A local path planning method and system based on spatial memory neural mechanism

Abstract

This application discloses a local path planning method and system based on spatial memory neural mechanisms, aiming to improve the robot's obstacle avoidance and shortcut discovery capabilities in complex environments. Inspired by the hippocampal navigation neural circuit, the algorithm achieves efficient memory search and path planning by simulating the functions of grid cells, position cells, and boundary cells. First, the robot explores feasible paths and records memory nodes, including grid cell representations, reward values, and cognitive value. Second, a hierarchical memory module is constructed, achieving information abstraction from fine to sparse from the dorsal to the ventral side, and information inheritance and memory search are achieved through fixed-interval connections. Finally, path planning is performed using direction decoding and boundary cell neural representations, dynamically selecting the optimal sub-target point. This algorithm requires no extensive training and can be quickly deployed to various robots to achieve intelligent autonomous navigation.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and path planning technology, and in particular to a local path planning method and system based on spatial memory neural mechanisms. Background Technology

[0002] In recent years, significant progress has been made in the study of biological navigation mechanisms in the field of neurobiology. Cognitive mapping theory, as one of the core theories explaining biological navigation, has gained widespread acceptance among researchers. This theory posits that animals achieve spatial localization and path planning by forming internal representations of their environment through brain centers (especially the hippocampus and entorhinal cortex). In this process, specific nerve cells in the hippocampus and entorhinal cortex, such as place cells, grid cells, and boundary cells, play crucial roles. The prefrontal cortex and basal ganglia are involved in cognitive functions such as strategy planning, decision-making, and action selection. This complex neural circuit collaboration enables organisms to rapidly adapt and navigate in dynamic and complex environments.

[0003] Building upon this foundation, researchers have proposed neuromorphic navigation algorithms to enhance robots' autonomous navigation capabilities in complex environments by simulating this biological navigation mechanism. The advantage of neuromorphic navigation algorithms lies in their ability to utilize bio-inspired mechanisms for more flexible and intelligent path planning and decision-making in unknown or dynamic environments. Therefore, developing neuromorphic navigation algorithms based on neural mechanisms can significantly improve robot performance in obstacle avoidance, path optimization, and task execution. However, despite the significant importance of this research direction in advancing the practical application of neuromorphic navigation technology for robots, it still faces numerous challenges. Summary of the Invention

[0004] Based on this, embodiments of this application provide a local path planning method and system based on spatial memory neural mechanisms, which, combined with an efficient memory search strategy, enhances obstacle avoidance and shortcut discovery capabilities in complex environments, thereby achieving a more intelligent and autonomous navigation effect.

[0005] Firstly, a local path planning method based on spatial memory neural mechanisms is provided, which includes:

[0006] Step 1: The robot is manually manipulated to explore a feasible path from the current position to the target point in the environment, and the surrounding environment is perceived by the lidar on the robot, which serves as the source of memory information for the robot to record the information of the path points.

[0007] Step 2: After obtaining the path point information, the recorded path point information is processed according to the hierarchical tissue memory network to achieve dynamic path planning; wherein, the hierarchical tissue memory network includes grid cells, position cells and boundary cells.

[0008] Step 3: Initialize parameters, calculate motion information from the current position to the target position, and obtain the target direction and the arrival indication signal value; when the signal threshold is reached, it is considered that the robot has reached the vicinity of the target and the planning ends; otherwise, select the motion direction.

[0009] Optionally, the robot can use its onboard LiDAR to perceive the surrounding environment and record waypoint information, including:

[0010] The information of each waypoint is represented as a tuple, which contains the grid cell representation pattern, reward value, and cognitive value. The grid cell representation pattern is the spatial coordinates of the waypoint, the reward value is generated based on the reverse memory playback of the target point, and the cognitive value represents the level of the memory node. The higher the level, the greater the cognitive value.

[0011] Optionally, the hierarchical memory network includes grid cells, place cells, and boundary cells, specifically including:

[0012] Generate a hierarchical tissue memory network based on the structure of the hippocampal navigation neural circuit;

[0013] The grid cells are distributed in the medial entorhinal cortex to provide a spatial coordinate system and perform vector navigation functions through the motion output module;

[0014] Location cells are distributed in the CA region of the hippocampus and are used to connect synapses to support the storage and retrieval of memories, as well as to support the hierarchical organization and efficient search of memories at different spatial sparsities.

[0015] Boundary cells are located in the lateral entorhinal cortex and are used to sense local obstacles and indicate whether passage is possible.

[0016] Optionally, in step 2, after obtaining the waypoint information, the recorded waypoint information is processed according to the hierarchical memory network to achieve dynamic path planning, including:

[0017] From the dorsal to the ventral side, at fixed intervals o l The upper layer inherits the memory information of the lower layer's exploration nodes, completing a refined to sparse abstract representation; among them, the tuple of the jl-th memory node in the l-th layer memory pool is:

[0018]

[0019] From the ventral to the dorsal side, at fixed intervals o l Connecting exploration nodes from upper to lower levels completes the memory search from abstraction to concreteness. The connection weight from level l to level (l-1) is:

[0020]

[0021] Optionally, in step 3, initialize parameters, calculate motion information from the current position to the target position, and obtain the target direction and arrival indication signal value, including:

[0022] Step 3.1: Calculate the motion information from the current position to the target position using the direction decoding module to obtain the target direction and the arrival indication signal value. If the signal threshold is reached, it is considered that the robot has reached the vicinity of the target, and the planning ends; otherwise, select a motion direction; otherwise, proceed to step 3.1.

[0023] Step 3.2, select the direction of movement. First, determine whether the target direction is passable:

[0024] s=B*T(θ target )

[0025] If s = 1, then the target direction θ is considered to be... target Passage is permitted, a = θ target This strategy is called a vector navigation strategy; conversely, if s = 0, then the target direction θ is considered to be... target It is not possible to proceed, a = θ, the strategy is called integrated navigation strategy;

[0026] Step 3.3: Update the current position p according to the direction in Step 3.2, and then proceed to Step 3.1.

[0027] Step 3.4: Calculate the direction of motion using the integrated navigation strategy.

[0028] Optionally, step 3.4: Calculate the direction of motion using an integrated navigation strategy, specifically including:

[0029] Step 3.4.1: Calculate the orientation of the memory nodes. For the l-th level memory pool, calculate the orientation of the memory nodes participating in the search, based on the search set I. l Provides instructions. When l = 1, I l = [1,2,...,n] E Otherwise, it is provided by step 3.4.3. The direction of the jl-th node is obtained by the direction decoding module:

[0030]

[0031] Step 3.4.2: Modulation of reward excitation and boundary inhibition to obtain the discharge state of the l-th layer memory node.

[0032]

[0033] This leads to the memory node search vector.

[0034]

[0035] Step 3.4.3: If l = 1, then output the optimal direction. Otherwise, perform a low-level search, searching set I. l-1 for:

[0036]

[0037] Secondly, a local path planning system based on spatial memory neural mechanisms is provided, the system comprising:

[0038] The recording unit allows the robot to explore a feasible path from its current location to a target point within the environment, and to record the path information using the robot's onboard LiDAR to perceive the surrounding environment and serve as the robot's memory information source.

[0039] The memory network unit is used to process the recorded path point information according to the hierarchical organization memory network after acquiring the path point information to achieve dynamic path planning; wherein, the hierarchical organization memory network includes grid cells, position cells and boundary cells.

[0040] The processing unit is used to initialize parameters, calculate motion information from the current position to the target position, and obtain the target direction and the arrival indication signal value. When the signal threshold is reached, it is considered that the robot has reached the vicinity of the target and the planning ends; otherwise, the motion direction is selected.

[0041] Thirdly, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the local path planning method described in any of the first aspects above.

[0042] Fourthly, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the local path planning method described in any of the first aspects above.

[0043] Fifthly, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the local path planning method described in any of the first aspects above.

[0044] The beneficial effects of the technical solutions provided in this application include at least the following:

[0045] (1) Flexible obstacle avoidance and shortcut discovery capabilities: Based on the navigation neural computing mechanism of mammals, this invention integrates external obstacle perception information with internal contextual memory to endow the robot with more flexible obstacle avoidance capabilities and the ability to discover unexplored shortcuts. Compared with the prior art, this invention adopts a strategy of multiple neuron models (such as boundary cells, position cells, and grid cells) working together to support the dynamic switching between support vector navigation and integrated navigation, thereby achieving more intelligent path planning decisions.

[0046] (2) Multi-scenario path planning with reduced mapping computation costs: Compared with traditional path planning techniques, this invention only requires manual manipulation of a trajectory to the target to provide global path guidance, without the need to fully explore and construct a detailed map of the scene in advance. This advantage is mainly due to the fact that this invention combines a highly efficient memory search strategy with global path and local obstacle avoidance, avoiding complete modeling of the entire environment and greatly saving the computational cost of mapping. Therefore, this invention can achieve an efficient path planning method in multiple scenarios and multiple tasks.

[0047] (3) Hierarchical memory structure for improved computational efficiency and interpretability: Compared with existing brain-like navigation technologies, this invention is not limited to the role of single neuron representation patterns in spatial navigation, but focuses more on the computational function of the hippocampal circuit. By introducing a multi-level brain-like memory structure (such as ventral sparse memory and dorsal fine memory), the computational efficiency and interpretability of the path planning algorithm are improved. This hierarchical structure allows the algorithm to avoid traversing all memory information during path search, instead adopting a "skip list"-like search method, reducing the computational complexity from linear to logarithmic level while maintaining optimal path length. Attached Figure Description

[0048] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0049] Figure 1 The flowchart shows a local path planning method based on spatial memory neural mechanisms.

[0050] Figure 2 Map showing hippocampal neural circuits and spatial cell distribution;

[0051] Figure 3 This is a schematic diagram of the hippocampus model control flow.

[0052] Figure 4 A schematic diagram of information inheritance and weighted connection in the ventral-dorsal axis of the hippocampus;

[0053] Figure 5 A schematic diagram for integrating navigation strategy calculations;

[0054] Figure 6 A flowchart outlining the strategy organization for the path planning algorithm;

[0055] Figure 7 Here are schematic diagrams of the simulation environment (A. Obstacle environment; B. Hairpin maze);

[0056] Figure 8 The path planning results in an obstacle environment are shown in the following diagrams: (A. Hierarchical memory pool visualization; B. Connection weight visualization; C. Strategy switching results on two paths: light blue represents vector navigation strategy, and purple represents integrated navigation strategy; D. Path planning results for 50 different starting points).

[0057] Figure 9 The path planning results in the hairpin maze are shown in the image below (A. Hierarchical memory pool visualization; B. Strategy switching results of the original maze; C. Strategy switching results of the three new mazes; light blue represents vector navigation strategy, and purple represents integrated navigation strategy).

[0058] Figure 10 Statistical charts of path planning results under single-layer and multi-layer memory structures (A. Path length statistics; B. Calculation time statistics). Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0060] In the description of this invention, the terms “comprising,” “having,” and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units that are expressly listed, but may also include other steps or units that are not expressly listed but are inherent to these processes, methods, products, or apparatuses, or steps or units added based on further optimizations of the inventive concept.

[0061] The purpose of this invention is to provide a local path planning algorithm based on the working mechanism of the biological brain, combined with an efficient memory search strategy, to improve obstacle avoidance and shortcut discovery capabilities in complex environments, and achieve a more intelligent and autonomous navigation effect.

[0062] Please refer to Figure 1The diagram illustrates a flowchart of a local path planning method based on spatial memory neural mechanisms provided in an embodiment of this application. This method may include the following steps:

[0063] Step 1: The robot is manually manipulated to explore a feasible path from its current location to the target point in the environment. The robot's LiDAR is used to perceive the surrounding environment and record the path information as a source of memory information for the robot.

[0064] Each path point is represented as a tuple containing a grid cell representation pattern, a reward value, and a cognitive value. The grid cell representation pattern is the spatial coordinates of the path point, the reward value is generated based on the reverse memory playback of the target point, and the cognitive value represents the level of the memory node. The higher the level, the greater the cognitive value.

[0065] Step 2: After obtaining the path point information, the recorded path point information is processed according to the hierarchical tissue memory network to achieve dynamic path planning; wherein, the hierarchical tissue memory network includes grid cells, position cells and boundary cells.

[0066] Step 3: Initialize parameters, calculate motion information from the current position to the target position, and obtain the target direction and the arrival indication signal value; when the signal threshold is reached, it is considered that the robot has reached the vicinity of the target and the planning ends; otherwise, select the motion direction.

[0067] This invention presents a local path planning algorithm based on the working mechanism of the biological brain. By combining an efficient memory search strategy, it enhances the robot's obstacle avoidance and shortcut discovery capabilities in complex environments, achieving more intelligent and autonomous navigation. The entire algorithm is inspired by the structure and function of the hippocampal navigation neural circuit, and its specific structure is as follows: Figure 2 As shown. The core of the algorithm consists of the following three key neuron types and their corresponding functional modules:

[0068] Grid cells: Located in the medial entorhinal cortex (MEC), they are responsible for providing a spatial coordinate system and performing vector navigation functions through the motion output module. Grid cells further obtain the direction of motion and the signal strength when approaching the target by decoding the motion output pattern.

[0069] Place cells: Located in the CA regions of the hippocampus, particularly CA1 and CA3, and perform different functions in these two regions. The recurrent synapses in the CA3 region support memory storage and retrieval, while the positional signal variations along the ventral-dorsal axis in the CA1 region support the hierarchical organization and efficient retrieval of memories at varying spatial sparsity.

[0070] Boundary cells: distributed in the lateral entorhinal cortex (LEC), mainly used to perceive local obstacle information and to indicate whether passage is possible.

[0071] Boundary cells, position cells, and grid cells collaborate to dynamically select the optimal sub-target point during real-time navigation, supporting the flexible selection and integration of multiple strategies to achieve dynamic path planning and obstacle avoidance. For simple and efficient deployment, the internal weights and parameters of each functional module are pre-configured, eliminating the need for extensive training. Therefore, the algorithm can be easily migrated and deployed to various robots and automated devices, enabling instant navigation. Furthermore, other high-level control logic implemented in the prefrontal cortex or basal ganglia is represented through explicit rules to ensure the system's robustness and adaptability.

[0072] Figure 3 A control diagram inspired by the hippocampus loop is shown, and the specific steps are as follows:

[0073] Step 1: Manually guide the robot to explore a feasible path from the current location to the target location. This serves as the source of the robot's memory information. The information in each memory node is represented using tuples:

[0074] Ω j =(g j r j v j ), j = 1, 2, ..., n E .

[0075] Among them, g j For path point p j The grid cell characterization pattern, r j and v j This corresponds to the reward value and cognitive value of that point. The reward value is generated by the reverse memory playback after the target point is determined, i.e. Cognitive value is determined by the layer in which the memory node is located; simply put, the higher the layer, the greater the cognitive value.

[0076] Step 2: Organize hierarchical memory modules according to the ventral-dorsal direction of the hippocampus, such as... Figure 4 As shown. Specifically, in the direction from the dorsal side to the ventral side, at fixed intervals o l The upper layer inherits the memory information of the lower layer's exploration nodes, completing a refined to sparse abstract representation. Therefore, the j-th node in the l-th layer memory pool... l The tuple for each memory node is:

[0077]

[0078] From the ventral to the dorsal side, at fixed intervals o l Connecting exploration nodes from upper to lower levels completes the memory search from abstraction to concreteness. Therefore, the connection weight from layer l to layer (l-1) is:

[0079]

[0080] Step 3: Enter path planning. First, initialize the following parameters: initial position (current position) p, initial and target position grid cell pattern g, exploration trajectory and multi-layer memory pool Ω obtained from steps 1-2, and arrival signal threshold U. thre The method in this application provides flexible selection of navigation strategies, and first introduces the orientation decoding module and the boundary cell neural representation method.

[0081] Direction decoding module: Performs vector decoding module D using the raster cell representations of the current and target positions.

[0082]

[0083] This module includes n D There are directional detectors, each with neural activity uk and a preferred direction θ. k Uniform sampling is performed within the range [0, 2π]. The target direction θ and the arrival indication signal value u are obtained by decoding the motion signal Θ:

[0084]

[0085] u=||Θ||2

[0086] Boundary cell neural activity: via an onboard lidar (circumferential scanning rays numbered i = 1, 2, ..., n) b By combining the given direction and a preset distance threshold, we can obtain binarized directional feasibility indication information (also known as boundary cell firing mode) B, i.e., when the detection direction... If the distance to an obstacle is higher than a threshold, the path is considered passable; if it is lower, the path is considered impassable. Using the angle fine-tuning method T(·), the angle θ to be considered is compared with the detection direction θ. i Alignment yields Φ:

[0087]

[0088] in, If and only if i = argmax(cos(θ-θ) i When activated, it is essentially still a binary discharge mode.

[0089] The detailed process of step 3 is as follows:

[0090] Step 3.1: Calculate the motion information from the current position to the target position using the direction decoding module to obtain the target direction θ. target And reach the indicated signal value u. If u thre If the robot is considered to have reached the vicinity of the target, then the planning stops; otherwise, proceed to step 3.2.

[0091] Step 3.2: Select direction of movement a. First, determine if the target direction is passable:

[0092] s=B*T(θ target )

[0093] If s = 1, then the target direction θ is considered to be... targe t can pass, a = θ target This strategy is called a vector navigation strategy; conversely, if s = 0, then the target direction θ is considered to be... target It is not allowed to pass. This strategy is called the integrated navigation strategy, see step 3.5;

[0094] Step 3.3: Update the current position p according to the direction in Step 3.2, and then proceed to Step 3.1.

[0095] Step 3.4: Calculate the direction of motion using the integrated navigation strategy. like Figure 5 As shown, the specific steps are as follows:

[0096] Step 3.4.1: Calculate the orientation of the memory nodes. For the l-th level memory pool, calculate the orientation of the memory nodes participating in the search, based on the search set I. l Provides instructions. When l = 1, I l = [1, 2, ..., n] E Otherwise, it is provided by step 3.4.3. The direction of the jl-th node is obtained by the direction decoding module:

[0097]

[0098] Step 3.4.2: Modulation of reward excitation and boundary inhibition to obtain the discharge state of the l-th layer memory node.

[0099]

[0100] Based on the "winner-takes-all" rule, the memory node search vector is obtained.

[0101]

[0102] Step 3.4.3: If l = 1, then output the optimal direction. ​Otherwise, perform a low-level search, searching set I. l-1 for:

[0103]

[0104] like Figure 6 A flowchart illustrating the technical implementation details is provided. To verify the usability and efficiency of the algorithm of this invention, this application... Figure 7 The tests were conducted in two different environments, and the generated red trajectories were based on the method described in step 1.

[0105] Example 1: Flexible obstacle avoidance in open environments with distributed obstacles

[0106] exist Figure 7 In the open environment shown, the exploration trajectory is first organized into a three-layer memory structure, such as... Figure 8 As shown in Figure A, the visualization results of the connectivity matrix are shown below. Figure 8 B. Next, the path planning process is executed. The algorithm of this invention can flexibly switch between a vector navigation strategy (sky blue trajectory) and an integrated navigation strategy (blue trajectory), such as... Figure 8 As shown in C. Figure 8 D shows 50 planned trajectories generated from different starting points, all of which successfully avoided obstacles and reached the vicinity of the target, demonstrating the algorithm's reliable obstacle avoidance capability.

[0107] Example 2: Finding shortcuts in complex mazes

[0108] exist Figure 7 In the complex maze environment shown in B, the exploration trajectory is first organized into a three-layer memory structure, such as... Figure 9 As shown in A. Path planning is then performed. Figure 9 B shows the trajectory planned on the original maze map. (Through...) Figure 9 As shown in C, for the three different shortcut-finding tasks, the algorithm successfully found the shortcuts, demonstrating its path optimization capabilities in complex environments.

[0109] Based on the results of Examples 1 and 2, the path planning algorithm of the present invention, by combining a memory search strategy, not only achieves flexible obstacle avoidance, but also discovers shortcuts in unexplored paths, demonstrating the efficiency and practicality of the algorithm.

[0110] Example 3: Efficient Search Based on Multi-Layer Memory Mechanism

[0111] In this embodiment, the multi-layered memory structure is removed, and only a single-layered memory module directly generated based on the exploration trajectory is used to retest Embodiments 1 and 2. Figure 10As shown, the differences in planned path length and computation time were compared. The results indicate that using a single-level memory module did not significantly change the path length, but increased the computation time by approximately 5-7 times. This further demonstrates the advantages of the multi-level memory search mechanism in improving computational efficiency. The test was conducted on a computer configured with Ubuntu 18.04. W-2195 CPU @ 2.30GHz.

[0112] Experimental results demonstrate that in Examples 1 and 2, boundary cells, by sensing the distance to external obstacles, indicate the accessibility of the target direction and serve as a direct indicator for switching between vector navigation and integrated navigation strategies. Grid cells support vector navigation, combining past memory and current sensory information to achieve integrated navigation. In Example 3, a hierarchical search approach, moving from ventral sparse memory to dorsal fine memory, not only avoids accessing all memory information but also significantly reduces computational complexity while ensuring no increase in the planned path length. Therefore, this invention can efficiently and effectively accomplish obstacle avoidance tasks in obstacle-rich environments and shortcut discovery tasks in maze environments.

[0113] This application also provides a local path planning system based on spatial memory neural mechanisms, comprising:

[0114] The recording unit allows the robot to explore a feasible path from its current location to a target point within the environment, and to record the path information using the robot's onboard LiDAR to perceive the surrounding environment and serve as the robot's memory information source.

[0115] The memory network unit is used to process the recorded path point information according to the hierarchical organization memory network after acquiring the path point information to achieve dynamic path planning; wherein, the hierarchical organization memory network includes grid cells, position cells and boundary cells.

[0116] The processing unit is used to initialize parameters, calculate motion information from the current position to the target position, and obtain the target direction and the arrival indication signal value. When the signal threshold is reached, it is considered that the robot has reached the vicinity of the target and the planning ends; otherwise, the motion direction is selected.

[0117] Specific limitations regarding the local path planning system based on spatial memory neural mechanisms can be found in the limitations of the local path planning method based on spatial memory neural mechanisms mentioned above, and will not be repeated here. Each module in the aforementioned local path planning system based on spatial memory neural mechanisms can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0118] In one embodiment, an electronic device, which may be a computer, is provided. The electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device is used for local path planning data based on spatial memory neural mechanisms. The network interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a local path planning method based on spatial memory neural mechanisms.

[0119] In one embodiment of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the local path planning method described above.

[0120] In one embodiment of this application, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the local path planning method described above.

[0121] The computer-readable storage medium and computer program product provided in this embodiment are similar in implementation principle and technical effect to the above method embodiments, and will not be repeated here.

[0122] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in M ​​forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0123] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0124] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A local path planning method based on spatial memory neural mechanisms, characterized in that, The method includes: Step 1: The robot is manually manipulated to explore a feasible path from the current position to the target point in the environment, and the surrounding environment is perceived by the lidar on the robot, which serves as the source of memory information for the robot to record the information of the path points. Step 2: After obtaining the path point information, the recorded path point information is processed according to the hierarchical tissue memory network to achieve dynamic path planning; wherein, the hierarchical tissue memory network includes grid cells, position cells and boundary cells. Step 3: Initialize parameters, calculate motion information from the current position to the target position, and obtain the target direction and the arrival indication signal value; when the signal threshold is reached, it is considered that the robot has reached the vicinity of the target and the planning ends; otherwise, select the motion direction; The hierarchical memory network of an organization includes grid cells, place cells, and boundary cells, specifically including: Generate a hierarchical tissue memory network based on the structure of the hippocampal navigation neural circuit; The grid cells are distributed in the medial entorhinal cortex to provide a spatial coordinate system and perform vector navigation functions through the motion output module; Location cells are distributed in the CA region of the hippocampus and are used to connect synapses to support the storage and retrieval of memories, as well as to support the hierarchical organization and efficient search of memories at different spatial sparsities. Boundary cells are located in the lateral entorhinal cortex and are used to sense local obstacles and indicate whether passage is possible.

2. The local path planning method according to claim 1, characterized in that, The robot uses its onboard LiDAR to perceive its surroundings and record path point information, including: The information of each waypoint is represented as a tuple, which contains the grid cell representation pattern, reward value, and cognitive value. The grid cell representation pattern is the spatial coordinates of the waypoint, the reward value is generated based on the reverse memory playback of the target point, and the cognitive value represents the level of the memory node. The higher the level, the greater the cognitive value.

3. The local path planning method according to claim 1, characterized in that, Step 2, after obtaining the waypoint information, processes the recorded waypoint information according to the hierarchical memory network to achieve dynamic path planning, including: From the dorsal to the ventral side, at fixed intervals o l The upper layer inherits the memory information of the lower layer's exploration nodes, completing a refined to sparse abstract representation; among them, the j-th node in the l-th layer memory pool... l The tuple for each memory node is: From the ventral to the dorsal side, at fixed intervals o l Connecting exploration nodes from upper to lower levels completes the memory search from abstraction to concreteness. The connection weight from level l to level (l-1) is:

4. The local path planning method according to claim 1, characterized in that, Step 3: Initialize parameters, calculate motion information from the current position to the target position, and obtain the target direction and arrival indication signal value, including: Step 3.1: Calculate the motion information from the current position to the target position using the direction decoding module to obtain the target direction and the arrival indication signal value. If the signal threshold is reached, it is considered that the robot has reached the vicinity of the target, and the planning ends; otherwise, select a motion direction; otherwise, proceed to step 3.

1. Step 3.2, select the direction of movement. First, determine whether the target direction is passable: s=B*T(θ target ) If s=1, then the target direction θ is considered to be... target Passage is permitted, a=θ target This is called a vector navigation strategy; conversely, if s=0, then the target direction is assumed to be θ. target It is not allowed to pass. This is called an integrated navigation strategy; Step 3.3: Update the current position p according to the direction in Step 3.2, and then proceed to Step 3.1; Step 3.4: Calculate the direction of motion using the integrated navigation strategy.

5. The local path planning method according to claim 4, characterized in that, Step 3.4: Calculate the direction of motion using an integrated navigation strategy, specifically including: Step 3.4.1: Calculate the orientation of the memory nodes. For the l-th level memory pool, calculate the orientation of the memory nodes participating in the search, based on the search set I. l Provides an indication that when l=1, I l = [1, 2, ..., n] E Otherwise, as provided in step 3.4.3, the j-th... l The orientation of each node is obtained by the orientation decoding module: Step 3.4.2: Modulation of reward excitation and boundary inhibition to obtain the discharge state of the l-th layer memory node. This leads to the memory node search vector. Step 3.4.3: If l = 1, then output the optimal direction. Otherwise, perform a low-level search, searching set I. l-1 for:

6. A local path planning system based on spatial memory neural mechanisms, characterized in that, The system includes: The recording unit allows the robot to explore a feasible path from its current location to a target point within the environment, and to record the path information using the robot's onboard LiDAR to perceive the surrounding environment and serve as the robot's memory information source. The memory network unit is used to process the recorded path point information according to the hierarchical organization memory network after acquiring the path point information to achieve dynamic path planning; wherein, the hierarchical organization memory network includes grid cells, position cells and boundary cells. The processing unit is used to initialize parameters, calculate motion information from the current position to the target position, and obtain the target direction and the arrival indication signal value. When the signal threshold is reached, it is considered that the robot has reached the vicinity of the target and the planning ends; otherwise, the motion direction is selected. The hierarchical memory network of an organization includes grid cells, place cells, and boundary cells, specifically including: Generate a hierarchical tissue memory network based on the structure of the hippocampal navigation neural circuit; The grid cells are distributed in the medial entorhinal cortex to provide a spatial coordinate system and perform vector navigation functions through the motion output module; Location cells are distributed in the CA region of the hippocampus and are used to connect synapses to support the storage and retrieval of memories, as well as to support the hierarchical organization and efficient search of memories at different spatial sparsities. Boundary cells are located in the lateral entorhinal cortex and are used to sense local obstacles and indicate whether passage is possible.

7. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program that, when executed by the processor, implements the local path planning method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the local path planning method as described in any one of claims 1 to 5.

9. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the local path planning method as described in any one of claims 1 to 5.

Images