Unmanned System Navigation Method, Device, Unmanned System, Equipment and Storage Medium

Abstract

The present application provides an unmanned system navigation method, device, unmanned system, equipment and storage medium, relating to the technical field of unmanned systems. In the present application, an unmanned system navigation model based on a cognitive map is pre-trained. The neurons included in the cognitive map construction module in the unmanned system navigation model are differential equation neurons with adjustable time constants, which can accelerate the operation speed within the neurons. In addition, by constructing a cognitive map encoding through the cognitive map construction module, the map construction speed is shortened, so that the navigation speed and accuracy of the unmanned system can be significantly improved in large-scale real scenarios.

Description

Technical Field

[0001] The present application relates to the technical field of unmanned systems, and in particular to an unmanned system navigation method, apparatus, unmanned system, equipment and storage medium. Background Art

[0002] In recent years, autonomous navigation technology for unmanned systems has rapidly developed and has been widely used in fields such as intelligent transportation, robotics, and drones. Existing navigation methods primarily include path planning navigation based on high-precision environmental maps and end-to-end navigation using deep learning. The former relies on high-precision sensors (such as lidar and high-definition cameras) to perceive the environment and constructs a refined geometric map using technologies such as SLAM. Based on this map, global and local path planning is performed to achieve precise navigation. However, HD maps are expensive to construct and maintain, difficult to update in real time, and require high computing resources, making them difficult to respond quickly in dynamic environments. The latter uses convolutional neural networks to generate navigation control commands directly from camera image input, bypassing the traditional navigation process. While capable of some environmental adaptability, its decision-making process is "black-box" and lacks interpretability, making it difficult to effectively handle complex tasks and changing scenarios. It is particularly susceptible to adaptability issues when navigating long distances or in repetitive environments.

[0003] Therefore, in view of the limitations of existing navigation technology, it is urgent to propose a new autonomous navigation method. Summary of the Invention

[0004] In view of the above problems, the embodiments of the present application provide an unmanned system navigation method, apparatus, unmanned system, device and storage medium to overcome the above problems or at least partially solve the above problems.

[0005] In a first aspect of an embodiment of the present application, a method for navigating an unmanned system based on a cognitive map is provided, the method comprising:

[0006] capturing, by a camera carried by the unmanned system, a sequence of environmental images, the sequence of environmental images comprising a plurality of environmental images captured within a time period starting from an initial moment, the unmanned system being at an initial position at the initial moment;

[0007] Extracting features from the environment image sequence through the visual perception module of the unmanned system navigation model, and generating landmark codes, obstacle codes, motion parameter value codes, and location codes based on the extracted image features;

[0008] encoding the target position relative to the initial position in the target instruction by means of the instruction encoding module of the unmanned system navigation model to generate a target position code;

[0009] Through the cognitive map construction module of the unmanned system navigation model, the landmark encoding, the obstacle encoding, the motion parameter value encoding, the location encoding, and the target location encoding are processed to generate a cognitive map encoding, where the cognitive map encoding represents the environmental map from the initial location to the target location, and the neurons included in the cognitive map construction module are differential equation neurons with adjustable time constants;

[0010] Through the decision-making action module of the unmanned system navigation model, the cognitive map encoding is processed to generate control parameter values for controlling the drive motors of the unmanned system, where the control parameter values include: motion direction and motion speed.

[0011] Optionally, the process of generating the cognitive map encoding by processing the landmark encoding, the obstacle encoding, the motion parameter value encoding, the location encoding, and the target location encoding through the cognitive map construction module of the unmanned system navigation model includes:

[0012] Input the landmark encoding, the obstacle encoding, the motion parameter value encoding, the location encoding, and the target location encoding into the differential equation neurons in the cognitive map construction module to obtain the cognitive map neuron state;

[0013] Process the cognitive map neuron state through a first fully connected layer to obtain the cognitive map encoding.

[0014] Optionally, the neurons included in the visual perception module are differential equation neurons with adjustable time constants;

[0015] Through the visual perception module of the unmanned system navigation model, feature extraction is performed on the environmental image sequence, and based on the extracted image features, landmark encoding, obstacle encoding, motion parameter value encoding, and location encoding are generated, including:

[0016] Preprocess each of the environmental images in the environmental image sequence to obtain feature vectors;

[0017] Input the feature vectors into the differential equation neurons in the visual perception module to obtain the visual neuron state;

[0018] Process the visual neuron state through a second fully connected layer to obtain the landmark encoding, the obstacle encoding, the motion parameter value encoding, and the location encoding.

[0019] Optionally, the neurons included in the instruction encoding module and the decision-making action module are both differential equation neurons with adjustable time constants; where the output of the i-th differential equation neuron is the neuron state , the input of the i-th differential equation neuron includes: an input feature vector , and the states of other differential equation neurons connected to the i-th differential equation neuron;

[0020] Any of the differential equation neurons conforms to the following expression:

[0021]

[0022] where represents the state of the -th differential equation neuron; represents the state of the -th differential equation neuron; represents the input feature vector; represents the time constant that controls the dynamic change of the differential equation neuron; represents the set of units connected to the -th differential equation neuron; represents the -th differential equation neuron transmitting information to the -th differential equation neuron's connection weight; represents the perception weight from the input feature vector to the -th differential equation neuron; represents the activation function.

[0023] Optionally, the decision-making action module of the unmanned system navigation model processes the cognitive map encoding to generate control parameter values for controlling the drive motor of the unmanned system, including:

[0024] Processing the cognitive map encoding through the decision-making action module to obtain the action neuron state;

[0025] Determining the movement direction of the unmanned system during obstacle avoidance according to the positive or negative of the action neuron state, where the movement direction includes left and right directions;

[0026] Determining the movement speed of the unmanned system during obstacle avoidance according to the value of the action neuron state.

[0027] Optionally, the method further includes:

[0028] Collecting a sample environmental image sequence through a camera carried by the unmanned system, and obtaining the actual values of the control parameters of the drive motor of the unmanned system during the collection of the sample environmental image sequence;

[0029] Mark the initial sample position according to the position of the unmanned system when the first sample environment image in the collected sample environment image sequence is acquired, and mark the sample target position according to the position of the unmanned system when the last sample environment image in the collected sample environment image sequence is acquired;

[0030] Input the sample environment image sequence and the sample target position into the unmanned system navigation model to be trained, and obtain the predicted control parameter values of the drive motors of the unmanned system;

[0031] Update the perception weights and connection weights of each differential equation neuron in the unmanned system navigation model to be trained according to the predicted control parameter values of the drive motors of the unmanned system and the actual control parameter values of the drive motors of the unmanned system, and obtain the trained unmanned system navigation model.

[0032] In a second aspect of the present application, there is provided a navigation device for an unmanned system based on a cognitive map, the device comprising:

[0033] An acquisition unit, configured to acquire an environment image sequence through a camera carried by the unmanned system, the environment image sequence including multiple environment images acquired within a time period starting from an initial moment, and the unmanned system being located at an initial position at the initial moment;

[0034] A feature extraction unit, configured to extract features from the environment image sequence through a visual perception module of the unmanned system navigation model, and generate a landmark code, an obstacle code, a motion parameter value code, and a location code according to the extracted image features;

[0035] An encoding unit, configured to encode a target position relative to the initial position in a target instruction through an instruction encoding module of the unmanned system navigation model to generate a target position code;

[0036] A first processing unit, configured to process the landmark code, the obstacle code, the motion parameter value code, the location code, and the target position code through a cognitive map construction module of the unmanned system navigation model to generate a cognitive map code, the cognitive map code representing an environmental map from the initial position to the target position, and the neurons included in the cognitive map construction module being differential equation neurons with adjustable time constants;

[0037] A second processing unit, configured to process the cognitive map code through a decision-making action module of the unmanned system navigation model to generate control parameter values for controlling the drive motors of the unmanned system, the control parameter values including: a motion direction and a motion speed.

[0038] Optionally, the cognitive map construction module of the unmanned system navigation model processes the landmark encoding, the obstacle encoding, the motion parameter value encoding, the location encoding, and the target location encoding to generate a cognitive map encoding. The first processing unit includes:

[0039] A first input subunit, configured to input the landmark encoding, the obstacle encoding, the motion parameter value encoding, the location encoding, and the target location encoding into the differential equation neurons in the cognitive map construction module to obtain cognitive map neuron states;

[0040] A first processing subunit, configured to process the cognitive map neuron states through a first fully connected layer to obtain a cognitive map encoding.

[0041] Optionally, the neurons included in the visual perception module are differential equation neurons with adjustable time constants;

[0042] The visual perception module of the unmanned system navigation model extracts features from the environmental image sequence, and generates a landmark encoding, an obstacle encoding, a motion parameter value encoding, and a location encoding according to the extracted image features. The feature extraction unit includes:

[0043] A second processing subunit, configured to preprocess each of the environmental images in the environmental image sequence to obtain feature vectors;

[0044] A second input subunit, configured to input the feature vectors into the differential equation neurons in the visual perception module to obtain visual neuron states;

[0045] A third processing subunit, configured to process the visual neuron states through a second fully connected layer to obtain the landmark encoding, the obstacle encoding, the motion parameter value encoding, and the location encoding.

[0046] Optionally, the decision-making action module of the unmanned system navigation model processes the cognitive map encoding to generate a control parameter value for controlling the drive motor of the unmanned system. The second processing unit includes:

[0047] A fourth processing subunit, configured to process the cognitive map encoding through the decision-making action module to obtain action neuron states;

[0048] A first determination subunit, configured to determine the motion direction of the unmanned system during obstacle avoidance according to the positive or negative of the action neuron states, where the motion direction includes left and right directions;

[0049] A second determination subunit, configured to determine the movement speed of the unmanned system during obstacle avoidance according to the value of the action neuron state.

[0050] Optionally, the device further includes:

[0051] An acquisition subunit, configured to acquire a sample environment image sequence through a camera carried by the unmanned system, and obtain actual values of control parameters of a driving motor of the unmanned system during the acquisition of the sample environment image sequence;

[0052] A marking subunit, configured to mark an initial sample position according to the position of the unmanned system when the first sample environment image in the acquired sample environment image sequence is acquired, and mark a sample target position according to the position of the unmanned system when the last sample environment image in the acquired sample environment image sequence is acquired;

[0053] A third input subunit, configured to input the sample environment image sequence and the sample target position into an unmanned system navigation model to be trained, and obtain predicted values of control parameters of the driving motor of the unmanned system;

[0054] An update subunit, configured to update the perception weights and connection weights of each differential equation neuron in the unmanned system navigation model to be trained according to the predicted values of the control parameters of the driving motor of the unmanned system and the actual values of the control parameters of the driving motor of the unmanned system, so as to obtain the trained unmanned system navigation model.

[0055] In a third aspect of the present application, there is provided an unmanned system based on a cognitive map, where the system includes the unmanned system navigation device as described in the second aspect of the present application, and / or the unmanned system is configured to execute the steps of the unmanned system navigation method as described in the first aspect of the present application.

[0056] In a fourth aspect of the present application, there is provided an electronic device, including a processor, a memory, and a program or instruction stored on the memory and executable on the processor, where when the program or instruction is executed by the processor, the steps of the unmanned system navigation method as described in the first aspect of the present application are implemented.

[0057] In a fifth aspect of the present application, there is provided a readable storage medium, where a program or instruction is stored on the readable storage medium, and when the program or instruction is executed by a processor, the steps of the unmanned system navigation method as described in the first aspect of the present application are implemented.

[0058] Advantages of the present application:

[0059] The present application proposes a navigation method for an unmanned system based on a cognitive map. The method includes: First, an environmental image sequence is collected through a camera carried by the unmanned system; Then, through the visual perception module of the unmanned system navigation model, feature extraction is performed on the environmental image sequence, and landmark encoding, obstacle encoding, motion parameter value encoding, and location encoding where it is located are generated according to the extracted image features; Also, through the instruction encoding module of the unmanned system navigation model, the target position relative to the initial position in the target instruction is encoded to generate a target position encoding; After that, through the cognitive map construction module of the unmanned system navigation model, the landmark encoding, the obstacle encoding, the motion parameter value encoding, the location encoding where it is located, and the target position encoding are processed to generate a cognitive map encoding. The neurons included in the cognitive map construction module are differential equation neurons with adjustable time constants; Finally, through the decision-making action module of the unmanned system navigation model, the cognitive map encoding is processed to generate control parameter values for controlling the drive motor of the unmanned system. The control parameter values include: motion direction and motion speed.

[0060] The present application uses a pre-trained unmanned system navigation model based on a cognitive map. The neurons included in the cognitive map construction module of the unmanned system navigation model are differential equation neurons with adjustable time constants, which can accelerate the operation speed within the neurons. In addition, by constructing a cognitive map encoding through the cognitive map construction module, the map construction speed is shortened, thereby significantly improving the navigation speed and accuracy of the unmanned system in large-scale real scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments of the present application. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0062] Figure 1 It is a schematic flowchart of the steps of a navigation method for an unmanned system based on a cognitive map provided by an embodiment of the present application;

[0063] Figure 2 It is a schematic flowchart of the steps of a training method for an unmanned system navigation model provided by an embodiment of the present application;

[0064] Figure 3 It is a schematic diagram of the overall architecture of an unmanned system navigation model based on a cognitive map provided by an embodiment of the present application;

[0065] Figure 4It is a schematic diagram of a navigation device for an unmanned system based on a cognitive map provided by an embodiment of the present application;

[0066] Figure 5 It is a schematic diagram of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0067] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings in the embodiments of the present application. Although the exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present application can be more thoroughly understood and the scope of the present application can be completely conveyed to those skilled in the art.

[0068] In the first aspect of the embodiments of the present application, a navigation method for an unmanned system based on a cognitive map is provided, as Figure 1 shown, the method includes:

[0069] Step S101, collecting an environmental image sequence through a camera carried by the unmanned system, where the environmental image sequence includes multiple environmental images collected within a time period starting from an initial moment, and the unmanned system is located at an initial position at the initial moment.

[0070] In this step, an environmental image sequence is continuously collected through a camera carried by the unmanned system; during this process, the unmanned system is at a known position at the initial moment, and this position serves as a benchmark for state update and path integration to ensure the accuracy of positioning and timely correction of errors. In some cases, preprocessing such as denoising, enhancement, distortion correction, and color calibration can also be performed on the collected image sequence to ensure data quality.

[0071] In addition, in some embodiments, the image sequence contains rich scene geometric features, obstacle information, visual landmarks and other information, and the high frame rate and high resolution acquisition requirements and rapid data transmission ensure that the system can capture subtle environmental changes in a timely manner and meet the real-time response requirements, thus providing a key guarantee for the efficient and accurate operation of the entire end-to-end navigation system.

[0072] Step S102, extracting features from the environmental image sequence through the visual perception module of the unmanned system navigation model, and generating landmark coding, obstacle coding, motion parameter value coding, and position coding where it is located according to the extracted image features.

[0073] In this step, the visual perception module of the unmanned system navigation model extracts deep features from the collected environmental image sequence. In some cases, by comprehensively analyzing information such as texture, edges, shapes, and colors in the images, it can automatically identify key elements and feature regions in the environment and generate multi-level encodings based on this, including landmark encodings for identifying significant landmarks, obstacle encodings for quantitatively describing the positions and shapes of potential obstacles, motion parameter value encodings reflecting the current motion state, and location encodings providing precise spatial positioning. These encodings provide rich and accurate information support for subsequent cognitive map construction, path planning, and navigation decision-making.

[0074] Step S103: Through the instruction encoding module of the unmanned system navigation model, encode the target position in the target instruction relative to the initial position to generate a target position encoding.

[0075] In this step, the instruction encoding module of the unmanned system navigation model processes the target position information contained in the target instruction and converts this information into a target position encoding using a preset encoding algorithm, thereby being able to provide a clear and precise target indication; through this encoding process, the relative position relationship of the target in the initial coordinate system can be captured, realizing a quantitative description and highly robust expression from the initial position to the target position.

[0076] Step S104: Through the cognitive map construction module of the unmanned system navigation model, process the landmark encoding, the obstacle encoding, the motion parameter value encoding, the location encoding, and the target position encoding to generate a cognitive map encoding. The cognitive map encoding represents the environmental map from the initial position to the target position. The neurons included in the cognitive map construction module are differential equation neurons with adjustable time constants.

[0077] In this step, the cognitive map construction module of the unmanned system navigation model uses the landmark encoding, obstacle encoding, motion parameter value encoding, position encoding, and target position encoding generated in the previous steps to comprehensively process and fuse this multi-modal information. By using differential equation neurons with adjustable time constants to dynamically model and iteratively update the input data, a cognitive map encoding reflecting the complete environmental structure between the initial position and the target position is generated. In this application, the neurons included in the cognitive map construction module are differential equation neurons with adjustable time constants, which compress the traditional multiplication and addition operations within a single neuron and use the Euler method as the differential solver. This not only significantly improves the operation speed and makes it run faster than existing artificial neural networks, but also its bio-inspired characteristics, that is, using differential equation neurons with adjustable time constants to simulate the place cells and grid cells in the hippocampus and entorhinal cortex of organisms, enable the unmanned system navigation model to capture and continuously evolve spatio-temporal information more precisely, thus ensuring that the system can construct and update the internal environmental model in real time in complex or dynamic scenarios, providing solid and accurate data support for subsequent path planning, navigation decision-making, and action execution.

[0078] Step S105: Through the decision-making action module of the unmanned system navigation model, process the cognitive map encoding to generate a control parameter value for controlling the drive motor of the unmanned system. The control parameter value includes: motion direction and motion speed.

[0079] In this step, the decision-making action module of the unmanned system navigation model uses the cognitive map encoding generated in the previous step. This cognitive map encoding integrates the landmark encoding, obstacle encoding, motion parameter value encoding, position encoding, and target position encoding extracted from the continuous environmental image sequence collected by the camera starting from the initial position, and after dynamic modeling and iterative update by differential equation neurons, comprehensively reflects the environmental structure and spatio-temporal information between the initial position and the target position. Based on this cognitive map encoding, the decision-making action module further analyzes the relative position information of obstacle distribution, motion state, and target instructions in the environment, and finally generates control parameter values including motion direction and motion speed. These parameters can accurately guide the drive motor of the unmanned system to achieve real-time and accurate path planning and navigation decision-making, ensuring that the system can always move forward smoothly towards the predetermined target in complex or dynamic environments.

[0080] In this application, an unmanned system equipped with a camera continuously acquires an environmental image sequence starting from an initial moment, providing continuous, high-resolution raw data with accurate timestamps for the entire system. Subsequently, the visual perception module preprocesses these image sequences and extracts depth features, generating landmark codes, obstacle codes, motion parameter value codes, and position codes, fully capturing the geometric, obstacle, and dynamic information of the environment. Then, the instruction coding module further encodes the target position information relative to the initial position to generate a target position code, thereby clarifying the target direction. Next, the cognitive map construction module fuses the aforementioned various codes and dynamically models and iteratively updates the input data by using differential equation neurons with adjustable time constants. The traditional multiplication and addition operations are compressed into a single neuron by using an Euler method differential solver, thus significantly improving the operation speed and generating a cognitive map code that comprehensively represents the environmental structure from the initial position to the target position. Finally, the decision-making action module analyzes the obstacle, motion state, and target information in the environment based on this cognitive map code, generates control parameter values including the motion direction and motion speed, and guides the drive motor to achieve precise motion control. By efficiently integrating multi-modal information and accelerating operations using bio-inspired differential equation neurons, this application not only significantly improves the real-time response ability and operation efficiency of the system but also can dynamically and accurately construct an environmental model and perform path planning, achieving high-precision, autonomous, and robust navigation decisions in complex dynamic environments.

[0081] In one embodiment, the cognitive map construction module of the unmanned system navigation model processes the landmark code, the obstacle code, the motion parameter value code, the position code, and the target position code to generate a cognitive map code, including:

[0082] Input the landmark code, the obstacle code, the motion parameter value code, the position code, and the target position code into the differential equation neurons in the cognitive map construction module to obtain the cognitive map neuron state;

[0083] Process the cognitive map neuron state through a first fully connected layer to obtain the cognitive map code.

[0084] In this embodiment, the landmark encoding, obstacle encoding, motion parameter value encoding, current position encoding, and target position encoding are input into the differential equation neurons in the cognitive map construction module. By utilizing the adjustable time constant characteristic of these neurons, dynamic modeling of environmental information is performed to obtain the cognitive map neuron state. Subsequently, the cognitive map neuron state is further processed through the first fully connected layer to extract effective features and generate the cognitive map encoding, thereby achieving an efficient representation of the environmental information from the initial position to the target position. In addition, in this embodiment, by adopting the cognitive map construction module, only the key information related to the navigation task is recorded, rather than storing the complete high-precision map details, thus significantly reducing the memory occupancy and improving the computational efficiency.

[0085] In some embodiments, the cognitive map functional area can use 24 differential equation neurons as recording nodes to simulate the place cells and grid cells in the hippocampus and entorhinal cortex of organisms, thereby completing the accurate position representation and path integration tasks. This functional area can learn and predict the map representation of the current position by processing continuous motion inputs (such as linear velocity and angular velocity). Specifically, this model receives, at each time step, the sine and cosine value encodings of the linear velocity and angular velocity extracted by the perception layer as inputs, and uses the cognitive map network composed of differential equation neurons to deeply capture the dynamic information in the time series. Combining with the Euler method differential solver to accelerate the calculation, the computational efficiency and real-time performance are enhanced. Finally, the position representation is generated through the subsequent linear output layer to achieve its own precise positioning and environmental map construction, providing efficient and stable environmental perception and decision-making support for the autonomous navigation of the unmanned system.

[0086] In one embodiment, the neurons included in the visual perception module are differential equation neurons with adjustable time constants.

[0087] Through the visual perception module of the unmanned system navigation model, feature extraction is performed on the environmental image sequence. According to the extracted image features, landmark encoding, obstacle encoding, motion parameter value encoding, and current position encoding are generated, including:

[0088] Preprocess each of the environmental images in the environmental image sequence to obtain feature vectors.

[0089] Input the feature vectors into the differential equation neurons in the visual perception module to obtain the visual neuron state.

[0090] Process the visual neuron state through the second fully connected layer to obtain the landmark encoding, the obstacle encoding, the motion parameter value encoding, and the current position encoding.

[0091] In this embodiment, the visual perception module includes differential equation neurons with adjustable time constants to enhance the ability to extract dynamic features from environmental image sequences and improve the efficiency and accuracy of information processing. Specifically, this process first preprocesses each environmental image in the environmental image sequence received by the visual perception module of the unmanned system navigation model to remove noise, enhance key features, and convert them into feature vectors suitable for neural network processing. Subsequently, the feature vectors are input into the differential equation neurons in the visual perception module, which through time constant regulation, realizes the temporal dynamic modeling of image features, captures environmental changes and motion information, and obtains the visual neuron states. Then, through the second fully connected layer, non-linear transformation and feature extraction are performed on the visual neuron states to generate key coding information for characterizing the environment where the unmanned system is located, including landmark coding, obstacle coding, motion parameter value coding, and location coding. These codings not only accurately describe the environmental features around the unmanned system but also provide efficient and stable input data for subsequent cognitive map construction and path planning, ensuring the real-time performance and robustness of the navigation system in complex environments.

[0092] Exemplarily, the visual perception module includes 20 differential equation neurons, and these 20 differential equation neurons are used for image feature extraction to improve the computing efficiency and information processing ability. Specifically, this process first preprocesses the environmental image sequence collected by the camera carried by the unmanned system, downsamples the image pixels to 256×144, and evenly divides it into 8×8 image blocks, a total of 32×18. Subsequently, each image block is sequentially expanded into a vector of length 64, and the vectors of all blocks are sequentially concatenated to form a 576×64 vector matrix. Then, this vector matrix is used as the input and enters the visual perception module, where 20 differential equation neurons are used to extract visual information from the 576×64 input matrix. Specifically, this vector matrix first passes through a 64×1 visual layer neuron fully connected input layer and is converted into a 576×1 feature vector, and then is input into 20 differential equation neurons through neuron perception weights to obtain 20 visual neuron states. Subsequently, these neuron states are further processed through the fully connected layer to generate landmark coding, obstacle coding, motion parameter value coding, and location coding. In this application, by projecting image information onto differential equation neurons for feature extraction, large-scale scanning calculations are avoided, so that the structural information in the image can be extracted more quickly, improving the extraction efficiency and computing speed of image features, enabling the unmanned system to achieve more efficient perception and navigation in complex environments.

[0093] In this embodiment, the environmental images collected by the unmanned system are first downsampled and segmented, converted into a feature vector matrix, and feature extraction is performed through the differential equation neurons in the visual perception module. For example, 20 differential equation neurons are used in the visual area to process the input feature vector matrix, and the encoded information representing landmarks, obstacles, motion parameter values, and the current position is mapped through a fully connected layer. Compared with the traditional feature extraction method based on convolutional neural networks, this method utilizes the time dynamic characteristics of differential equation neurons to capture the spatial structure and motion information of the image in a shorter calculation time, thereby improving the real-time performance and computational efficiency of visual perception, enabling the unmanned system to perceive the environment more quickly and complete path planning and navigation decisions.

[0094] In one embodiment, processing the cognitive map encoding through the decision-making action module of the unmanned system navigation model to generate a control parameter value for controlling the drive motor of the unmanned system includes:

[0095] Processing the cognitive map encoding through the decision-making action module to obtain the action neuron state;

[0096] Determining the movement direction of the unmanned system during obstacle avoidance according to the positive or negative of the action neuron state, where the movement direction includes left and right directions;

[0097] Determining the movement speed of the unmanned system during obstacle avoidance according to the value of the action neuron state.

[0098] In this embodiment, the decision-making action module generates a control parameter value for controlling the drive motor of the unmanned system based on the processing of the cognitive map encoding to achieve precise obstacle avoidance and path planning. First, the cognitive map encoding is transmitted as input to the decision-making action module and the action neuron state is obtained through calculation. In practical applications, the action neuron state can be the last component of the neuron state vector after iteration by the Euler method. Subsequently, according to the positive or negative value of the action neuron state, the movement direction of the unmanned system during obstacle avoidance is determined, that is, the driving trajectory is adjusted left or right to avoid obstacles and maintain a stable navigation path. At the same time, according to the numerical value of the action neuron state, the movement speed of the unmanned system is calculated to ensure that it can flexibly adjust the direction and reasonably control the speed during obstacle avoidance, avoiding instability caused by drastic speed changes. In practical applications, the action area neuron state is a number between -1 and 1, representing the action degree of the unmanned system. For example, the speeds of the unmanned system moving left and right during obstacle avoidance, -1 represents moving left at 1 m / s, and 0.5 represents moving right at 0.5 m / s.

[0099] In this embodiment, the differential equation neurons are used to analyze environmental information, enabling the unmanned system to adaptively adjust its motion parameters, improve navigation efficiency, and enhance motion stability and safety in complex environments.

[0100] In one embodiment, the neurons included in both the instruction encoding module and the decision-making action module are differential equation neurons with adjustable time constants; among them, the output of the th differential equation neuron is the neuron state , and the input of the th differential equation neuron includes: the input feature vector , and the states of other differential equation neurons connected to the th differential equation neuron.

[0101] Any one of the differential equation neurons conforms to the following expression (1):

[0102] (1)

[0103] Where represents the state of the th differential equation neuron; represents the state of the th differential equation neuron; represents the input feature vector; represents the time constant that controls the dynamic change of the differential equation neuron; represents the set of units connected to the th differential equation neuron; represents the connection weight for the th differential equation neuron to transmit information to the th differential equation neuron; represents the perception weight from the input feature vector to the th differential equation neuron; represents the activation function.

[0104] In this application, the neurons included in both the instruction encoding module and the decision-making action module are also differential equation neurons with adjustable time constants. Among them, the output of the th differential equation neuron is the neuron state , and the input of the th differential equation neuron includes: the input feature vector , and the states of other differential equation neurons connected to the th differential equation neuron.

[0105] It should be noted that in the unmanned system navigation model, for different modules, the input feature vector are different. For example, in the visual perception module, the input feature vector is a visual feature vector. In the cognitive map construction module, the input feature vector is a feature vector representing landmark coding, obstacle coding, motion parameter value coding, location coding, and target location coding. In the instruction coding module, the input feature vector is a feature vector including the target location in the target instruction relative to the initial location. In the decision-making action module, the input feature vector is a feature vector representing the cognitive map coding.

[0106] Among them, any differential equation neuron conforms to the following expression (1):

[0107] (1)

[0108] Among them, represents the state of the th differential equation neuron; represents the state of the th differential equation neuron; represents the input feature vector; represents the time constant that controls the dynamic change of the differential equation neuron; represents the set of units connected to the th differential equation neuron; represents the connection weight from the th differential equation neuron to the th differential equation neuron; represents the perception weight from the input feature vector to the th differential equation neuron, represents the activation function.

[0109] In some embodiments, the present application also provides a training method for the unmanned system navigation model. The method is as Figure 2 shown and includes:

[0110] Step S201: Collect a sample environmental image sequence through the camera carried by the unmanned system, and obtain the actual values of the control parameters of the drive motor of the unmanned system during the collection of the sample environmental image sequence.

[0111] In this step, the camera carried by the unmanned system is used to collect a sequence of sample environment images, and the actual values of the control parameters of the driving motor of the unmanned system during the collection process are recorded synchronously for subsequent model training and optimization. Specifically, during the sample data collection phase, the unmanned system travels along a preset path and obtains a sequence of environment images in the form of continuous frames through the carried camera. At the same time, the actual values of the control parameters of the driving motor, including the movement direction, movement speed, etc., are obtained in real time to accurately reflect the driving state of the unmanned system in different environments.

[0112] Step S202: Mark the initial sample position according to the position of the unmanned system when the first sample environment image in the collected sequence of sample environment images is obtained, and mark the sample target position according to the position of the unmanned system when the last sample environment image in the collected sequence of sample environment images is obtained.

[0113] In this step, the initial sample position is marked according to the position of the unmanned system when the first sample environment image in the collected sequence of sample environment images is obtained. At the same time, the sample target position is marked according to the position of the unmanned system when the last sample environment image in the collected sequence of sample environment images is obtained. Specifically, during the sample data collection process, the position information of the unmanned system is recorded synchronously with the environment images to ensure the spatio-temporal consistency of the data. By determining the start and end positions of the sequence of sample environment images, the movement trajectory of the unmanned system during this data collection process can be clarified, providing a reference for the path planning and behavior decision-making of the subsequent navigation model. In addition, this step helps to establish a mapping of position information based on environmental features, enabling the training model to better understand the spatial relationship from the initial position to the target position and improving the learning effect and path inference ability of the navigation system.

[0114] Step S203: Input the sequence of sample environment images and the sample target position into the navigation model of the unmanned system to be trained, and obtain the predicted values of the control parameters of the driving motor of the unmanned system.

[0115] In this step, the sample environment image sequence and the sample target position are input into the unmanned system navigation model to be trained to obtain the control parameter prediction value of the drive motor of the unmanned system. Specifically, the unmanned system navigation model to be trained receives the collected sample environment image sequence, and uses the visual perception module to extract environmental features, generates landmark codes, obstacle codes, motion parameter value codes and location codes, and at the same time, combines the marked sample target positions to construct a cognitive map code to characterize the environmental information of the unmanned system from the initial sample position to the sample target position. On this basis, the decision-making action module infers the unmanned system's travel strategy based on the cognitive map code, and outputs the control parameter prediction value for the drive motor control, including the direction of movement and the speed of movement. This step realizes the construction of the training input of the unmanned system navigation model, so that the model can learn the mapping relationship from environmental perception to motion decision-making in the subsequent training process, and improve the autonomous navigation capability of the unmanned system in complex environments.

[0116] Step S204: Update the perception weights and connection weights of each differential equation neuron in the unmanned system navigation model to be trained based on the predicted value of the control parameter of the drive motor of the unmanned system and the actual value of the control parameter of the drive motor of the unmanned system to obtain the trained unmanned system navigation model.

[0117] In one embodiment, there is provided Figure 3 The overall architecture diagram of the unmanned system navigation model based on cognitive map is shown in Figure 3 As shown:

[0118] The framework diagram shows the architecture of image acquisition, preprocessing, and unmanned system navigation model, as well as the process of driving the unmanned system. The unmanned system navigation model is divided into multiple modules, including: visual perception module, instruction encoding module, cognitive map construction module, and decision-making action module.

[0119] Among them, the visual perception module is divided into visual functional area and integration functional area, the instruction encoding module includes target instruction functional area, the cognitive map construction module includes cognitive map functional area, and the decision-making action module is divided into decision-making functional area and action functional area.

[0120] like Figure 3 As shown, the input of the architecture is Figure 3 The dashed box A in the figure represents the camera on the unmanned system that collects visual information about the external environment and obtains a sequence of environmental images. The example in the figure is a forest scene, which shows that the model can perform navigation tasks in a natural and complex environment. The input of this architecture is Figure 3 The dotted box in also includes the relative position information of the target position.

[0121] An intermediate processing end of this architecture isFigure 3 The dashed box B in it is the visual perception module. The left side is the visual function area, representing the visual perception network for extracting image features. This module inputs the original image obtained by the camera into the differential equation neurons to complete the preliminary extraction and encoding of landmarks, obstacles, motion parameter values, and the location. The right side is the integration function area, which is used to perform further feature extraction on the landmark encoding, obstacle encoding, motion parameter value encoding, and location encoding extracted by the visual function area.

[0122] An intermediate processing end of this architecture is Figure 3 The dashed box C in it is the instruction encoding module, including the target instruction function area, which is used to encode according to the input target relative position information to obtain the target position encoding.

[0123] An intermediate processing end of this architecture is Figure 3 The dashed box D in it is the cognitive map construction module, including the cognitive map function area, which appears in the form of a similar SLAM map or grayscale environment map in the figure, representing the core concept of the "cognitive map". Here, the key encodings (such as landmarks, obstacles, motion parameter values, current location, etc.) obtained by the visual perception module and the target position encoding obtained by the instruction encoding module are jointly input into the network constructed by differential equation neurons, so as to generate or update the "cognitive map encoding". Among them, the network structure constructed by differential equation neurons is also shown in the dashed box D.

[0124] An intermediate processing end of this architecture is Figure 3 The dashed box E in it is the decision-making action module, which is divided into a decision-making function area and an action function area. Among them, the left side is the decision-making function area, which is connected to the output of the cognitive map module and is used to make navigation decisions according to the environmental encoding and target position information. This network will output control parameters such as the motion direction and motion speed. The right side is the action function area, which is connected to the unmanned system in the dashed box F at the output end, indicating that the control parameters such as the motion direction and motion speed output by the decision-making function area are converted into control instructions for the drive motor (such as motor speed or servo angle) and output to the unmanned system by the action function area, so that the unmanned system can perform actual actions. The waveform diagram therein reflects the timing characteristics of the control signal. Figure 3 The dashed box F in it is connected to the unmanned system, indicating that the control parameters such as the motion direction and motion speed output by the decision-making function area are converted into control instructions for the drive motor (such as motor speed or servo angle) and output to the unmanned system by the action function area, so that the unmanned system can perform actual actions. The waveform diagram therein reflects the timing characteristics of the control signal.

[0125] In this step, based on the predicted control parameter values of the drive motors of the unmanned system and the actual control parameter values of the drive motors of the unmanned system, the perception weights and connection weights of each differential equation neuron in the unmanned system navigation model to be trained are updated to obtain the trained unmanned system navigation model. Specifically, this process adopts an error backpropagation mechanism, uses the deviation between the predicted control parameter values and the actual values as the loss signal, and adjusts the weights of the differential equation neurons in the visual perception module, the cognitive map construction module, and the decision-making action module to optimize the feature extraction, environment modeling, and path decision-making capabilities. Since the differential equation neurons can perform multiplication and addition operations within a single neuron and use the Euler method differential solver, the optimization speed is faster than that of traditional neural networks, making the training process more efficient. In addition, the differential equation neurons in the cognitive map construction module learn the dynamic information in the time series, improve the accuracy of the unmanned system's self-positioning and path inference, thereby enhancing the autonomous navigation performance of the unmanned system in complex environments and achieving precise obstacle avoidance and stable motion control.

[0126] This application proposes a method for navigating an unmanned system based on a cognitive map. The method includes: first, collecting an environmental image sequence through a camera carried by the unmanned system; then, through the visual perception module of the unmanned system navigation model, extracting features from the environmental image sequence, and generating landmark encoding, obstacle encoding, motion parameter value encoding, and location encoding based on the extracted image features; also, through the instruction encoding module of the unmanned system navigation model, encoding the target position relative to the initial position in the target instruction to generate target position encoding; then, through the cognitive map construction module of the unmanned system navigation model, processing the landmark encoding, the obstacle encoding, the motion parameter value encoding, the location encoding, and the target position encoding to generate cognitive map encoding. The neurons included in the cognitive map construction module are differential equation neurons with adjustable time constants; finally, through the decision-making action module of the unmanned system navigation model, processing the cognitive map encoding to generate control parameter values for controlling the drive motors of the unmanned system. The control parameter values include: motion direction and motion speed.

[0127] This application uses a pre-trained unmanned system navigation model based on a cognitive map. The neurons included in the cognitive map construction module of the unmanned system navigation model are differential equation neurons with adjustable time constants, which can accelerate the operation speed within the neurons. In addition, by constructing the cognitive map encoding through the cognitive map construction module, the map construction speed is shortened, so that the navigation speed and accuracy of the unmanned system can be significantly improved in large-scale real scenarios.

[0128] Based on the same inventive concept, in the second aspect of the present application, a navigation device for an unmanned system based on a cognitive map is provided, as Figure 4 shown. The device includes:

[0129] An acquisition unit 201, configured to collect an environmental image sequence through a camera carried by the unmanned system. The environmental image sequence includes multiple environmental images collected within a time period starting from an initial moment, and the unmanned system is located at an initial position at the initial moment;

[0130] A feature extraction unit 202, configured to perform feature extraction on the environmental image sequence through a visual perception module of the unmanned system navigation model, and generate a landmark code, an obstacle code, a motion parameter value code, and a location code according to the extracted image features;

[0131] An encoding unit 203, configured to encode a target position relative to the initial position in a target instruction through an instruction encoding module of the unmanned system navigation model to generate a target position code;

[0132] A first processing unit 204, configured to process the landmark code, the obstacle code, the motion parameter value code, the location code, and the target position code through a cognitive map construction module of the unmanned system navigation model to generate a cognitive map code. The cognitive map code represents an environmental map from the initial position to the target position, and the neurons included in the cognitive map construction module are differential equation neurons with adjustable time constants;

[0133] A second processing unit 205, configured to process the cognitive map code through a decision-making action module of the unmanned system navigation model to generate control parameter values for controlling a driving motor of the unmanned system. The control parameter values include: a motion direction and a motion speed.

[0134] Optionally, the processing of the landmark code, the obstacle code, the motion parameter value code, the location code, and the target position code through the cognitive map construction module of the unmanned system navigation model to generate a cognitive map code, the first processing unit 204 includes:

[0135] A first input sub-unit, configured to input the landmark code, the obstacle code, the motion parameter value code, the location code, and the target position code into the differential equation neurons in the cognitive map construction module to obtain a cognitive map neuron state;

[0136] A first processing sub-unit, configured to process the cognitive map neuron state through a first fully connected layer to obtain a cognitive map code.

[0137] Optionally, the neurons included in the visual perception module are differential equation neurons with adjustable time constants;

[0138] Through the visual perception module of the unmanned system navigation model, feature extraction is performed on the environmental image sequence. According to the extracted image features, landmark coding, obstacle coding, motion parameter value coding, and location coding are generated. The feature extraction unit 202 includes:

[0139] A second processing subunit for preprocessing each of the environmental images in the environmental image sequence to obtain feature vectors;

[0140] A second input subunit for inputting the feature vectors into the differential equation neurons in the visual perception module to obtain visual neuron states;

[0141] A third processing subunit for processing the visual neuron states through a second fully connected layer to obtain the landmark coding, the obstacle coding, the motion parameter value coding, and the location coding.

[0142] Optionally, through the decision-making action module of the unmanned system navigation model, the cognitive map coding is processed to generate a control parameter value for controlling the drive motor of the unmanned system. The second processing unit 205 includes:

[0143] A fourth processing subunit for processing the cognitive map coding through the decision-making action module to obtain action neuron states;

[0144] A first determination subunit for determining the movement direction of the unmanned system during obstacle avoidance according to the positive or negative of the action neuron states. The movement direction includes left and right directions;

[0145] A second determination subunit for determining the movement speed of the unmanned system during obstacle avoidance according to the value of the action neuron states.

[0146] Optionally, the device further includes:

[0147] An acquisition subunit for acquiring a sample environmental image sequence through a camera carried by the unmanned system and obtaining the actual control parameter values of the drive motor of the unmanned system during the acquisition of the sample environmental image sequence;

[0148] A marking subunit for marking an initial sample position according to the position of the unmanned system when the first sample environmental image in the acquired sample environmental image sequence is collected, and marking a sample target position according to the position of the unmanned system when the last sample environmental image in the acquired sample environmental image sequence is collected;

[0149] A third input subunit, configured to input the sample environment image sequence and the sample target position into an unmanned system navigation model to be trained, and obtain a predicted value of the control parameter of the driving motor of the unmanned system;

[0150] An update subunit, configured to update the perception weights and connection weights of each differential equation neuron in the unmanned system navigation model to be trained according to the predicted value of the control parameter of the driving motor of the unmanned system and the actual value of the control parameter of the driving motor of the unmanned system, so as to obtain the trained unmanned system navigation model.

[0151] Based on the same inventive concept, in the third aspect of the present application, there is provided an unmanned system based on a cognitive map, where the system includes the unmanned system navigation device as described in the second aspect of the present application, and / or the unmanned system is configured to execute the steps of the unmanned system navigation method as described in the first aspect of the present application.

[0152] Based on the same inventive concept, in the fourth aspect of the present application, there is provided an electronic device 100 as Figure 5 shown, including a processor 120, a memory 110, and a program or instruction stored on the memory 110 and executable on the processor 120. When the program or instruction is executed by the processor 120, the steps of the unmanned system navigation method as described in the first aspect of the present application are implemented.

[0153] In the fifth aspect of the present application, there is provided a readable storage medium, where a program or instruction is stored on the readable storage medium. When the program or instruction is executed by a processor, the steps of the unmanned system navigation method as described in the first aspect of the present application are implemented.

[0154] In each embodiment of this specification, the key points described are the differences from other embodiments. For the same or similar parts among the various embodiments, reference may be made to each other.

[0155] Those skilled in the art should understand that the embodiments of the present application may be provided as a method, an apparatus, or a computer program product. Therefore, the embodiments of the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0156] Embodiments of the present application are described with reference to the flowcharts and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing terminal devices generate a device for implementing the functions specified in one process Figure 1 or in one or more processes and / or blocks Figure 1 or in one or more blocks.

[0157] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device that implements the functions specified in one process Figure 1 or in one or more processes and / or blocks Figure 1 or in one or more blocks.

[0158] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, such that a series of operation steps are executed on the computer or other programmable terminal device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable terminal device provide steps for implementing the functions specified in one process Figure 1 or in one or more processes and / or blocks Figure 1 or in one or more blocks.

[0159] Although the preferred embodiments of the embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concepts. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present application.

[0160] Finally, it should also be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or terminal device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or terminal device comprising the said element.

[0161] The above provides a detailed introduction to a method, device, unmanned system, equipment and storage medium for unmanned system navigation. Specific examples are used in this text to elaborate on the principle and implementation of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation and application scope. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A navigation method for an unmanned system based on a cognitive map, characterized in that, The method includes: Collecting an environmental image sequence through a camera carried by an unmanned system, where the environmental image sequence includes multiple environmental images collected within a time period starting from an initial moment, and the unmanned system is located at an initial position at the initial moment; Extracting features from the environmental image sequence through a visual perception module of an unmanned system navigation model, and generating a landmark code, an obstacle code, a motion parameter value code, and a location code according to the extracted image features; Encoding a target position relative to the initial position in a target instruction through an instruction encoding module of an unmanned system navigation model to generate a target position code; Processing the landmark code, the obstacle code, the motion parameter value code, the location code, and the target position code through a cognitive map construction module of an unmanned system navigation model to generate a cognitive map code, where the cognitive map code represents an environmental map from the initial position to the target position; Processing the cognitive map code through a decision-making action module of an unmanned system navigation model to generate control parameter values for controlling a driving motor of the unmanned system, where the control parameter values include: a motion direction and a motion speed; Neurons included in each of the cognitive map construction module, the visual perception module, the instruction encoding module, and the decision-making action module are all differential equation neurons with adjustable time constants, and the expression that the differential equation neurons conform to is: Among them, represents the state of the th differential equation neuron; represents the state of the th differential equation neuron; represents the input feature vector; represents the time constant that controls the dynamic change of the differential equation neuron; represents the set of units connected to the th differential equation neuron; represents the connection weight for the th differential equation neuron to transmit information to the th differential equation neuron; represents the perception weight from the input feature vector to the th differential equation neuron; represents the activation function; In the visual perception module, the input feature vector is the visual feature vector. In the cognitive map construction module, the input feature vector is the feature vector representing landmark coding, obstacle coding, motion parameter value coding, location coding, and target location coding. In the instruction coding module, the input feature vector is the feature vector containing the target location relative to the initial location in the target instruction. In the decision-making action module, the input feature vector is the feature vector representing the cognitive map coding.

2. The method for navigating an unmanned system based on a cognitive map according to claim 1, wherein The processing of the landmark code, the obstacle code, the motion parameter value code, the location code, and the target position code through the cognitive map construction module of the unmanned system navigation model to generate a cognitive map code includes: Inputting the landmark code, the obstacle code, the motion parameter value code, the location code, and the target position code into a differential equation neuron in the cognitive map construction module to obtain a cognitive map neuron state; Processing the cognitive map neuron state through a first fully connected layer to obtain a cognitive map code.

3. The navigation method of the unmanned system based on the cognitive map according to claim 1, wherein Extracting features from the environmental image sequence through a visual perception module of an unmanned system navigation model, and generating a landmark code, an obstacle code, a motion parameter value code, and a location code according to the extracted image features, including: Preprocessing each of the environmental images in the environmental image sequence to obtain a feature vector; Inputting the feature vector into a differential equation neuron in the visual perception module to obtain a visual neuron state; Processing the visual neuron state through a second fully connected layer to obtain the landmark code, the obstacle code, the motion parameter value code, and the location code.

4. The navigation method for an unmanned system based on a cognitive map according to claim 1, wherein, The processing of the cognitive map code through the decision-making action module of the unmanned system navigation model to generate control parameter values for controlling the driving motor of the unmanned system includes: Processing the cognitive map code through the decision-making action module to obtain an action neuron state; Determining the motion direction of the unmanned system during obstacle avoidance according to the positive or negative of the action neuron state, where the motion direction includes left and right directions; Determine the movement speed of the unmanned system during obstacle avoidance according to the numerical value of the action neuron state.

5. The method for navigating an unmanned system based on a cognitive map according to any one of claims 1-4, characterized in that, The method further includes: Collect a sequence of sample environment images through a camera carried by the unmanned system, and obtain the actual value of the control parameter of the drive motor of the unmanned system during the collection of the sequence of sample environment images; Mark the initial sample position according to the position of the unmanned system when the first sample environment image in the sequence of sample environment images is collected, and mark the sample target position according to the position of the unmanned system when the last sample environment image in the sequence of sample environment images is collected; Input the sequence of sample environment images and the sample target position into the unmanned system navigation model to be trained, and obtain the predicted value of the control parameter of the drive motor of the unmanned system; Update the perception weight and connection weight of each differential equation neuron in the unmanned system navigation model to be trained according to the predicted value of the control parameter of the drive motor of the unmanned system and the actual value of the control parameter of the drive motor of the unmanned system, and obtain the trained unmanned system navigation model.

6. A navigation device for an unmanned system based on a cognitive map, characterized in that, The device includes: A collection unit for collecting a sequence of environment images through a camera carried by the unmanned system. The sequence of environment images includes multiple environment images collected during a time period starting from an initial moment. The unmanned system is located at an initial position at the initial moment; A feature extraction unit for extracting features from the sequence of environment images through the visual perception module of the unmanned system navigation model, and generating a landmark code, an obstacle code, a motion parameter value code, and a location code according to the extracted image features; An encoding unit for encoding the target position relative to the initial position in the target instruction through the instruction encoding module of the unmanned system navigation model to generate a target position code; A first processing unit for processing the landmark code, the obstacle code, the motion parameter value code, the location code, and the target position code through the cognitive map construction module of the unmanned system navigation model to generate a cognitive map code, where the cognitive map code represents the environmental map from the initial position to the target position; A second processing unit for processing the cognitive map code through the decision-making action module of the unmanned system navigation model to generate a control parameter value for controlling the drive motor of the unmanned system. The control parameter value includes: a motion direction and a motion speed; each neuron included in the cognitive map construction module, the visual perception module, the instruction encoding module, and the decision-making action module is a differential equation neuron with an adjustable time constant, and the differential equation neuron conforms to the expression: Among them, represents the state of the th differential equation neuron; represents the state of the th differential equation neuron; represents the input feature vector; represents the time constant that controls the dynamic change of the differential equation neuron; represents the set of units connected to the th differential equation neuron; represents the connection weight for the th differential equation neuron to transmit information to the th differential equation neuron; represents the perception weight from the input feature vector to the th differential equation neuron; represents the activation function; in the visual perception module, the input feature vector is the visual feature vector, in the cognitive map construction module, the input feature vector is the feature vector representing landmark coding, obstacle coding, motion parameter value coding, location coding, and target location coding, in the instruction coding module, the input feature vector is the feature vector containing the target location relative to the initial position in the target instruction, in the decision-making action module, the input feature vector is the feature vector representing the cognitive map coding.

7. An unmanned system based on a cognitive map, characterized in that, The system includes the unmanned system navigation device as claimed in claim 6, and / or the unmanned system is used to execute the steps of the unmanned system navigation method as claimed in any one of claims 1-5.

8. An electronic device, characterized in that, It includes a processor, a memory, and a program or instruction stored on the memory and executable on the processor. When the program or instruction is executed by the processor, it realizes the steps of the unmanned system navigation method as claimed in any one of claims 1-5.

9. A readable storage medium, characterized in that, The program or instructions are stored on the readable storage medium, and when the program or instructions are executed by the processor, the steps of the unmanned system navigation method described in any one of claims 1-5 are implemented.

Images