Robot Automatic Navigation and Return Method, Device, Equipment and Storage Medium

Abstract

The present invention relates to the technical field of robot navigation, and discloses a method, device, equipment and storage medium for automatic navigation and return of a robot. The method includes: extracting multi-scale environmental feature data from the environmental image data collected by the mobility robot to obtain environmental feature data; constructing an initial navigation path; performing object detection and feature re-identification on the real-time environmental image to obtain a real-time environmental perception result; performing extended Kalman filter fusion processing on multi-sensor data to obtain a robot-environment interaction state feature; analyzing the straight-line segment and turning segment trajectories according to the robot-environment interaction state feature to construct a smooth navigation trajectory; continuously updating the environmental feature data and the working state data of the mobility robot during navigation to construct a dynamic environmental map; when a return signal is received, quickly planning an automatic return path based on the dynamic environmental map. The present invention can improve the autonomy and working efficiency of the mobility robot.

Description

Technical Field

[0001] The present invention relates to the technical field of robot navigation, and particularly to a method, device, equipment and storage medium for automatic navigation and return of a robot. Background Art

[0002] Robots can not only provide services such as food delivery and guidance, but also greatly improve work efficiency and customer experience. However, existing mobility robots often have difficulty in achieving stable and reliable autonomous navigation and return functions in complex and changeable indoor environments, which severely limits their practicality and work efficiency. Traditional navigation methods usually rely on pre-established static maps and cannot effectively cope with uncertain factors in dynamic environments, such as moving obstacles and crowded areas.

[0003] In addition, existing navigation algorithms often have problems of low computational efficiency when dealing with large-scale environmental data and are difficult to meet real-time requirements. At the same time, due to the lack of in-depth understanding of environmental characteristics and effective fusion of multi-sensor data, existing methods are prone to problems such as positioning deviation and unreasonable path planning in complex scenarios. These problems not only affect the accuracy and smoothness of navigation, but may also lead to delays, collisions and other situations that affect service quality during the execution of tasks or return of the robot. Summary of the Invention

[0004] The present invention provides a method, device, equipment and storage medium for automatic navigation and return of a robot, which can improve the autonomy and work efficiency of a mobility robot.

[0005] In a first aspect, the present invention provides a method for automatic navigation and return of a robot, the method for automatic navigation and return of the robot comprising:

[0006] Performing multi-scale environmental feature extraction on environmental image data collected by a mobility robot to obtain environmental feature data;

[0007] Constructing an initial navigation path according to the environmental feature data and historical navigation data;

[0008] Based on the initial navigation path, performing target detection and feature re-identification on real-time environmental images to obtain a real-time environmental perception result;

[0009] According to the real-time environmental perception result, performing extended Kalman filter fusion processing on multi-sensor data to obtain a robot-environment interaction state feature;

[0010] Performing straight-line segment and turning segment trajectory analysis according to the robot-environment interaction state feature to construct a smooth navigation trajectory;

[0011] During the navigation process, continuously update the environmental feature data and the working state data of the mobility robot, and construct a dynamic environmental map; when a return signal is received, quickly plan an automatic return path based on the dynamic environmental map.

[0012] In a second aspect, the present invention provides a robot automatic navigation and return device, and the robot automatic navigation and return device includes:

[0013] An extraction module, configured to perform multi-scale environmental feature extraction on the environmental image data collected by the mobility robot to obtain environmental feature data;

[0014] A construction module, configured to construct an initial navigation path according to the environmental feature data and historical navigation data;

[0015] An identification module, configured to perform target detection and feature re-identification on the real-time environmental image based on the initial navigation path to obtain a real-time environmental perception result;

[0016] A processing module, configured to perform extended Kalman filter fusion processing on multi-sensor data according to the real-time environmental perception result to obtain a robot-environment interaction state feature;

[0017] An analysis module, configured to perform straight-line segment and turning segment trajectory analysis according to the robot-environment interaction state feature to construct a smooth navigation trajectory;

[0018] A planning module, configured to continuously update the environmental feature data and the working state data of the mobility robot during the navigation process, and construct a dynamic environmental map; when a return signal is received, quickly plan an automatic return path based on the dynamic environmental map.

[0019] In a third aspect of the present invention, a computer device is provided, including: a memory and at least one processor, and instructions are stored in the memory; the at least one processor calls the instructions in the memory to enable the computer device to execute the above-mentioned robot automatic navigation and return method.

[0020] In a fourth aspect of the present invention, a computer-readable storage medium is provided, and instructions are stored in the computer-readable storage medium, and when it runs on a computer, it enables the computer to execute the above-mentioned robot automatic navigation and return method.

[0021] In the technical solution provided by the present invention, the multi-scale environmental feature extraction technology combined with global-local contrast learning can capture environmental information more comprehensively and accurately, improving the accuracy and robustness of environmental perception. The adaptive uncertainty path planning model effectively copes with various uncertainty factors in complex dynamic environments by dynamically constructing an uncertain parameter set, improving the adaptability and reliability of path planning. The TensorRT-accelerated YOLOv5 object detection and OSNet feature re-identification technologies significantly improve the efficiency and accuracy of real-time environmental perception. The extended Kalman filter fuses and processes multi-sensor data, effectively improving the accuracy of robot-environment interaction state estimation. The differential flatness trajectory generator combines polynomial interpolation and Bezier curves to generate smooth navigation trajectories that satisfy dynamic constraints, improving the comfort and safety of navigation. The continuous update of the dynamic environmental map and the fast return path planning technology enable the robot to respond to environmental changes in a timely manner and achieve safe and efficient automatic return. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0023] Figure 1 It is a schematic diagram of the steps of the robot automatic navigation and return method in the embodiment of the present invention;

[0024] Figure 2 It is a schematic diagram of the structure of the robot automatic navigation and return device in the embodiment of the present invention;

[0025] Figure 3 It is a schematic block diagram of the structure of the computer device in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] An embodiment of the present invention provides a method, device, equipment and storage medium for automatic navigation and return of a robot. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or equipment comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or equipment.

[0027] For ease of understanding, the specific process of the embodiment of the present invention will be described below. Please refer to Figure 1 , an embodiment of the method for automatic navigation and return of a robot in the embodiment of the present invention includes:

[0028] Step S1: Extract multi-scale environmental feature data from the environmental image data collected by the mobility robot to obtain environmental feature data;

[0029] It can be understood that the execution subject of the present invention can be a device for automatic navigation and return of a robot, or a terminal or a server. Specifically, it is not limited here. An embodiment of the present invention will be described by taking the server as the execution subject as an example.

[0030] Specifically, the environmental image data collected by the mobility robot is processed, and the Gaussian pyramid decomposition method is used to perform multi-scale decomposition on the image to obtain a multi-scale image set. The Gaussian pyramid decomposition performs recursive downsampling on the original image, enabling the system to analyze the image at different scales and capture the global and local information of the image. For the obtained multi-scale image set, a multi-layer convolutional neural network is used to extract convolutional features. The convolutional operations include shallow, middle, and deep convolutions, thereby generating a multi-scale feature map set. The features extracted by shallow convolution are usually low-level edge and texture information, while the features extracted by deep convolution are more abstract and high-level semantic information. The multi-scale feature map set is input into the feature pyramid network, where the feature maps at different scales are upsampled and fused to generate a feature map with global semantic information, that is, the global environmental feature map. The feature pyramid network adaptively fuses features at different scales, enabling the final feature map to retain both detailed information and overall semantic information, including the overall layout of the scene and the position and shape of objects. The global environmental feature map is input into the spatial attention module, which generates a key region weight map by calculating the spatial correlation of the feature map. The role of the spatial attention mechanism is to highlight the particularly important regions in the environment, that is, the regions that have a significant impact on robot navigation. By calculating the spatial correlation between different positions in the feature map, the regions that are most critical for environmental understanding are identified, and higher weights are assigned to these regions to generate the key region weight map. Based on the key region weight map, the global environmental feature map is weighted to obtain the local environmental feature map. The global environmental feature map is used as the anchor sample, the local environmental feature map is used as the positive sample, and other feature maps within the batch are randomly selected as negative samples to construct a triplet sample set. The InfoNCE loss function is calculated for the triplet sample set. InfoNCE is a contrastive learning loss that learns a more discriminative feature representation by minimizing the distance between the anchor sample and the positive sample while maximizing the distance between the anchor sample and the negative sample. In this way, the ability to identify similar environmental features is enhanced, and the discrimination between different environments is improved. Based on the feature similarity score, the weights of the global environmental feature map and the key region weight map are adjusted to obtain more optimized environmental feature data. Through the feedback of the feature similarity score, the degree of attention to different features is continuously adjusted, so that the environmental feature data not only contains global information but also can more effectively highlight the information of the key regions.

[0031] Step S2: Construct an initial navigation path according to the environmental feature data and historical navigation data;

[0032] Specifically, based on the environmental feature data, the current driving environment is spatially gridded, and the continuous spatial environment is discretized into a series of grid cells of a fixed size, thus converting the complex environment into a discrete representation that is convenient for analysis, forming a two-dimensional matrix containing environmental states, namely the two-dimensional environmental state matrix. Each grid cell in the two-dimensional environmental state matrix represents the state information of the corresponding position in the environment, such as the presence or absence of obstacles or the ease of passage. The two-dimensional environmental state matrix is time-stamped and aligned with the historical navigation data. By combining the current environment with the previous navigation data, a spatio-temporal sequence dataset containing environmental states and historical trajectories is constructed. The spatio-temporal sequence dataset is segmented by a sliding window, and multiple overlapping subsequence samples are extracted from the entire spatio-temporal sequence dataset, so that subsequent analysis can better capture the potential temporal correlation and change characteristics in the environment and navigation data. For each subsequence sample, wavelet transform is performed to extract the features of the subsequence in the time-frequency domain. Wavelet transform is a signal processing method that can effectively reveal the change patterns of data at different time scales. To reduce the computational complexity and remove noise, after wavelet transform, only the significant wavelet coefficients are retained through threshold screening, and a feature vector with reduced dimension is obtained. Based on the reduced-dimension feature vector, a latent space representation and a reconstruction error are generated. The latent space representation is a higher-level abstraction of the reduced-dimension feature vector, while the reconstruction error is used to describe the reconstruction quality of these features. On this basis, the kernel density estimation algorithm is used to construct the uncertainty parameter distribution of the current driving environment. Kernel density estimation is a non-parametric statistical method that effectively estimates the probability density of sample data and obtains the probability distribution of uncertainty factors in the driving environment. The uncertainty parameter distribution describes various uncertain factors existing in the current environment and their possibilities, such as the possible positions of obstacles, the speed of dynamic changes, etc. Based on the uncertainty parameter distribution, the Latin hypercube sampling method is used to generate multiple groups of representative uncertainty parameter samples. Latin hypercube sampling is an effective sampling technique that can generate uniformly distributed samples in the multi-dimensional parameter space, thus ensuring that different uncertainty factors can be fully explored. For the multiple groups of generated uncertainty parameter samples, they are respectively input into the path planner based on the rapidly-exploring random tree. The rapidly-exploring random tree is a path planning algorithm that gradually expands the path through random sampling and can quickly find a passage path in a complex environment. For each group of uncertainty parameter samples, the rapidly-exploring random tree planner generates a candidate path, thus obtaining multiple candidate paths, and each candidate path corresponds to a possible environmental state. To select the best initial navigation path from the candidate paths, a multi-objective evaluation is performed on all candidate paths. The evaluation objectives include multiple factors such as path length, driving time, energy consumption, safety, etc. To comprehensively consider the advantages and disadvantages of different objectives, the comprehensive score of each candidate path is calculated by weighted summation.According to the comprehensive score, all candidate paths are sorted in descending order, and the candidate path with the highest score is selected as the initial navigation path.

[0033] Step S3: Based on the initial navigation path, perform object detection and feature re-identification on the real-time environmental image to obtain the real-time environmental perception result;

[0034] Specifically, real-time environmental images are collected based on the initial navigation path, and the real-time environmental images are input into the YOLOv5 network accelerated by TensorRT for processing. The YOLOv5 network is an efficient object detection model, which consists of three main parts: the backbone network, the neck network, and the detection head network. The role of the backbone network is to extract the basic features of the environmental images. It adopts the CSPDarknet53 structure, which contains 5 CSP modules. Each module is composed of multiple residual units and 1×1 convolutional layers, and can efficiently extract the low-level and high-level features of the images, generating the first multi-scale feature map. Through these feature maps, different objects and feature information in the environment are extracted. The first multi-scale feature map is input into the neck network of YOLOv5. The neck network adopts a feature pyramid structure, which contains 3 bottom-up feature fusion layers and 3 top-down feature fusion layers. The role of these fusion layers is to effectively fuse features of different scales, generating the second multi-scale feature map. Through the fusion of multi-scale features, the target features of different sizes and scales in the environment are captured, enabling object detection to adapt to complex environmental conditions and ensuring that various important objects and obstacles can be detected during the robot's navigation. The detection head network of YOLOv5 is used to process the second multi-scale feature map. The detection head network contains 3 parallel detection branches, and each detection branch includes 2 convolutional layers and 1 output layer, which are used to output the object detection results. The detection results include the bounding box coordinates, confidence, and class probability of the target object. Non-maximum suppression processing is performed on the detection results to remove redundant detection boxes, and only the target detection box with the highest confidence is retained, obtaining the filtered target detection results. The image region corresponding to the filtered target detection box is cropped and resized, and the cropped image region contains the detected target object. In order to identify and distinguish the target object, the cropped image region is input into the OSNet network accelerated by TensorRT for feature re-identification. The OSNet network contains a lightweight backbone network and a feature aggregation module. The lightweight backbone network adopts a depthwise separable convolutional structure and contains 4 stages, and each stage is composed of multiple lightweight residual units. Through the processing of the lightweight backbone network, the third multi-scale feature map is obtained. After the feature extraction of the lightweight backbone network is completed, the feature aggregation module is used to process these features. The feature aggregation module adopts an adaptive feature fusion strategy, which includes a channel attention mechanism and a spatial attention mechanism. The channel attention mechanism is used to highlight the feature channels that are important for classification and re-identification, while the spatial attention mechanism is used to enhance the significant features at specific spatial positions. Through the adaptive feature fusion strategy, a fused global feature vector is generated. L2 normalization processing is performed on the fused global feature vector to obtain the target re-identification feature. The object detection results after non-maximum suppression processing are associated and matched with the normalized target re-identification features to obtain the real-time environmental perception results.

[0035] Step S4. According to the real-time environment perception result, perform extended Kalman filter fusion processing on the multi-sensor data to obtain the robot-environment interaction state characteristics;

[0036] Specifically, perform spatio-temporal alignment on the object detection bounding boxes and re-identification features in the real-time environment perception results to obtain feature-enhanced environment perception data. Synchronize the feature-enhanced environment perception data with the original sensor data from the inertial measurement unit, encoder, lidar, and depth camera to obtain a multi-source heterogeneous sensor dataset, which includes visual information of the robot's interaction with the environment, as well as other important sensor measurement data such as pose and distance. Analyze the noise characteristics of various types of sensor data in the multi-source heterogeneous sensor dataset. For example, the angular velocity data measured by the IMU is usually affected by drift noise, while the accuracy of the distance measured by the lidar is affected by environmental light and surface reflection characteristics. Model the noise of each type of sensor data to establish a sensor measurement noise covariance matrix and a process noise covariance matrix, which are used to describe the uncertainty of the sensor measurement values and are used to weigh the trust levels of different sensors during the Kalman filtering process. Construct a non-linear state transition equation and an observation equation. The state transition equation is used to describe the dynamic changes of the robot's state, while the observation equation is used to describe the relationship between the measurement values of each sensor and the state. In this method, the state vector includes information such as the robot's position, pose, velocity, and angular velocity, and the observation vector includes the actual measurement values of each sensor. In the extended Kalman filter framework, since the state transition equation and the observation equation are non-linear, linearize these equations. Perform a first-order Taylor expansion on the non-linear state transition equation to obtain the Jacobian matrix of the state prediction equation; similarly, perform a first-order Taylor expansion on the non-linear observation equation to obtain the Jacobian matrix of the observation prediction equation. The Jacobian matrix is used to describe the linearization characteristics of the system near the current state, so as to apply the theoretical framework of the Kalman filter in a non-linear environment. Based on the state estimate value at the previous moment and the current control input, use the Jacobian matrix of the state prediction equation to predict the state of the robot, obtaining a prior state estimate value and a prior error covariance matrix. The prior state estimate value is a prediction of the robot's current state, and the prior error covariance matrix describes the credibility of this prediction. Substitute the prior state estimate value into the observation prediction equation to calculate the observation prediction value. Subtract the observation prediction value from the actual observation value to obtain the observation residual. The observation residual reflects the difference between the actual observation value and the predicted value, and this difference is usually caused by the incomplete accuracy of the system model and sensor noise, so it is used to further correct the state estimate. To improve the accuracy of the state estimate, calculate the Kalman gain based on the prior error covariance matrix, the Jacobian matrix of the observation prediction equation, and the observation noise covariance matrix. The calculation of the Kalman gain is one of the core steps of the extended Kalman filter, which is used to weigh the trust levels between the prior prediction and the actual observation. If the observation noise is small, the Kalman gain will rely more on the observation value; if the prior error of the prediction is small, the Kalman gain will trust the prior prediction more.By using the Kalman gain, the prior state estimate value is corrected to obtain the robot-environment interaction state characteristics.

[0037] Step S5: According to the robot-environment interaction state characteristics, perform straight-line segment and turning segment trajectory analysis to construct a smooth navigation trajectory;

[0038] Specifically, analyze the robot-environment interaction state characteristics, extract the current position, attitude, speed, and acceleration of the robot, and use this information as the constraint conditions for constructing the starting point of the trajectory. At the same time, determine the constraint conditions for the end point of the trajectory based on the initial navigation path to obtain a complete set of trajectory boundary conditions. The set of trajectory boundary conditions includes the state information of the starting and ending points of the trajectory, including position, speed, and acceleration, and these constraint conditions are used to ensure that the trajectory can be smoothly connected and meet the requirements of navigation. Based on the set of trajectory boundary conditions, segment the initial navigation path and divide it into straight segments and turning segments to obtain a path segment set. The path segment set is used to describe the types of each section that the robot needs to pass through during the movement process, and different types of path segments correspond to different trajectory construction methods. For the straight segments in the path segment set, use the fifth-order polynomial interpolation method to construct the trajectory equation. The fifth-order polynomial interpolation method can ensure the smoothness of the trajectory while meeting the constraint conditions of position, speed, and acceleration. To simplify the calculation, normalize the time to the interval [0, 1], and use the position, speed, and acceleration information of the starting and ending points of the trajectory to solve the coefficients of the polynomial to obtain the trajectory equation of the straight segment. For the turning segments in the path segment set, use the third-order Bezier curve to construct the trajectory equation. The characteristic of the third-order Bezier curve is that it can flexibly adjust the shape of the curve by setting control points, which is suitable for describing turning trajectories. When constructing the turning segment trajectory, set the four control points as the starting point of the turning segment, the point on the tangent direction of the starting point, the point on the tangent direction of the ending point, and the ending point respectively. By adjusting the positions of the middle two control points, flexibly control the curvature of the turning segment to ensure that the robot can pass through smoothly during the turning process without sharp direction changes. The obtained turning segment trajectory equation can better adapt to the dynamic characteristics of the robot while meeting the requirements of smoothness. Combine the trajectory equations of the straight segments and turning segments to construct a complete trajectory equation. To make the trajectory adapt to the motion state of the robot at different time points, parameterize the time of the complete trajectory equation to obtain a parameterized trajectory equation. Time parameterization enables the trajectory to be adjusted according to the actual speed of the robot, ensuring that the trajectory planning is not only reasonable in space but also meets the time constraints of the robot. Perform differential operations on the parameterized trajectory equation to obtain the expressions of the speed, acceleration, and jerk of the trajectory. Through these expressions, analyze the dynamic characteristics of the trajectory to ensure that the changes in speed and acceleration of the trajectory conform to the physical characteristics of the robot. Combine the dynamic constraints of the robot to verify the feasibility of the trajectory to check whether the trajectory meets the dynamic performance of the robot. If the trajectory does not meet the constraint conditions in some aspects, such as too high speed or too large acceleration, adjust the trajectory parameters and regenerate the trajectory to ensure that the finally generated trajectory is feasible and safe. After generating a parameterized trajectory equation that meets the constraints, use the method of sampling at equal time intervals to generate a discrete sequence of trajectory points from the parameterized trajectory equation.Each trajectory point contains the position, attitude, velocity, and acceleration information of the robot, which is used to control the state of the robot at each time point. After generating the sequence of trajectory points, perform smoothness and continuity checks on the trajectory points to ensure that the changes in position, velocity, and acceleration between adjacent trajectory points meet the smoothness requirements. If the changes between adjacent trajectory points are too drastic, it will cause unstable movement during the robot's travel. Adjust these points to ensure the smoothness and continuity of the trajectory and obtain a smooth navigation trajectory.

[0039] Step S6: Continuously update the environmental feature data and the working state data of the mobility robot during navigation to construct a dynamic environmental map; when receiving a return signal, quickly plan an automatic return path based on the dynamic environmental map.

[0040] Specifically, during the robot navigation process, image data of the motion environment is collected in real time, and the image data is subjected to multi-scale convolution processing and feature fusion to obtain a motion multi-scale feature map. The motion multi-scale feature map is input into the global-local contrast learning module for processing, and more refined and differentiated features are extracted through contrast learning between the global and local parts to obtain the latest environmental feature data. The environmental feature data is continuously updated, which can dynamically reflect the changes in the environment where the robot is located. The latest environmental feature data is subjected to spatio-temporal registration to construct a short-term environmental feature sequence matrix. The spatio-temporal registration process arranges the environmental feature data in chronological order to form a short-term feature sequence and capture the temporal features of environmental changes. Through spatio-temporal correlation feature analysis of the short-term environmental feature sequence matrix, a dynamic environmental feature representation vector is extracted, which reflects the dynamic changes and various important spatio-temporal information in the current environment. At the same time, based on the odometer data and IMU data of the mobility robot, real-time pose estimation of the robot is performed to obtain the position information of the robot. Pose estimation combines the data of the odometer and the inertial measurement unit, which can provide accurate position information for the robot. The dynamic environmental feature representation vector is associated with the robot position information to obtain a local dynamic environmental map. The local dynamic environmental map is a refined description of the environment where the robot is currently located, including the dynamic features of the robot's current position and its surrounding environment. The local dynamic environmental map is subjected to spatial registration and fused with the global environmental map to update the global dynamic environmental map. The global dynamic environmental map is a dynamic description of the entire navigation area. As the local dynamic environmental map is continuously updated, the global map can dynamically reflect the changes in the environment. Data association and tracking of dynamic obstacles in the real-time motion environment are performed to obtain a set of motion trajectories of the obstacles. When the system receives the return signal, the current position coordinates of the mobility robot are extracted as the starting point of the return path, and at the same time, the coordinates of the navigation starting position are used as the ending point of the return path to construct a pair of return mission target points. The pair of return mission target points and the global dynamic environmental map are input into the A* algorithm for path search to obtain an initial return path composed of grid coordinates. The A* algorithm is a heuristic path search algorithm that can find a feasible path from the starting point to the ending point on a known map, ensuring the optimality and feasibility of the path. In order to make the initial return path smoother, it is fitted with a cubic B-spline curve, and the control point interval is set to 0.5 meters to generate a smooth path curve. The application of the cubic B-spline curve enables the path to maintain a smooth curvature change between each control point, avoiding sharp turns or sudden direction changes in the path. The system uniformly samples the fitted path every 0.1 meters to obtain a path point sequence containing multiple path points, and each path point contains information such as position, attitude, speed, and acceleration. At the same time, the path point sequence, the dynamic environmental feature representation vector, and the set of obstacle motion trajectories are combined together to construct a return scenario description tensor that comprehensively describes the return scenario.The return scenario description tensor contains all the environmental and path information required for the robot's current return mission. The return scenario description tensor is input into the adaptive uncertainty path planning model, which adopts a variational inference network structure and consists of an encoder and a decoder. The role of the encoder is to probabilistically encode the return scenario, generate the latent variable distribution of the scenario, and sample multiple sets of scenario samples from it. In this way, different environmental uncertainties are modeled to generate multiple possible scenarios. The multiple sets of scenario samples are input into the decoder, and the decoder generates multiple candidate return paths. For the generated candidate return paths, an optimization solution is performed to evaluate the feasibility, advantages, and disadvantages of each path in different environments, and finally an optimal automatic return path is obtained.

[0041] In the embodiments of the present invention, the multi-scale environmental feature extraction technology combined with global-local contrast learning can capture environmental information more comprehensively and accurately, improving the accuracy and robustness of environmental perception. The adaptive uncertainty path planning model effectively copes with various uncertainty factors in complex dynamic environments by dynamically constructing an uncertain parameter set, improving the adaptability and reliability of path planning. TensorRT-accelerated YOLOv5 object detection and OSNet feature re-identification technologies significantly improve the efficiency and accuracy of real-time environmental perception. The extended Kalman filter fuses multi-sensor data, effectively improving the accuracy of robot-environment interaction state estimation. The differential flatness trajectory generator combines polynomial interpolation and Bezier curves to generate smooth navigation trajectories that satisfy dynamic constraints, improving the comfort and safety of navigation. The continuous update of the dynamic environmental map and the fast return path planning technology enable the robot to respond to environmental changes in a timely manner and achieve safe and efficient automatic return.

[0042] In a specific embodiment, the process of executing step S1 may specifically include the following steps:

[0043] Perform Gaussian pyramid decomposition on the environmental image data to obtain a multi-scale image set, and perform multi-layer convolutional feature extraction on the multi-scale image set to obtain a multi-scale feature map set containing shallow, middle, and deep features;

[0044] Input the multi-scale feature map set into the feature pyramid network for upsampling and fusion of features at different scales to obtain a global environmental feature map, and input the global environmental feature map into the spatial attention module. By calculating the spatial correlation of the feature map, a key region weight map is obtained;

[0045] Weight the global environmental feature map based on the key region weight map to obtain a local environmental feature map. Use the global environmental feature map as the anchor sample, the local environmental feature map as the positive sample, and randomly select other samples within the batch as negative samples to construct a triplet sample set;

[0046] Calculate the InfoNCE loss for the triple sample set, and obtain the feature similarity score by minimizing the distance between the anchor sample and the positive sample and maximizing the distance between the anchor sample and the negative sample.

[0047] Based on the feature similarity score, adjust the weights of the global environmental feature map and the key region weight map to obtain the environmental feature data.

[0048] Specifically, perform multi-scale processing on the environmental image data to capture environmental features at different scales. Gaussian pyramid decomposition is a multi-scale image processing technique that obtains a series of images with gradually decreasing resolutions by recursively performing Gaussian smoothing and downsampling on the original image. Each layer of the image represents a different scale, and these images form a multi-scale image set. The process of the Gaussian pyramid is expressed as:

[0049] ;

[0050] Among them, represents the -th layer image of the pyramid, represents the Gaussian smoothing and downsampling operation, while represents the original image. In this way, a multi-scale environmental representation is obtained. Perform multi-layer convolutional feature extraction on the multi-scale image set, and use the convolutional layer in the convolutional neural network to extract features from each layer of the image. Through convolutional layers with different depths, shallow features, middle features, and deep features are respectively extracted to form a multi-scale feature map set. Shallow features correspond to edge and texture information in the image, while deep features can capture more abstract semantic information. Let represent the output features of the -th layer image in the -th convolutional layer, then the feature representation of each layer is obtained through the convolutional operation:

[0051] ;

[0052] Among them, represents the activation function (such as the ReLU function), represents the -th convolutional kernel, * represents the convolutional operation, Denotes the bias term. Through this step, feature maps of different levels are generated and integrated into a multi-scale feature map set, which reflects both the local details and the global structure of the image. The multi-scale feature map set is input into the Feature Pyramid Network for upsampling and fusion of features at different scales. The Feature Pyramid Network is a structure for object detection tasks. Through the upsampling operation, low-resolution, deep feature maps are fused with high-resolution, shallow feature maps to generate a global environmental feature map containing rich context information. The upsampling process lifts the low-resolution feature map to a higher resolution through interpolation, and at the same time, features at different scales are added and fused through skip connections to obtain the global environmental feature map. The global environmental feature map is input into the spatial attention module to identify the most critical regions in the environment by calculating the spatial correlation of the feature map. The spatial attention mechanism strengthens the significant regions in the feature map by learning weights, highlighting the regions that are important for the navigation task. The calculation of the spatial attention module is expressed as:

[0053] ;

[0054] where, and denote the query matrix and the key matrix obtained through convolutional transformation respectively. Softmax is a normalization operation used to map the correlation of features to the range from 0 to 1. denotes the obtained key region weight map. Through this weight map, the global environmental feature map is weighted to obtain the local environmental feature map. The global environmental feature map is used as the anchor sample, the local environmental feature map is used as the positive sample, and other samples are randomly selected from the current batch as negative samples to construct a triplet sample set. This sample construction method can effectively enhance the discriminative ability of the model, ensuring that the model can correctly distinguish the critical regions and non-critical regions in the environment. Calculate the InfoNCE loss for the triplet sample set. The InfoNCE loss is a loss function for contrastive learning, whose goal is to minimize the distance between the anchor sample and the positive sample, while maximizing the distance between the anchor sample and the negative sample, thereby improving the distinguishability of features. The loss function is expressed as:

[0055] ;

[0056] where, denotes the similarity calculation between features (such as cosine similarity). , , denote the feature representations of the anchor sample, the positive sample, and the th negative sample respectively. It is a temperature coefficient used to control the difficulty of contrastive learning. By minimizing the loss function, it ensures that the features of positive samples and anchor samples are closer, while the features of negative samples and anchor samples are farther apart, improving the reliability of feature representation. Based on the feature similarity score, the global environmental feature map and the key region weight map are weighted and adjusted, thereby optimizing the environmental feature data to obtain the environmental feature data.

[0057] In a specific embodiment, the process of executing step S2 may specifically include the following steps:

[0058] Based on the environmental feature data, perform spatial grid processing on the current driving environment, discretize the continuous space into grid cells of a fixed size to obtain a two-dimensional environmental state matrix, and align the time stamps of the two-dimensional environmental state matrix with the historical navigation data to construct a spatio-temporal sequence dataset containing environmental states and historical trajectories;

[0059] Perform sliding window segmentation on the spatio-temporal sequence dataset, extract multiple overlapping subsequence samples, perform wavelet transform on each subsequence sample to extract time-frequency domain features, and retain significant coefficients through threshold screening to obtain a dimensionality-reduced feature vector;

[0060] Generate a latent space representation and a reconstruction error based on the dimensionality-reduced feature vector, and based on the latent space representation and the reconstruction error, use the kernel density estimation algorithm to construct the uncertainty parameter distribution of the current driving environment;

[0061] Based on the uncertainty parameter distribution, perform Latin hypercube sampling to generate multiple groups of representative uncertainty parameter samples, and input the multiple groups of uncertainty parameter samples into the path planner based on the rapidly-exploring random tree respectively to generate multiple candidate paths;

[0062] Perform multi-objective evaluation and weighted summation on the multiple candidate paths, calculate the comprehensive score of each candidate path, and perform descending order sorting on the multiple candidate paths based on the comprehensive score, and select the candidate path with the highest comprehensive score as the initial navigation path.

[0063] Specifically, perform spatial grid processing on the current driving environment based on the environmental feature data. By dividing the continuous space into grid cells of a fixed size, the entire environment is discretized into a two-dimensional grid matrix. Assume the size of the grid cell is , through spatial discretization processing of the environmental features, a two-dimensional environmental state matrix is obtained, where each element represents in the grid The status information on it, such as whether there are obstacles, the difficulty of passage, etc. Align the two-dimensional environmental status matrix with the historical navigation data in terms of time stamps, associate the status of the current environment with the past navigation trajectories, and construct a spatio-temporal sequence dataset containing environmental status and historical trajectories. Through the alignment operation, capture the dynamic changes of the environment and the position information of the robot at different time points, forming a more comprehensive environmental description. represents a time series, where each time point has a corresponding environmental status matrix and the position trajectory information of the robot, constituting a spatio-temporal sequence dataset. Perform sliding window segmentation on the spatio-temporal sequence dataset to extract multiple overlapping subsequence samples. Sliding window segmentation can capture the short-term change characteristics of the environment. Each window contains the status information within a continuous period of time, forming subsequence samples. Let the sliding window size be , and the sliding step size be . Through segmentation, multiple overlapping subsequence samples are obtained, reflecting the characteristics of the environment and the trajectory at different time periods. For each subsequence sample, perform wavelet transform to extract its characteristics in the time-frequency domain. Wavelet transform is a time-frequency analysis tool that decomposes a signal into different frequency components, revealing the characteristics of the signal at different time scales. Let the subsequence sample be , and through wavelet transform, a wavelet coefficient matrix is obtained, where represents the scale parameter, represents the time translation parameter. The formula of wavelet transform is expressed as:

[0064] ;

[0065] where, is the mother wavelet function, and * represents the complex conjugate. The wavelet coefficient matrix obtained through wavelet transform contains the characteristic information of the subsequence at different frequencies and time points. To simplify the calculation and remove noise, retain the significant wavelet coefficients through threshold screening to obtain a dimensionality-reduced feature vector , which can effectively describe the change characteristics of the environment. Based on the dimensionality-reduced feature vector, generate a latent space representation and a reconstruction error. The latent space representation is a further compression and abstraction of the original features, which can reveal the latent structure of the feature data. At the same time, the reconstruction error is used to measure the difference between the original features and the reconstructed features. Based on the latent space representation and the reconstruction error, use the kernel density estimation algorithm to construct the uncertainty parameter distribution of the current driving environment. Kernel density estimation is a non-parametric statistical method used to estimate the probability distribution of data. Let the set of dimensionality-reduced feature vectors be , and the kernel density estimation is expressed as:

[0066] ;

[0067] Among them, is a kernel function (such as a Gaussian kernel), is the bandwidth parameter. Through kernel density estimation, the probability distribution of the uncertainty parameters in the environment is obtained. Based on the uncertainty parameter distribution, Latin hypercube sampling is performed to generate multiple groups of representative uncertainty parameter samples. Latin hypercube sampling is an effective sampling method that uniformly generates samples in a high-dimensional parameter space, ensuring sufficient exploration of environmental uncertainties. For the generated uncertainty parameter samples, each group of samples is respectively input into a path planner based on the rapidly-exploring random tree (RRT) to generate multiple candidate paths. RRT is a path planning algorithm based on random sampling that can quickly find a feasible path from the starting point to the target point in a complex environment. For the generated multiple candidate paths, multi-objective evaluation is performed, and the comprehensive score of each candidate path is calculated by the method of weighted summation. The objectives of multi-objective evaluation include path length, travel time, energy consumption, safety, etc., and each objective has different weights. Let the candidate path be and its comprehensive score is expressed as:

[0068] ;

[0069] Among them, represents the weight of the th objective, represents the score of the candidate path on the th objective. Through the method of weighted summation, different candidate paths are comprehensively evaluated, and the multiple candidate paths are sorted in descending order according to the comprehensive score, and the candidate path with the highest comprehensive score is selected as the initial navigation path.

[0070] In a specific embodiment, the process of executing step S3 may specifically include the following steps:

[0071] Collect real-time environmental images based on the initial navigation path and input the real-time environmental images into the YOLOv5 network accelerated by TensorRT, where the YOLOv5 network includes a backbone network, a neck network, and a detection head network;

[0072] The backbone network adopts the CSPDarknet53 structure, which contains 5 CSP modules. Each CSP module contains multiple residual units and a 1×1 convolutional layer, and the first multi-scale feature map is obtained through the backbone network;

[0073] The neck network adopts a feature pyramid structure, which contains 3 bottom-up feature fusion layers and 3 top-down feature fusion layers, and the second multi-scale feature map is obtained through the neck network;

[0074] The detection head network consists of 3 parallel detection branches, each branch contains 2 convolutional layers and 1 output layer. The object detection results, including bounding box coordinates, confidence, and class probabilities, are obtained through the detection head network.

[0075] The non-maximum suppression process is performed on the object detection results to obtain the filtered object detection boxes. The image regions corresponding to the filtered object detection boxes are cropped and resized, and then input into the TensorRT-accelerated OSNet network, where the OSNet network includes a lightweight backbone network and a feature aggregation module.

[0076] The lightweight backbone network adopts a depthwise separable convolution structure and consists of 4 stages. Each stage contains multiple lightweight residual units. The third multi-scale feature map is obtained through the lightweight backbone network.

[0077] The feature aggregation module adopts an adaptive feature fusion strategy and includes a channel attention mechanism and a spatial attention mechanism. The fused global feature vector is obtained through the feature aggregation module.

[0078] The L2 normalization process is performed on the fused global feature vector to obtain the object re-identification features. The object detection results and the object re-identification features are associated and matched to obtain the real-time environment perception results.

[0079] Specifically, ensure that the robot or autonomous vehicle can continuously collect real-time image data of the surrounding environment during driving. These image data are captured by a camera and used as the input for subsequent object detection and recognition. The TensorRT-accelerated YOLOv5 network is used to improve the inference speed and efficiency. The YOLOv5 network has a three-part structure: a backbone network, a neck network, and a detection head network. These modules cooperate with each other to efficiently complete the object detection task. The backbone network is the feature extraction foundation part of YOLOv5 and adopts the CSPDarknet53 structure. CSPDarknet53 contains 5 CSP modules, and each CSP module consists of multiple residual units and 1 convolutional layer. The residual unit effectively solves the gradient vanishing problem in the deep neural network, enabling the network to be stably trained in deeper situations. Through the residual unit and convolution operations, the backbone network extracts feature information at different levels and generates the first multi-scale feature map. Assume the input real-time image is , then it is represented by the backbone network as:

[0080] ;

[0081] where represents the convolution operation and residual structure in the backbone network, It is a multi-scale feature map extracted from the input image, containing features such as basic edges and textures in the image. The first multi-scale feature map is input into the neck network of YOLOv5. The neck network adopts a feature pyramid structure, including 3 bottom-up feature fusion layers and 3 top-down feature fusion layers. These feature fusion layers combine feature maps of different scales through upsampling and downsampling operations to achieve cross-level information integration and obtain the second multi-scale feature map . The second multi-scale feature map is input into the detection head network of YOLOv5. The detection head network contains 3 parallel detection branches, and each branch contains 2 convolutional layers and 1 output layer. Each branch is responsible for detecting objects at different scales, so as to detect objects of different sizes simultaneously. The detection head network finally outputs the object detection results, including the coordinates of the bounding box, confidence, and class probability. Let the object detection result be , where represents the bounding box coordinates of the th object, represents the class, represents the confidence of the object. To remove redundant and inaccurate detection results, non-maximum suppression is performed on the object detection results, and the box with the highest confidence is selected from the overlapping detection boxes, and other redundant boxes are removed to obtain the filtered object detection boxes . The image regions corresponding to the object detection boxes are cropped and resized for feature re-identification. To achieve feature re-identification of the object, the cropped image regions are input into the OSNet network accelerated by TensorRT. The OSNet network consists of a lightweight backbone network and a feature aggregation module. The lightweight backbone network adopts a depthwise separable convolution structure and contains 4 stages, and each stage consists of multiple lightweight residual units. Depthwise separable convolution can effectively reduce the computational complexity of the model while maintaining good feature extraction ability. Through the processing of the lightweight backbone network, the third multi-scale feature map is obtained. These feature maps contain the basic appearance features and structural information of the object. The feature aggregation module is used to process the third multi-scale feature map . The feature aggregation module adopts an adaptive feature fusion strategy, including a channel attention mechanism and a spatial attention mechanism. The role of the channel attention mechanism is to adaptively adjust the importance of each feature channel and highlight the channels that make significant contributions to object recognition; the spatial attention mechanism is used to emphasize the significant features in the spatial position of the feature map. In this way, a fused global feature vector is generated, which is expressed as:

[0082] ;

[0083] where Represents the combination of channel attention and spatial attention operations. The fused global feature vector contains significant feature information of the target and is used for the re-identification task of the target. To ensure the consistency of feature comparison between different targets, the fused global feature vector is subjected to L2 normalization. The normalized feature vector Satisfies:

[0084] ;

[0085] Wherein, Represents the L2 norm of the feature vector. Through normalization, the length of the feature vector is standardized, ensuring that the features extracted under different scales and lighting conditions have the same measurement standard, making the similarity comparison between different targets more fair and effective. The normalized re-identification features of the target are associated and matched with the target detection results to obtain the perception results of the real-time environment. The perception results include basic information such as the position and category of the target, as well as the feature description of the target, which are used to identify and distinguish different targets.

[0086] In a specific embodiment, the process of executing step S4 may specifically include the following steps:

[0087] Perform spatio-temporal alignment on the target detection box and re-identification features in the real-time environment perception results to obtain feature-enhanced environment perception data, and synchronize the feature-enhanced environment perception data with the original sensor data from the inertial measurement unit, encoder, lidar, and depth camera to obtain a multi-source heterogeneous sensor dataset;

[0088] Analyze the noise characteristics of various sensor data in the multi-source heterogeneous sensor dataset to establish a sensor measurement noise covariance matrix and a process noise covariance matrix;

[0089] Construct a non-linear state transition equation and an observation equation, where the state vector includes the robot's position, attitude, velocity, and angular velocity, and the observation vector includes the measurement values of each sensor;

[0090] Perform a first-order Taylor expansion on the non-linear state transition equation to obtain the Jacobian matrix of the state prediction equation, and perform a first-order Taylor expansion on the non-linear observation equation to obtain the Jacobian matrix of the observation prediction equation;

[0091] Based on the state estimate value and control input of the previous moment, use the Jacobian matrix of the state prediction equation to perform state prediction to obtain the prior state estimate value and the prior error covariance matrix;

[0092] Substitute the prior state estimate value into the observation prediction equation, calculate the observation prediction value, and subtract it from the actual observation value to obtain the observation residual;

[0093] The Kalman gain is calculated according to the prior error covariance matrix, the Jacobian matrix of the observation prediction equation and the observation noise covariance matrix, and the Kalman gain is used to correct the prior state estimate to obtain the robot-environment interaction state characteristics.

[0094] Specifically, the visual information of environmental perception is sorted and aligned. The real-time environmental perception results include the position of the target detection box and the feature information obtained by re-identification, which are from different time points. These data are aligned in time and space to ensure that all information can jointly describe the environmental state at the same time point. The time and space alignment unifies the data collected at different times to the same moment through interpolation and matching to obtain feature-enhanced environmental perception data. The feature-enhanced environmental perception data is synchronized with the original sensor data from other sensors (such as inertial measurement unit (IMU), encoder, lidar and depth camera). The time synchronization process is achieved through timestamp matching, which unifies all data on the same time axis to obtain a multi-source heterogeneous sensor data set. The noise characteristics of various sensor data in the multi-source heterogeneous sensor data set are analyzed to establish the sensor measurement noise covariance matrix and process noise covariance matrix. Different types of sensors are affected by noise from different sources during the measurement process. For example, the acceleration data of the IMU is affected by drift noise, while the lidar is affected by ambient light and the reflection characteristics of the object surface when measuring distance. In order to accurately describe the statistical characteristics of these noises, the noise of each sensor is modeled and the measurement noise covariance matrix is ​​obtained. and the process noise covariance matrix Each element of the covariance matrix describes the degree of mutual dependence between the measurements, which can be used to weigh the trust of different sensors in the filtering process. Construct nonlinear state transfer equations and observation equations to describe the change of the robot state over time and the measurement process of each sensor. State vector Contains information such as the robot's position, posture, speed, and angular velocity, and the observation vector Including the actual measurement values ​​of each sensor. The state transfer equation is expressed as:

[0095] ;

[0096] in, Indicates at time status, represents the state transfer function, which describes the control input Under the action, at all times Status How to transfer to Moment . represents process noise, describing the random disturbance in state transition. The observation equation is expressed as:

[0097] ;

[0098] wherein, represents the observed value at time . represents the observation function, which describes how the state is measured by various sensors. represents the measurement noise. Since both the state transition equation and the observation equation are nonlinear, these equations are linearized so as to use the extended Kalman filter for state estimation. The first-order Taylor expansion is performed on the nonlinear state transition equation to obtain the Jacobian matrix of the state prediction equation. At the same time, the first-order Taylor expansion is performed on the nonlinear observation equation to obtain the Jacobian matrix of the observation prediction equation. The Jacobian matrix is used to describe the linearized characteristics of the system near the current state, so as to apply the theoretical framework of the Kalman filter in a nonlinear environment. Based on the state estimate value at the previous moment and the control input , the Jacobian matrix of the state prediction equation is used to perform state prediction on the robot, and the prior state estimate value and the prior error covariance matrix

[0099] ;

[0100] ;

[0101] wherein, represents the error covariance matrix at the previous moment, represents the process noise covariance matrix. Substitute the prior state estimate value into the observation prediction equation to calculate the observation prediction value , and subtract it from the actual observed value to obtain the observation residual

[0102] ;

[0103] The observation residual reflects the difference between the actual observed value and the predicted value. This difference is caused by the incomplete accuracy of the system model and sensor noise, and is used to further correct the state estimate. In order to improve the accuracy of the state estimate, based on the prior error covariance matrix , the Jacobian matrix of the observation prediction equation, and the observation noise covariance matrix , calculate the Kalman gain

[0104] ;

[0105] The role of the Kalman gain is to weigh the degree of trust between the prior prediction and the actual observation. If the observation noise is small, the Kalman gain will rely more on the observed value; if the prior error of the prediction is small, the Kalman gain will trust the prior prediction more. The prior state estimate value is corrected using the Kalman gain to obtain the final robot-environment interaction state feature:

[0106] ;

[0107] In a specific embodiment, the process of executing step S5 may specifically include the following steps:

[0108] Analyze the robot-environment interaction state feature, extract the robot's current position, attitude, speed, and acceleration as the trajectory starting point constraint conditions, and determine the trajectory ending point constraint conditions according to the initial navigation path to obtain the trajectory boundary condition set;

[0109] Based on the trajectory boundary condition set, segment the initial navigation path, divide the initial navigation path into straight line segments and turning segments to obtain the path segment set;

[0110] For the straight line segments in the path segment set, use the fifth-order polynomial interpolation method to construct the trajectory equation, normalize the time to the interval [0, 1], and use the position, speed, and acceleration constraints of the trajectory starting point and ending point to solve the polynomial coefficients to obtain the straight line segment trajectory equation;

[0111] For the turning segments in the path segment set, use the third-order Bezier curve to construct the trajectory equation, where the four control points are respectively set as the starting point of the turning segment, the point on the tangent direction of the starting point, the point on the tangent direction of the ending point, and the ending point, and control the curvature by adjusting the positions of the middle two control points to obtain the turning segment trajectory equation;

[0112] Combine the straight line segment trajectory equation and the turning segment trajectory equation to construct a complete trajectory equation, and perform time parameterization on the complete trajectory equation to obtain the parameterized trajectory equation;

[0113] Perform differential operations on the parameterized trajectory equation to obtain the expressions of speed, acceleration, and jerk, and combine the robot dynamics constraints to verify the feasibility of the trajectory. If the constraint conditions are not met, adjust the trajectory parameters and regenerate;

[0114] Based on the parameterized trajectory equation, use the equal time interval sampling method to generate a discrete trajectory point sequence. Each trajectory point contains position, attitude, speed, and acceleration information, and check the smoothness and continuity of the trajectory point sequence to ensure that the changes in position, speed, and acceleration between adjacent trajectory points meet the smoothness requirements to obtain the smooth navigation trajectory.

[0115] Specifically, analyze the current environmental interaction state characteristics of the robot to extract the current position of the robot , attitude , speed and acceleration as the constraint conditions for the starting point of the trajectory. At the same time, according to the information such as the end position, attitude, speed, etc. of the initial navigation path, determine the end constraint conditions of the trajectory, including the end position , end attitude , end speed and acceleration . Through these constraint conditions, define a set of trajectory boundary conditions for subsequent trajectory planning. Segment the initial navigation path based on the set of trajectory boundary conditions. The initial navigation path consists of multiple consecutive points. Divide the path into straight segments and turning segments to obtain a set of path segments. Adopt different trajectory construction methods for different types of road segments to improve the accuracy and adaptability of trajectory planning. For the straight segments in the set of path segments, use the fifth-order polynomial interpolation method to construct the trajectory equation. The fifth-order polynomial interpolation has sufficient degrees of freedom to satisfy the boundary constraints of position, speed, and acceleration while ensuring the smoothness of the trajectory. Assume that time is normalized to the interval [0, 1], and the fifth-order polynomial trajectory equation is expressed as:

[0116] ;

[0117] ;

[0118] where and are the polynomial coefficients to be solved. By using the constraint conditions such as the position, speed, and acceleration of the starting point and end point of the trajectory, solve these coefficients to obtain the fifth-order polynomial equation for describing the straight segment trajectory. For the turning segments in the set of path segments, use the third-order Bezier curve to construct the trajectory equation. The characteristic of the Bezier curve is that it adjusts the shape of the curve through control points, which is suitable for describing the trajectory of the robot during the turning process. The third-order Bezier curve trajectory of the turning segment is expressed as:

[0119] ;

[0120] where are four control points respectively, is the time parameter. The control points and are the starting point and end point of the turning segment respectively, and the control points and are respectively located on the tangent directions of the starting point and end point. By adjusting and At the position, control the curvature of the curve so that the trajectory of the turning section is smoother and meets the dynamic requirements. Combine the trajectory equations of the straight section and the turning section to construct a complete trajectory equation. To adapt the trajectory to the motion state of the robot at different time points, perform time parameterization on the complete trajectory equation to obtain a parameterized trajectory equation. The purpose of time parameterization is to associate the trajectory with time, enabling the system to execute the trajectory according to the actual motion time and ensuring the smoothness and dynamic consistency of the trajectory. Perform differential operations on the parameterized trajectory equation to obtain expressions for velocity and acceleration. For the trajectory equations and , the velocity and acceleration are respectively expressed as:

[0121] ;

[0122] ;

[0123] Through these expressions, analyze the velocity and acceleration of the trajectory at each time point. Combine the dynamic constraints of the robot, such as maximum velocity, maximum acceleration, and maximum steering angle, etc., to verify the feasibility of the trajectory. If the trajectory does not meet these dynamic constraints, for example, the acceleration of a certain section exceeds the maximum acceleration that the robot can withstand, then adjust the parameters of the trajectory and regenerate the trajectory until the trajectory meets all the constraint conditions. After generating a parameterized trajectory equation that meets the constraint conditions, use the method of sampling at equal time intervals to generate a discrete sequence of trajectory points from the parameterized trajectory equation. Each trajectory point contains information such as the position, attitude, velocity, and acceleration of the robot. For the time interval , the sequence of trajectory points is expressed as . The generated sequence of trajectory points is subjected to smoothness and continuity checks to ensure that the changes in position, velocity, and acceleration between adjacent trajectory points are continuous and smooth. Through the smoothness check, avoid unstable situations when the robot executes the trajectory, such as sudden changes in velocity or direction, and obtain a smooth navigation trajectory.

[0124] In a specific embodiment, the process of executing step S6 may specifically include the following steps:

[0125] Perform multi-scale convolution processing and feature fusion on the motion environment images collected in real time during navigation to obtain a motion multi-scale feature map, and input the motion multi-scale feature map into the global-local contrast learning module for processing to obtain the latest environmental feature data;

[0126] Perform spatio-temporal registration on the latest environmental feature data to construct a short-term environmental feature sequence matrix, and perform spatio-temporal correlation feature analysis on the short-term environmental feature sequence matrix to obtain a dynamic environmental feature representation vector;

[0127] Based on the odometer data and IMU data of the mobility robot, real-time pose estimation of the robot is carried out to obtain the robot's position information, and the dynamic environment feature representation vector is associated with the robot's position information to obtain a local dynamic environment map;

[0128] Spatial registration is performed on the local dynamic environment map to update the global dynamic environment map, and data association and tracking are carried out on the dynamic obstacles in the real-time motion environment to obtain a set of obstacle motion trajectories;

[0129] When a return signal is received, the current position coordinates of the mobility robot are extracted as the starting point, and the navigation starting position coordinates are used as the ending point to construct a pair of return mission target points. Then, the pair of return mission target points and the global dynamic environment map are input into the A* algorithm for path search to obtain an initial return path composed of grid coordinates;

[0130] The initial return path is fitted with a cubic B-spline curve, with the control point interval set to 0.5 meters, to generate a smooth path curve. Then, path points are uniformly sampled every 0.1 meter along the path to obtain a sequence of path points. The sequence of path points, the dynamic environment feature representation vector, and the set of obstacle motion trajectories are combined to construct a return scenario description tensor;

[0131] The return scenario description tensor is input into an adaptive uncertainty path planning model. The adaptive uncertainty path planning model adopts a variational inference network structure, which includes an encoder and a decoder. The encoder is used to probabilistically encode the return scenario to generate a latent variable distribution of the scenario, and multiple groups of scenario samples are sampled from it. The multiple groups of scenario samples are input into the decoder to generate multiple candidate return paths, and the multiple candidate return paths are optimized to obtain an automatic return path.

[0132] Specifically, real-time image data of the surrounding environment is collected using an on-vehicle camera or a camera on the mobility robot. The real-time image is input into a multi-scale convolutional network for processing. The purpose of multi-scale convolutional processing is to extract features from different scales so as to capture both detailed information and global semantic information. Suppose the input environmental image is Through processing by the multi-scale convolutional network, a motion multi-scale feature map is obtained 。The motion multi-scale feature map is input into the global-local contrast learning module for processing, and the discriminative ability of the environmental feature data is enhanced through contrast learning. In the contrast learning module, a contrast is made between the global features and the local features, and those features that can effectively distinguish different environments are learned through the contrast loss to obtain the latest environmental feature data. The contrast learning process is described by the InfoNCE loss function, and the goal of this loss function is to minimize the distance between positive samples and maximize the distance between negative samples. Through contrast learning, more robust environmental features are extracted. The latest environmental feature data is subjected to spatio-temporal registration to construct a short-term environmental feature sequence matrix. The environmental feature data collected at different time points are aligned and organized to form an environmental feature sequence matrix containing multiple time segments. Let the environmental feature data be , where represents the number of time steps, and a feature matrix is obtained through spatio-temporal registration, where represents the dimension of the feature. Spatio-temporal correlation feature analysis is performed on this feature matrix to extract the dynamically changing patterns in the environment, and a dynamic environmental feature representation vector is obtained, which effectively describes the dynamic change characteristics of the current environment. At the same time, based on the odometer data and inertial measurement unit (IMU) data of the mobility robot, real-time pose estimation of the robot is performed to obtain the current position and pose of the robot. The odometer data provides the distance and direction information of the robot's travel, while the IMU data provides the measurements of the robot in terms of acceleration and angular velocity. By fusing these data, the pose of the robot at the current moment is estimated to obtain the position information of the robot, where represents the two-dimensional planar position of the robot, and represents the pose angle of the robot. The dynamic environmental feature representation vector is associated with the robot position information to obtain a local dynamic environmental map. The local dynamic environmental map is subjected to spatial registration and fused with the global environmental map to update the global dynamic environmental map. The global dynamic environmental map is used to represent the environmental state of the entire navigation area, and the change of the environment is reflected by continuously updating the global map. At the same time, data association and tracking are performed on the dynamic obstacles in the real-time motion environment to obtain a set of motion trajectories of the obstacles, where represents the motion trajectory of the th obstacle, and these trajectory data are used for obstacle avoidance operations in subsequent path planning. When the system receives the return signal, the current position coordinates of the mobility robot are extracted as the starting point of the return path, and at the same time, the coordinates of the navigation starting position are used as the end point to construct a pair of return mission target points. The pair of return mission target points and the global dynamic environmental map are input into the A* algorithm for path search. The A* algorithm is a classic heuristic path search algorithm that can find the optimal path from the starting point to the ending point on a known map. Through the A* algorithm, an initial return path composed of grid coordinates is generated. To make the initial return path smoother, it is fitted with a cubic B-spline curve. The cubic B-spline curve constructs a smooth curve by setting control points, and the interval of the control points is set to 0.5 meters, so as to ensure that the generated path curve is smooth enough in space and can meet the driving requirements of the robot. The representation form of the spline curve is:

[0133] ;

[0134] where, represents the spline curve, represents the control point, is the cubic B-spline basis function, is the parameter of the curve. By controlling the positions of these control points, the shape of the spline curve is adjusted so that the path has smooth speed and acceleration characteristics at the starting point and the ending point. Along the fitted path, the system samples uniformly every 0.1 meter to obtain a sequence of path points , where each path point contains position, attitude, speed, and acceleration information. At the same time, the sequence of path points, the dynamic environment feature representation vector and the set of obstacle motion trajectories are combined together to construct a return scenario description tensor , which contains all the environmental and dynamic information that needs to be considered during the return process. The return scenario description tensor is input into the adaptive uncertainty path planning model. This model adopts a variational inference network structure, which includes an encoder and a decoder. The role of the encoder is to perform probabilistic encoding on the return scenario and generate the latent variable distribution of the scenario. Let the return scenario description be , and through the encoder , the latent variable distribution is obtained:

[0135] ;

[0136] where, represents the latent variable, and represent the mean and covariance of the latent variable respectively, which are determined by the encoder parameters Decision. By sampling from this latent variable distribution, multiple sets of scenario samples are obtained to model the uncertainties in the environment. The multiple sets of scenario samples are input into the decoder, and the decoder generates multiple candidate return paths. The role of the decoder is to map the latent variables back to specific path representations, and the generated candidate return paths can cover different possible environmental changes and obstacle movements. For the multiple generated candidate return paths, an optimization solution is performed to find an optimal return path. The goal of the optimization solution is to achieve the best balance in terms of safety, travel time, energy consumption, etc. for the return path.

[0137] The above described the method for automatic navigation and return of a robot in the embodiments of the present invention. Next, the device for automatic navigation and return of a robot in the embodiments of the present invention will be described. Please refer to Figure 2 One embodiment of the device for automatic navigation and return of a robot in the embodiments of the present invention includes:

[0138] An extraction module, configured to perform multi-scale environmental feature extraction on the environmental image data collected by the mobility robot to obtain environmental feature data;

[0139] A construction module, configured to construct an initial navigation path according to the environmental feature data and historical navigation data;

[0140] An identification module, configured to perform target detection and feature re-identification on the real-time environmental image based on the initial navigation path to obtain a real-time environmental perception result;

[0141] A processing module, configured to perform extended Kalman filter fusion processing on the multi-sensor data according to the real-time environmental perception result to obtain the robot-environment interaction state features;

[0142] An analysis module, configured to perform straight-line segment and turning segment trajectory analysis according to the robot-environment interaction state features to construct a smooth navigation trajectory;

[0143] A planning module, configured to continuously update the environmental feature data and the working state data of the mobility robot during navigation to construct a dynamic environmental map; when receiving a return signal, quickly plan an automatic return path based on the dynamic environmental map.

[0144] Through the collaborative cooperation of the above-mentioned various components, by adopting the multi-scale environmental feature extraction technology combined with global-local contrast learning, it is possible to capture environmental information more comprehensively and accurately, improving the accuracy and robustness of environmental perception. The adaptive uncertainty path planning model effectively copes with various uncertainty factors in complex dynamic environments by dynamically constructing an uncertain parameter set, improving the adaptability and reliability of path planning. The TensorRT-accelerated YOLOv5 object detection and OSNet feature re-identification technologies significantly enhance the efficiency and accuracy of real-time environmental perception. The extended Kalman filter fuses and processes multi-sensor data, effectively improving the accuracy of robot-environment interaction state estimation. The differential flatness trajectory generator combines polynomial interpolation and Bezier curves to generate smooth navigation trajectories that meet dynamic constraints, improving the comfort and safety of navigation. The continuous update of the dynamic environmental map and the fast return path planning technology enable the robot to respond to environmental changes in a timely manner and achieve safe and efficient automatic return.

[0145] Referring to Figure 3 , an embodiment of the present invention also provides a computer device, which may be a server, and its internal structure may be as Figure 3 shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected through a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program, when executed by the processor, implements the above method.

[0146] Those skilled in the art can understand that Figure 3 the structure shown in

[0147] is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0148] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium provided in the present invention and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or an external cache. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0149] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, systems, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0150] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs, etc., which can store program codes.

[0151] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A robot automatic navigation and return method, characterized in that: The method comprises: Perform multi-scale environmental feature extraction on the environmental image data collected by the walking robot to obtain environmental feature data; Constructing an initial navigation path according to the environmental feature data and the historical navigation data; Based on the initial navigation path, target detection and feature re-identification are performed on the real-time environment image to obtain a real-time environment perception result; According to the real-time environment perception results, the multi-sensor data is subjected to extended Kalman filter fusion processing to obtain the robot-environment interaction state characteristics; Performing straight line segment and turning segment trajectory analysis according to the robot-environment interaction state characteristics to construct a smooth navigation trajectory; During the navigation process, the environmental feature data and the working status data of the walking robot are continuously updated to construct a dynamic environmental map; when a return signal is received, an automatic return path is quickly planned based on the dynamic environmental map; the process of constructing the dynamic environmental map includes: performing multi-scale convolution processing and feature fusion on the motion environment image collected in real time during the navigation process to obtain a motion multi-scale feature map, and inputting the motion multi-scale feature map into a global-local contrast learning module for processing to obtain the latest environmental feature data; performing spatiotemporal registration on the latest environmental feature data to construct a short-term environmental feature sequence matrix, and performing spatiotemporal correlation feature analysis on the short-term environmental feature sequence matrix to obtain a dynamic environmental feature representation vector; based on the odometer data and IMU data of the walking robot, the robot real-time posture estimation is performed to obtain the robot position information, and the dynamic environmental feature representation vector is associated with the robot position information to obtain a local dynamic environmental map; the local dynamic environmental map is spatially registered to update the global dynamic environmental map, and data association and tracking are performed on dynamic obstacles in the real-time motion environment to obtain an obstacle motion trajectory set.

2. The robot automatic navigation and return method according to claim 1, characterized in that: The multi-scale environmental feature extraction is performed on the environmental image data collected by the walking robot to obtain environmental feature data, including: Performing Gaussian pyramid decomposition on the environmental image data to obtain a multi-scale image set, and performing multi-layer convolution feature extraction on the multi-scale image set to obtain a multi-scale feature atlas containing shallow, middle and deep features; Input the multi-scale feature atlas set into the feature pyramid network to perform upsampling and fusion of features of different scales to obtain a global environment feature map, and input the global environment feature map into the spatial attention module to obtain a key area weight map by calculating the spatial correlation of the feature map; The global environment feature map is weighted based on the key area weight map to obtain a local environment feature map, and the global environment feature map is used as an anchor sample, the local environment feature map is used as a positive sample, and other samples in the batch are randomly selected as negative samples to construct a triplet sample set; Calculate the InfoNCE loss for the triple sample set, and obtain the feature similarity score by minimizing the distance between the anchor sample and the positive sample and maximizing the distance between the anchor sample and the negative sample; The global environment feature map and the key area weight map are weighted based on the feature similarity score to obtain environment feature data.

3. The robot automatic navigation and return method according to claim 2, characterized in that: The constructing the initial navigation path according to the environmental feature data and the historical navigation data comprises: Based on the environmental feature data, the current driving environment is spatially gridded, the continuous space is discretized into grid units of fixed size, a two-dimensional environmental state matrix is ​​obtained, and the two-dimensional environmental state matrix is ​​time-stamp aligned with the historical navigation data to construct a spatiotemporal sequence data set including environmental states and historical trajectories; Sliding window segmentation is performed on the spatiotemporal sequence data set to extract multiple overlapping subsequence samples, wavelet transform is performed on each subsequence sample to extract time-frequency domain features, and significant coefficients are retained through threshold screening to obtain a reduced-dimensional feature vector; Generating a latent space representation and a reconstruction error according to the dimension reduction feature vector, and constructing an uncertainty parameter distribution of the current driving environment using a kernel density estimation algorithm based on the latent space representation and the reconstruction error; Performing Latin hypercube sampling based on the uncertainty parameter distribution to generate multiple groups of representative uncertainty parameter samples, and inputting the multiple groups of uncertainty parameter samples into a path planner based on a rapidly expanding random tree to generate multiple candidate paths; The multiple candidate paths are evaluated by multiple objectives and weighted summation is performed to calculate a comprehensive score for each candidate path, and the multiple candidate paths are sorted in descending order based on the comprehensive score, and the candidate path with the highest comprehensive score is selected as the initial navigation path.

4. The robot automatic navigation and return method according to claim 3, characterized in that: The performing target detection and feature re-identification on the real-time environment image based on the initial navigation path to obtain a real-time environment perception result includes: Based on the initial navigation path, a real-time environment image is collected, and the real-time environment image is input into a YOLOv5 network accelerated by TensorRT, wherein the YOLOv5 network includes a backbone network, a neck network, and a detection head network; The backbone network adopts the CSPDarknet53 structure, including 5 CSP modules, each CSP module includes multiple residual units and 1×1 convolutional layers, and the first multi-scale feature map is obtained through the backbone network; The neck network adopts a feature pyramid structure, including 3 bottom-up feature fusion layers and 3 top-down feature fusion layers, and obtains a second multi-scale feature map through the neck network; The detection head network includes three parallel detection branches, each branch includes two convolutional layers and one output layer, and the target detection results, including bounding box coordinates, confidence and category probability, are obtained through the detection head network; Performing non-maximum suppression processing on the target detection result to obtain a filtered target detection frame, cropping and resizing the image area corresponding to the filtered target detection frame, and inputting it into the OSNet network accelerated by TensorRT, wherein the OSNet network includes a lightweight backbone network and a feature aggregation module; The lightweight backbone network adopts a deep separable convolution structure, including 4 stages, each stage including multiple lightweight residual units, and a third multi-scale feature map is obtained through the lightweight backbone network; The feature aggregation module adopts an adaptive feature fusion strategy, including a channel attention mechanism and a spatial attention mechanism, and obtains a fused global feature vector through the feature aggregation module; The fused global feature vector is subjected to L2 normalization processing to obtain a target re-identification feature, and the target detection result and the target re-identification feature are associated and matched to obtain a real-time environment perception result.

5. The robot automatic navigation and return method according to claim 4, characterized in that: According to the real-time environment perception result, the multi-sensor data is subjected to extended Kalman filter fusion processing to obtain the robot-environment interaction state characteristics, including: Performing spatiotemporal alignment on the target detection boxes and re-identification features in the real-time environment perception results to obtain feature-enhanced environment perception data, and performing time synchronization on the feature-enhanced environment perception data with the original sensor data from the inertial measurement unit, encoder, lidar, and depth camera to obtain a multi-source heterogeneous sensor data set; Performing noise characteristic analysis on various sensor data in the multi-source heterogeneous sensor data set, and establishing a sensor measurement noise covariance matrix and a process noise covariance matrix; Construct nonlinear state transfer equations and nonlinear observation equations, where the state vector includes the robot's position, posture, velocity and angular velocity, and the observation vector includes the measurement values ​​of each sensor; Performing a first-order Taylor expansion on the nonlinear state transfer equation to obtain a Jacobian matrix of the state prediction equation, and performing a first-order Taylor expansion on the nonlinear observation equation to obtain a Jacobian matrix of the observation prediction equation; Based on the state estimation value and control input at the previous moment, the Jacobian matrix of the state prediction equation is used to perform state prediction to obtain a priori state estimation value and a priori error covariance matrix; Substituting the prior state estimate into the observation prediction equation, calculating the observation prediction value, and subtracting it from the actual observation value to obtain the observation residual; The Kalman gain is calculated according to the prior error covariance matrix, the Jacobian matrix of the observation prediction equation and the observation noise covariance matrix, and the prior state estimation value is corrected by using the Kalman gain to obtain the robot-environment interaction state characteristics.

6. The robot automatic navigation and return method according to claim 5, characterized in that: The step of analyzing the trajectory of straight segments and turning segments according to the robot-environment interaction state characteristics to construct a smooth navigation trajectory includes: Analyze the robot-environment interaction state characteristics, extract the robot's current position, posture, speed and acceleration as the trajectory starting point constraint conditions, and determine the trajectory end point constraint conditions according to the initial navigation path to obtain a trajectory boundary condition set; Based on the trajectory boundary condition set, segmenting the initial navigation path, dividing the initial navigation path into straight segments and turning segments, and obtaining a path segment set; For the straight line segments in the path segment set, a trajectory equation is constructed using a quintic polynomial interpolation method, the time is normalized to the interval [0, 1], and the position, velocity, and acceleration constraints of the trajectory start and end points are used to solve the polynomial coefficients to obtain the straight line segment trajectory equation; For the turning segment in the path segment set, a third-order Bezier curve is used to construct a trajectory equation, wherein four control points are respectively set as the starting point of the turning segment, a point in the tangent direction of the starting point, a point in the tangent direction of the end point, and the end point, and the curvature is controlled by adjusting the positions of the two middle control points to obtain the turning segment trajectory equation; Combining the straight segment trajectory equation and the turning segment trajectory equation to construct a complete trajectory equation, and performing time parameterization on the complete trajectory equation to obtain a parameterized trajectory equation; Perform differential operation on the parameterized trajectory equation to obtain expressions of velocity, acceleration and acceleration, and verify the feasibility of the trajectory in combination with the robot dynamics constraints. If the constraints are not met, adjust the trajectory parameters and regenerate them; Based on the parameterized trajectory equation, an equal time interval sampling method is adopted to generate a discrete trajectory point sequence, each trajectory point contains position, attitude, velocity and acceleration information, and the trajectory point sequence is checked for smoothness and continuity to ensure that the position, velocity and acceleration changes between adjacent trajectory points meet the smoothness requirements, so as to obtain a smooth navigation trajectory.

7. The robot automatic navigation and return method according to claim 6, characterized in that: When the return signal is received, quickly planning an automatic return path based on the dynamic environment map includes: When a return signal is received, the current position coordinates of the walking robot are extracted as the starting point, the navigation start position coordinates are taken as the end point, a return mission target point pair is constructed, and the return mission target point pair and the global dynamic environment map are input into the A* algorithm for path search to obtain an initial return path composed of grid coordinates; The initial return path is fitted with a cubic B-spline curve, with the control point interval set to 0.5 meters, to generate a smooth path curve, and a path point sequence is obtained by uniformly sampling every 0.1 meters along the path, and the path point sequence, the dynamic environment feature representation vector and the obstacle motion trajectory set are combined to construct a return scene description tensor; The return scene description tensor is input into an adaptive uncertainty path planning model, which adopts a variational inference network structure and includes an encoder and a decoder. The encoder is used to probabilistically encode the return scene to generate a latent variable distribution of the scene, and multiple groups of scene samples are sampled therefrom. The multiple groups of scene samples are input into a decoder to generate multiple candidate return paths, and the multiple candidate return paths are optimized to obtain an automatic return path.

8. A robot automatic navigation and return device, characterized in that: Used to execute the robot automatic navigation and return method according to any one of claims 1 to 7, the robot automatic navigation and return device comprises: An extraction module is used to extract multi-scale environmental features from environmental image data collected by the walking robot to obtain environmental feature data; A construction module, used for constructing an initial navigation path according to the environmental feature data and the historical navigation data; A recognition module, used to perform target detection and feature re-recognition on the real-time environment image based on the initial navigation path to obtain a real-time environment perception result; A processing module, used to perform extended Kalman filter fusion processing on multi-sensor data according to the real-time environment perception result to obtain robot-environment interaction state characteristics; An analysis module, used for analyzing the trajectory of straight segments and turning segments according to the robot-environment interaction state characteristics, and constructing a smooth navigation trajectory; The planning module is used to continuously update the environmental feature data and the working status data of the walking robot during the navigation process to build a dynamic environmental map; when a return signal is received, the automatic return path is quickly planned based on the dynamic environmental map.

9. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and is characterized in that when the processor executes the computer program, the robot automatic navigation and return method described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor is enabled to execute the robot automatic navigation and return method according to any one of claims 1 to 7.

Images