] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean 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.

[0168] Those skilled in the art will appreciate that the units and algorithm steps of each example 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 performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0169] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0170] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interface, indirect coupling or communication connection of the device or unit, which can be electrical, mechanical or other forms.

[0171] The units described as separate components may or may not be physically separate, and 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 these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0172] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0173] If the 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 the present invention, or the portion that contributes to the prior art, or the portion 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 for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical disks.

[0174] 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 modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A robot motion planning method inspired by spatial relation memory, characterized in that The method comprises: Obtain movement starting position information, movement target position information, obstacle position information and map information; Based on the selective sampling strategy, according to the movement starting position information, the movement target position information and the map information, a first random geometric graph is constructed in the robot's configuration space by using a K-NN algorithm; Based on a preset confidence threshold, inputting the first random geometric graph and the obstacle position information into a smart edge selector for edge screening to obtain a second random geometric graph; Wherein, the intelligent edge selector includes an encoding layer, a transformer layer and a hidden layer; The encoding layer is constructed based on the attention mechanism; The hidden layer is constructed according to the multi-layer perceptron model structure; The step of inputting the first random geometric graph and the obstacle position information into the smart edge selector for edge screening based on a preset confidence threshold to obtain a second random geometric graph includes: Obtaining node information and edge information according to the first random geometric graph; Inputting the node information, the edge information, and the obstacle position information into the coding layer for feature extraction to obtain a first node feature, a first edge feature, and an obstacle position feature; Performing encoding processing through the transformer layer according to the first node feature, the first edge feature, and the obstacle position feature to obtain a node-obstacle feature and an edge-obstacle feature; Inputting the node-obstacle features and the edge-obstacle features into the hidden layer to perform confidence calculation to obtain edge confidence; Filtering the edges of the first random geometric graph according to the edge confidence and a preset confidence threshold to obtain a second random geometric graph; According to the second random geometric graph, heuristic values ​​are calculated by a memory-aware predictor to obtain node heuristic values; The memory-aware predictor comprises a subgraph extraction layer, a graph encoding layer, a linear layer, a pooling layer, and an output layer; The step of calculating the heuristic value by the memory-aware predictor according to the second random geometric graph to obtain the node heuristic value includes: Inputting the second random geometric graph into the subgraph extraction layer to perform subgraph extraction to obtain an Ego subgraph and a Cut subgraph; Based on a preset random walk process, a probability distribution calculation is performed according to the Ego subgraph and the Cut subgraph to obtain the random walk probability of the Ego subgraph and the random walk probability of the Cut subgraph; Inputting the second random geometric graph into the graph encoding layer for feature extraction to obtain second node features and second edge features; Inputting the random walk probability of the Ego subgraph and the random walk probability of the Cut subgraph into the graph coding layer for coding processing to obtain a first hidden representation vector feature and a second hidden representation vector feature; splicing the first hidden representation vector feature and the second hidden representation vector feature to the vertex vector of the second node feature through the linear layer to obtain a first spliced ​​vertex vector; Inputting the first spliced ​​vertex vector into the pooling layer for aggregation iteration to obtain a second spliced ​​vertex vector; Inputting the second spliced ​​vertex vector and the second edge feature into the output layer to perform heuristic value calculation to obtain a node heuristic value of the second random geometric graph; Based on the differentiable A* module, according to the node heuristic value and the second random geometric graph, a shortcut path exploration method is used to perform an optimal path search to obtain a second optimal path.

2. The robot motion planning method inspired by spatial relationship memory according to claim 1, characterized in that: The method comprises constructing a first random geometric graph in the robot's configuration space by using a K-NN algorithm based on the selective sampling strategy and the motion starting position information, the motion target position information, and the map information, including: Perform robot position mapping based on the movement starting position information, movement target position information, and map information to obtain robot positioning information; Based on the robot's configuration space, a node set is constructed using a selective sampling strategy according to the movement starting position information, the map information, and the robot positioning information; According to the node set, a K-NN algorithm is used to search for neighboring nodes to obtain a neighboring node set; Perform edge connections based on the node set and the neighboring node set to obtain a first random geometric graph.

3. The robot motion planning method inspired by spatial relation memory according to claim 1, characterized in that: The method of performing an optimal path search using a shortcut path exploration method based on the differentiable A* module according to the node heuristic value and the second random geometric graph to obtain a second optimal path includes: Inputting the node heuristic value into a differentiable A* module, performing an optimal path search in the second random geometric graph, and obtaining a first optimal path; According to the first optimal path, redundant nodes are deleted using a shortcut path exploration method to obtain a second optimal path.

4. A robot motion planning device inspired by spatial relational memory, wherein the robot motion planning device inspired by spatial relational memory is used to implement the robot motion planning method inspired by spatial relational memory as described in any one of claims 1 to 3, characterized in that: The device comprises: An information acquisition module is used to obtain movement starting position information, movement target position information, obstacle position information and map information; A first random geometric graph construction module is configured to construct a first random geometric graph in the robot's configuration space using a K-NN algorithm based on a selective sampling strategy and according to the motion starting position information, the motion target position information, and the map information; a second random geometric graph construction module, configured to input the first random geometric graph and the obstacle position information into the smart edge selector for edge screening based on a preset confidence threshold, thereby obtaining a second random geometric graph; a heuristic value calculation module, configured to calculate a heuristic value based on the second random geometric graph by using the memory-aware predictor to obtain a node heuristic value; The optimal path acquisition module is used to search for an optimal path based on the differentiable A* module, according to the node heuristic value and the second random geometric graph, using a shortcut path exploration method to obtain a second optimal path.

5. A robot motion planning device, characterized in that: The robot motion planning device comprises: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 3 is implemented.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program code, which can be called by a processor to execute the method according to any one of claims 1 to 3.

Images