High-precision positioning and mapping system of automatic driving carrying robot

Abstract

The invention discloses a high-precision positioning and mapping system for an automatic driving carrying robot, and the system comprises the steps: collecting environment data through a plurality of sensors, carrying out the real-time preprocessing through an edge computing device, and generating perception information and a preliminary environment model; based on the preliminary environment model, constructing a quantitative environment model, and improving the positioning precision of the robot by using a correction algorithm; analyzing environment change by using a deep learning and reinforcement learning algorithm, adaptively adjusting sensor parameters, and generating an optimized sensing result; a three-dimensional map is generated in real time based on the optimized sensing result and the quantitative model, and dynamic updating and error optimization are carried out according to positioning correction and environment change; based on a dynamic three-dimensional map and positioning information, a multi-level path planning structure is constructed, and path generation, evaluation and real-time optimization are realized; and a control instruction is executed according to the optimized path output by the path planning module, and a motion track is dynamically adjusted in combination with real-time sensing feedback, so that safe and efficient task execution is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of autonomous driving, specifically to a high-precision positioning and map construction system for autonomous delivery robots. BACKGROUND

[0002] Currently, autonomous driving technology has been widely applied in logistics transportation, unmanned delivery and other fields. In particular, in complex and dynamic environments, how to improve the positioning accuracy and map construction capability of autonomous delivery robots has become a key to efficient and safe operation. Most existing technologies rely on GPS signals for positioning, but in urban complex environments or underground areas where GPS signals are limited, the positioning accuracy often cannot meet the needs. In addition, traditional map construction methods are easily affected by sensor errors, making it difficult to balance real-time and accuracy. With the continuous development of sensor technology, edge computing and deep learning algorithms, new positioning and map construction methods provide more accurate and reliable solutions for autonomous delivery robots, but there is still room for optimization. SUMMARY

[0003] The present application provides a high-precision positioning and map construction method for autonomous delivery robots, which includes: S10, collecting environmental data through multiple sensors, using edge computing devices to pre-process the data in real time, and generating efficient perception information and preliminary environmental models; S20, constructing a quantitative environmental model based on the preliminary environmental model, and using a correction algorithm to improve the positioning accuracy of the robot, especially when GPS signals are limited; S30, using deep learning and reinforcement learning algorithms to analyze environmental changes and adaptively adjust sensor parameters to generate optimized perception results; S40, based on the optimized perception results and the quantitative model, generating a three-dimensional map in real time, and dynamically updating and optimizing errors according to the positioning correction and environmental changes to ensure the consistency and accuracy of the map in the time and space dimensions; S50, based on the dynamic three-dimensional map and positioning information, constructing a multi-level path planning structure to realize path generation, evaluation and real-time optimization; S60, executing control instructions according to the optimized path output by the path planning module, and dynamically adjusting the motion trajectory combined with real-time perception feedback to achieve safe and efficient task execution.

[0004] As described above, a high-precision positioning and map construction method for autonomous delivery robots, which collects environmental data through multiple sensors, uses edge computing devices to pre-process the data in real time, and generates efficient perception information and preliminary environmental models, including: S101, collect and synchronize data from multiple sensors such as vision, lidar, ultrasonic and inertial measurement unit, perform time and space alignment to form a unified perception dataset; S102, denoise, outlier rejection and feature extraction of the perception dataset by edge computing device, generate a preliminary environment model containing spatial features and semantic information, provide input for subsequent quantitative modeling.

[0005] The high-precision positioning and map construction method of the autonomous driving delivery robot as described above, wherein the quantitative environment model is constructed based on the preliminary environment model, and the robot positioning accuracy is improved by using the correction algorithm, especially when the GPS signal is limited, comprising: S201, quantitatively model the preliminary environment model by multi-sensor fusion algorithm, and convert obstacles and dynamic elements into calculable spatial feature units; S202, use the correction algorithm to correct the positioning error in real time and improve the positioning accuracy of the robot in complex environments.

[0006] The high-precision positioning and map construction method of the autonomous driving delivery robot as described above, wherein the deep learning and reinforcement learning algorithm is used to analyze environmental changes and adaptively adjust sensor parameters to generate optimized perception results, comprising: S301, analyze the perception data by a deep learning model to identify obstacles, pedestrians and environmental change elements; S302, based on the reinforcement learning algorithm, automatically adjust the working mode and parameters of the sensor according to real-time environmental feedback to optimize the stability of data acquisition and feature extraction.

[0007] The high-precision positioning and map construction method of the autonomous driving delivery robot as described above, wherein based on the optimized perception results and the quantitative model, a three-dimensional map is generated in real time, and dynamic updating and error optimization are performed according to the positioning correction and environmental changes, comprising: S401, fuse the quantitative model and real-time perception data, generate a three-dimensional environment map by sparse point cloud reconstruction algorithm, and establish topological node relationship to describe the spatial structure; S402, dynamically update the map node and boundary information according to the positioning correction feedback and environmental change data, perform error correction and time series smoothing to ensure high consistency between the map and the actual environment.

[0008] The high-precision positioning and map construction method of the autonomous driving delivery robot as described above, wherein based on the dynamic three-dimensional map and positioning information, a multi-level path planning structure is constructed to realize path generation, evaluation and real-time optimization, comprising: S501, construct an initial path graph based on node connectivity on the three-dimensional map, and generate a preliminary driving path by heuristic search algorithm; S502, combine the perception feedback with the risk factors, use the dynamic programming algorithm to adjust the path node weight in real time, and generate a multi-objective optimization path.

[0009] The high-precision positioning and map construction method of the automatic driving carrier robot as described above, wherein the control instructions are executed according to the optimized path output by the path planning module, and the motion trajectory is dynamically adjusted in combination with real-time perception feedback, comprising: S601, convert the planned path into executable motion control instructions, and monitor the pose deviation, obstacle change and power response parameters in the execution process; S602, when it is detected that the path is blocked or the deviation is out of limit, automatically trigger the local obstacle avoidance module, re-plan the local path based on the real-time map, and update the control instructions to maintain the task continuity.

[0010] The application also provides a high-precision positioning and map construction system of an automatic driving carrier robot, comprising: A perception data acquisition module for acquiring environmental data through multiple sensors and providing the data to an edge computing device; An edge computing preprocessing module for denoising, correcting, synchronizing and feature extracting the sensor data to generate a preliminary environmental model; An environmental modeling and positioning correction module for constructing a quantitative environmental model based on the preliminary environmental model and improving the positioning accuracy in real time through a correction algorithm; An intelligent perception and sensor adjustment module for analyzing environmental changes using deep learning and reinforcement learning algorithms and adaptively adjusting the working mode and parameters of the sensors; A three-dimensional map construction and correction module for generating a three-dimensional map based on the quantitative model and perception results, and dynamically updating and optimizing errors; A path planning and execution module including a path generation unit and a control execution unit for path planning, evaluation, optimization and dynamic execution control based on the three-dimensional map and positioning information, to ensure efficient and safe completion of the task.

[0011] The beneficial effects achieved by this invention are as follows: By combining multi-sensor data fusion, edge computing, and deep learning algorithms, this invention solves the problems of positioning accuracy and map building faced by traditional autonomous driving robots in complex environments. Through real-time processing of data from multiple sensors, including vision, LiDAR, and inertial measurement units, this invention can provide high-precision positioning correction and environmental modeling even when GPS signals are limited or obstructed. Simultaneously, the introduction of deep learning and reinforcement learning algorithms enables the robot to intelligently analyze environmental changes and adaptively adjust sensor parameters, further improving perception efficiency and accuracy. Furthermore, by dynamically updating the 3D map and optimizing path planning, this invention ensures that the robot can safely and efficiently perform tasks in dynamic and complex environments, significantly improving the overall performance and reliability of autonomous driving robots. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0013] Figure 1 This is a flowchart of a high-precision positioning and map building method for an autonomous driving transport robot provided in Embodiment 1 of this application; Figure 2 This is a schematic diagram of a high-precision positioning and mapping system for an autonomous driving transport robot provided in Embodiment 2 of this application. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Example 1 like Figure 1 As shown, Embodiment 1 of this application provides a high-precision positioning and map building method for an autonomous driving transport robot, including the following steps: S10. Collect environmental data through multiple sensors, use edge computing devices to preprocess the data in real time, and generate efficient perception information and preliminary environmental models. The core of this step is to obtain detailed data of the environment through the cooperation of multiple sensors such as vision, lidar, ultrasonic sensor and inertial measurement unit. These data will be precisely time-synchronized and spatially aligned to ensure that they can be fused and processed in the same coordinate system, thereby providing reliable data support for subsequent quantitative modeling, path planning and task execution. Through real-time processing of edge computing devices, the system can quickly extract effective information from perception data. This includes the following sub-steps: S101, collect and synchronize data from multiple sensors such as vision, lidar, ultrasonic and inertial measurement unit, perform time and space alignment to form a unified perception dataset; The carrier robot synchronously collects environmental data through multiple sensors. These sensors include vision sensors, lidar, ultrasonic sensors and inertial measurement units. Vision sensors provide environmental images, lidar measures the spatial position and distance of objects, ultrasonic sensors provide near-distance obstacle data, and inertial measurement units record the displacement, acceleration and other dynamic information of the robot. In order to ensure the consistency of the data, the system uses timestamp synchronization algorithm to time-align the data of each sensor, ensuring that all sensor data is collected under the same time reference.

[0016] In addition, in order to effectively fuse data from different sensors, the system also uses coordinate transformation algorithm to convert all sensor data into a unified spatial coordinate system. Through these synchronization and alignment processes, the system can ensure the consistency of sensor data.

[0017] S102, through the edge computing device, the perception dataset is denoised, abnormal values are removed and features are extracted, generating a preliminary environment model containing spatial features and semantic information, providing input for subsequent quantitative modeling.

[0018] After completing data synchronization, the system will denoise and remove outliers from the collected data. Due to environmental interference or other factors, the original data may contain noise, especially in complex environments, image, distance data and motion state information may be affected by weather changes, lighting conditions and sensor characteristics. Therefore, the system uses denoising algorithm to filter the data, remove external interference and ensure that the data for subsequent processing is more accurate. At the same time, the system will identify and remove invalid data points through outlier detection algorithm, such as sensor failure or erroneous data generated under extreme environmental conditions.

[0019] After data cleaning, the system begins feature extraction, transforming environmental data into meaningful spatial and semantic information. For example, lidar data is used to extract the location and shape of obstacles, while visual data can identify and label object types in the environment (such as pedestrians, vehicles, etc.). Through these feature extractions, the edge computing device generates a preliminary environmental model. This model not only includes the geometric structure of the physical space, but also contains the category information of objects, forming an environmental model with semantic labels.

[0020] S20, based on the preliminary environmental model, a quantitative environmental model is constructed, and a correction algorithm is used to improve the positioning accuracy of the robot, especially when the GPS signal is limited; This step uses multi-sensor fusion technology to combine data from lidar, visual sensors, IMU, and other devices to quantitatively model the preliminary environmental model, forming an environmental model that can be efficiently calculated and analyzed. Through real-time positioning correction algorithms, the system can eliminate errors in the environment, especially in areas where GPS signals are weak or invalid, ensuring that the robot's positioning accuracy meets task requirements. Specifically, the following sub-steps are included: S201, through multi-sensor fusion algorithm, the preliminary environmental model is quantitatively modeled, and the obstacles and dynamic elements are converted into calculable spatial feature units; The preliminary environmental model provides basic spatial structure information in the environment, such as the location, distance, and shape of obstacles. However, these data are still relatively rough and need to be further processed to meet the robot's positioning and navigation requirements.

[0021] The multi-sensor fusion algorithm combines data from visual sensors, lidar, ultrasonic sensors, and inertial measurement units to generate a more detailed and quantitative environmental model. By fusing data from different sensors, the system can accurately mark the location and shape of obstacles in space, and even identify the motion trajectory of dynamic obstacles. For example, lidar provides distance information that can accurately depict the geometry of obstacles, while visual sensors can identify the category and state of objects (such as stationary or moving). After fusion processing, these information can be converted into quantitative spatial feature units and stored as an environmental model that can be used by path planning and obstacle avoidance algorithms.

[0022] S202, use correction algorithm to correct positioning error in real time, improve the positioning accuracy of the robot in complex environment.

[0023] In complex environments, especially in areas with high-rise buildings, underground spaces, and other places, GPS signals often cannot provide stable or sufficiently accurate positioning information. Therefore, this step uses different types of correction algorithms to select the most suitable algorithm, such as Kalman filtering or particle filtering, to correct the robot's positioning information in real time according to the specific application scenario.

[0024] The system predicts the robot's position by combining its current motion state (such as acceleration and angular velocity) with spatial information provided by multiple sensors. In the absence of GPS support, the correction algorithm relies on data from other sensors for calculation. By comparing with the pre-constructed quantitative environment model, the system can detect the positioning error of the robot in real time and automatically adjust its position and attitude.

[0025] Kalman filtering is suitable for scenarios with relatively linear system noise, while particle filtering is more suitable for handling complex nonlinear situations. The system will choose the appropriate correction algorithm based on actual needs. In this process, in addition to the inertial data provided by the inertial measurement unit, real-time data from lidar, vision sensors, ultrasonic sensors, and other sensors will be used to correct the robot's positioning error, ensuring that the robot can maintain high-precision positioning capabilities in complex environments, especially in scenarios where GPS signals are limited.

[0026] S30, use deep learning and reinforcement learning algorithms to analyze environmental changes and adaptively adjust sensor parameters to generate optimized perception results, improve perception efficiency and accuracy; In complex and dynamic environments, sensors may encounter different lighting, occlusions, and dynamic objects, affecting the quality of data. Through deep learning models, the system can process data from vision, lidar, ultrasonic, and other sensors in real time, identifying and extracting environmental features such as obstacles, pedestrians, and other key elements. At the same time, reinforcement learning algorithms help the system adaptively optimize sensor configurations based on real-time environmental changes, ensuring optimal perception results in different environmental conditions. Specific sub-steps include: S301, analyze perception data through deep learning models to identify obstacles, pedestrians, and environmental change elements; The system uses multiple deep learning models to analyze data from different sensors in real time, identifying obstacles, pedestrians, and other dynamic objects. Specifically, for lidar data, the system uses PointNet (a deep learning model for processing point cloud data) to identify and classify three-dimensional obstacles in the environment. PointNet can extract spatial structure features from lidar scanning data, accurately identifying the location, shape, and size of obstacles. By using this model, the system can process lidar point cloud data and effectively identify static and dynamic obstacles.

[0027] For ultrasonic sensor data, the system employs a Convolutional Neural Network (CNN) to automatically identify the shape and location of surrounding obstacles by extracting distance and reflection signal data returned by the sensors. The data from ultrasonic sensors typically has low resolution, but through the convolutional and pooling layers of the CNN, the system can effectively extract useful environmental feature information and perform obstacle detection and classification.

[0028] For image data acquired by visual sensors such as cameras, the system uses YOLO (You Only Look Once) or Faster R-CNN object detection networks to identify pedestrians, vehicles, and other dynamic objects in the environment. These deep learning models can quickly locate and classify objects in real-time video streams, improving the efficiency and accuracy of visual data processing.

[0029] Through the combined application of deep learning models, the system can process data from multiple sensors and efficiently identify and classify various objects in dynamic environments. The system integrates these identification results to generate high-precision, usable perception data, which in turn provides accurate information for path planning, decision-making, and control.

[0030] S302, based on the reinforcement learning algorithm, automatically adjust the working mode and parameters of the sensor according to the real-time environmental feedback, optimize the stability of data acquisition and feature extraction.

[0031] In traditional autonomous driving systems, the working parameters of sensors are usually fixed, while the invention uses deep Q network (DQN) reinforcement learning method to dynamically optimize the working mode of sensors, so that they can automatically adjust to obtain the best perception effect under changing environmental conditions.

[0032] In this process, the system simulates the performance of sensors under different environmental conditions through reinforcement learning algorithm training, so that the robot can automatically adjust the working parameters of the sensor according to the current environmental state (such as changes in light, appearance of obstacles, etc.). For example, when the environmental light is too strong, the system can automatically adjust the exposure time of the visual sensor to avoid image overexposure; in low light conditions, the system may increase the exposure or enhance the contrast of the image. At the same time, the scanning frequency and range of laser radar and ultrasonic sensors will also be optimized according to real-time feedback to improve the accuracy of obstacle detection.

[0033] The reinforcement learning algorithm learns through interaction with the environment, constantly optimizing the parameters of the sensors. The system uses the feedback information perceived as a reward signal to adjust the working mode of each sensor, so that the robot can maintain efficient and stable perception performance in various environments. For example, in rainy or smoggy weather, the system will increase the sampling frequency of the laser radar to enhance the perception ability in low visibility conditions.

[0034] In this way, the system can ensure that the sensor works efficiently and stably under various environmental conditions. The reinforcement learning algorithm adjusts the working parameters of the sensor adaptively according to real-time environmental feedback, optimizes the stability of data acquisition and feature extraction, and ensures that the generated perception data is accurate and stable, thereby improving the perception ability of the entire autonomous driving system.

[0035] S40, based on the optimized perception result and the quantization model, a three-dimensional map is generated in real time, and dynamic updating and error optimization are performed according to the positioning correction and environmental change, to ensure the consistency and accuracy of the map in the time and space dimensions; This step uses multi-source sensor data and real-time feedback mechanism to build and maintain a high-precision three-dimensional environment model, so that the robot can accurately understand the real-time changes of the surrounding environment and respond. Specifically, the following sub-steps are included: S401, fuse the quantization model and real-time perception data, generate a three-dimensional environment map through sparse point cloud reconstruction algorithm, and establish topological node relationship to describe the spatial structure; The system fuses the quantization environment model with the real-time collected perception data, and generates a three-dimensional environment map using sparse point cloud reconstruction algorithm. The sparse point cloud algorithm uses multi-source data such as laser radar, visual sensor and ultrasonic sensor to first generate a rough three-dimensional point cloud map. Then, the system performs point cloud sparsification through optimization algorithm, retains key environmental feature information, reduces redundant data, and improves calculation efficiency.

[0036] After further processing of the generated point cloud data, the system will construct the spatial structure of the environment according to these point cloud data. This process helps the system establish the connection relationship between different objects and obstacles in space through the construction of topological node relationship. For example, the system can identify the spatial distance between obstacles and establish the corresponding topological structure, so that the three-dimensional map not only reflects the geometric shape of the object, but also accurately describes its relative position and relationship in space.

[0037] At this time, the generated three-dimensional map contains key spatial elements such as obstacles, roads, pedestrians and dynamic objects, and provides high-precision spatial layout.

[0038] S402, according to the positioning correction feedback and environmental change data, dynamically update the map node and boundary information, perform error correction and time series smoothing, and ensure high consistency between the map and the actual environment.

[0039] The system dynamically updates the node and boundary information of the three-dimensional map based on real-time feedback from the localization correction algorithm and environmental changes, and performs error correction and time series smoothing processing. As the robot moves in the environment, sensor data and positioning information will change, which may cause some information in the map to deviate or become outdated. Therefore, the system needs to correct the map data in real time to ensure that the map remains highly consistent with the actual environment.

[0040] Firstly, the system adjusts the error of the robot's current position in real time through localization correction feedback. By comparing the current positioning information with the existing data in the map, the system can detect positioning errors and correct them in a timely manner. For example, if the robot enters an area that is not fully covered or where the GPS signal is weak, the system will use lidar or IMU data to correct the positioning, preventing the map from deviating.

[0041] Subsequently, the system dynamically updates the map based on environmental change data (such as newly appearing obstacles, dynamic changes of pedestrians or objects). By analyzing sensor data in real time, the system can identify new changes in the environment and update the map nodes in a timely manner, adjusting the position or state of obstacles. At the same time, the system uses time series smoothing algorithms to optimize the data in the map, reducing errors caused by data fluctuations and ensuring the stability and accuracy of the map updating process.

[0042] S50, based on the dynamic three-dimensional map and positioning information, a multi-level path planning structure is constructed to realize path generation, evaluation and real-time optimization; This step ensures that the robot can efficiently and safely plan the driving path in a complex environment through a multi-level path planning framework. The initial path is generated through heuristic search algorithms, and in a dynamic environment, the perception data and risk factors are combined to optimize the path using dynamic programming algorithms to adjust the feasibility and safety of the path in real time. Specifically, the following sub-steps are included: S501, based on node connectivity, an initial path graph is constructed on the three-dimensional map, and a preliminary driving path is generated through heuristic search algorithms; The system constructs an initial path graph by analyzing the node connectivity of the three-dimensional map. The three-dimensional map contains a large number of environmental nodes (such as obstacles, road markings, pedestrians, etc.), which represent key positions that the robot may encounter during navigation. The system generates a path graph based on the connectivity of these nodes (such as whether there is a direct path between two nodes or whether it is blocked by obstacles).

[0043] To ensure that the path graph effectively reflects the real environment, the system generates a preliminary driving path through heuristic search algorithms such as A* or Dijkstra. These heuristic algorithms select the optimal path based on the weight of each node (e.g., the distance between nodes, the impact of obstacles, etc.). The heuristic search method prioritizes paths closer to the target location by calculating the estimated cost of each node, ensuring the efficiency and reliability of path planning.

[0044] After the preliminary path graph is generated, the system obtains a preliminary navigation path.

[0045] S502, in combination with the perception feedback and risk factors, the dynamic programming algorithm is used to adjust the path node weight in real time, and a multi-objective optimization path is generated.

[0046] The system combines perception data and real-time changes in the environment, and uses a dynamic programming algorithm to optimize the initial path. Path planning not only needs to consider the shortest path or optimal path, but also needs to be adjusted in real time according to the risk factors in the environment. For example, the appearance of dynamic obstacles (such as pedestrians, vehicles, etc.) or changes in environmental conditions (such as road changes, traffic conditions, etc.) may affect the feasibility and safety of the original path.

[0047] The dynamic programming algorithm adjusts the node weight of the path based on the current environmental perception feedback at each step. The basis for weight adjustment includes: Perception feedback: data from visual, lidar, ultrasonic, etc. sensors provide the latest information about the surrounding environment.

[0048] Risk factors: such as the size of obstacles, the motion trajectory of pedestrians or vehicles, the road capacity, etc. are used to adjust the risk weight of the path nodes.

[0049] In this process, the dynamic programming algorithm adjusts each node of the path in real time, so that the robot can choose a safe and effective path in a complex environment. For example, when the system detects an obstacle in front of it, it will automatically recalculate the path to avoid the obstacle and choose a detour path. At the same time, the algorithm will adjust the weight of the path according to the multi-objective optimization of the path (such as shortest time, lowest risk, minimum energy consumption, etc.), to generate a comprehensive optimal path.

[0050] S60, execute control instructions according to the optimized path output by the path planning module, and dynamically adjust the motion trajectory in combination with real-time perception feedback to achieve safe and efficient task execution.

[0051] During path execution, the system monitors the position and pose deviation of the robot when executing the path, and automatically triggers the local obstacle avoidance module for real-time adjustment in response to changes in obstacles and path blockage, etc. In this way, the system ensures that the robot can flexibly respond to changes in complex environments and ensure task continuity and efficient completion. Specifically, the following sub-steps are included: S601, converting the planned path into executable motion control instructions, and monitoring the pose deviation, obstacle changes and dynamic response parameters during execution; The system first converts the optimized path provided by the path planning module into specific motion control instructions that the robot can execute. These instructions include the robot's speed, acceleration, turning angle, etc., to ensure that the robot can travel along the predetermined path. During path execution, the system monitors the robot's state in real time, especially the robot's position and pose. By comparing the current actual position with the predetermined path, the system can calculate the pose deviation of the robot from the target path and assess whether the motion trajectory needs to be adjusted. In addition, the system also monitors environmental changes, such as the appearance of new obstacles, changes in the position of existing obstacles, and other factors that may affect the robot's motion (such as traffic conditions, weather changes, etc.). When the system finds that the robot deviates significantly from the path, it will adjust the control instructions in real time to ensure that the robot can continue to move towards the target.

[0052] S602, when the path is blocked or the deviation exceeds the limit, automatically trigger the local obstacle avoidance module, re-plan the local path based on the real-time map, and update the control instructions to maintain task continuity.

[0053] When the system detects that the path is blocked or the pose deviation exceeds the preset range during path execution, the local obstacle avoidance module will be automatically activated. The system quickly analyzes the position and size of the current obstacles based on real-time perception data, especially the information feedback from laser radar, vision sensor and ultrasonic sensor, and calculates the specific reasons for blocking the path. At this time, the system not only re-plans the local path, but also generates a new driving route according to the changes in obstacles, the current position of the robot and the requirements of the target task.

[0054] The local obstacle avoidance module re-plans the path based on the real-time map to ensure that the robot can bypass the obstacles and return to the optimal driving path. The re-planned path takes into account multiple factors such as the robot's current speed, acceleration limit, dynamic response, etc., to ensure the feasibility and efficiency of the new path. At this time, the system updates the control instructions according to the new local path, adjusts the robot's driving direction, speed and other motion parameters, to ensure that the robot can successfully avoid obstacles and continue to execute tasks.

[0055] Embodiment Two As Figure 2As shown, the second embodiment of the present application provides a high-precision positioning and map construction system for an autonomous delivery robot, which comprises: A perception data acquisition module 21 is configured to acquire environmental data through various sensors and provide the data to the edge computing device. The perception data acquisition module is responsible for acquiring environmental data through various sensors and providing real-time input for subsequent data processing and analysis. The core function of this module is to use visual sensors, lidar, ultrasonic sensors, and inertial measurement units (IMUs) to collect multi-dimensional information about the environment. Through data fusion and synchronization, the consistency of sensor data is ensured, and high-quality input is provided to the edge computing module. Specifically, the following sub-modules are included: A multi-source sensor data acquisition sub-module 211 acquires environmental images through visual sensors, measures the spatial distance between objects and the robot using lidar, provides detection information for close-range obstacles using ultrasonic sensors, and provides acceleration, angular velocity, and displacement data about the robot using IMUs. The coordinated work of these sensors provides comprehensive environmental perception, ensuring that the robot can monitor objects and obstacles around it in real time in a dynamic environment.

[0056] A data synchronization and spatial alignment sub-module 212 ensures efficient fusion of multiple sensor data by using precise timestamp synchronization algorithms to time-align all sensor data. Then, through coordinate transformation algorithms, the data collected by each sensor is converted to a unified spatial coordinate system, allowing all perception data to be processed and analyzed in the same framework.

[0057] An edge computing preprocessing module 22 is configured to denoise, correct, synchronize, and feature extract sensor data to generate a preliminary environmental model. The edge computing preprocessing module is the core module of data processing, mainly responsible for denoising, outlier rejection, synchronization, and feature extraction of collected sensor data. Through real-time computation and data cleaning, it ensures that subsequent environmental modeling and path planning tasks can be performed based on high-quality data. Specifically, the following sub-modules are included: A data denoising and outlier processing sub-module 221 can eliminate external interference factors and ensure data quality by filtering raw data using denoising algorithms. At the same time, it uses outlier detection techniques to eliminate invalid data points under fault or extreme conditions, providing a reliable data foundation for subsequent data modeling.

[0058] Feature extraction and preliminary environment modeling submodule 222, after data cleaning, the system will extract features from the perception data, and convert the perception data into meaningful spatial and semantic information. Through the spatial information processing of the laser radar data, the system extracts the position, shape and relative distance of the obstacles; through the analysis of image data by the vision sensor, dynamic objects such as pedestrians and vehicles are identified. Through these feature extraction, the edge computing device generates a preliminary environment model, which contains the geometric structure and semantic information of the physical space.

[0059] Environment modeling and positioning correction module 23, for constructing a quantitative environment model based on the preliminary environment model, and improving the positioning accuracy in real time through correction algorithm; The main function of this module is to construct a quantitative environment model based on the preliminary environment model, and to improve the positioning accuracy of the robot in real time by using correction algorithm, especially in the scene where GPS signal is limited. Through multi-sensor fusion, the system can generate an accurate environment model and correct the position of the robot in real time. It includes the following submodules: Multi-sensor fusion modeling submodule 231, in this submodule, the system fuses data from vision sensors, laser radars, ultrasonic sensors and IMU through multi-sensor fusion algorithm, to generate a more accurate quantitative environment model. This model can effectively represent the position, shape and trajectory of dynamic obstacles, providing basic data for path planning and obstacle avoidance algorithms.

[0060] Positioning correction and error correction submodule 232, in the area where GPS signal is weak or invalid, the system corrects the positioning information of the robot in real time through correction algorithms such as Kalman filter or particle filter, combined with IMU and laser radar data. The system compares the current position of the robot with the environment data, calculates the positioning error, and automatically corrects it to ensure the positioning accuracy of the robot in complex environments.

[0061] Intelligent perception and sensor adjustment module 24, for analyzing environmental changes and adaptively adjusting the working mode and parameters of the sensor by using deep learning and reinforcement learning algorithms; This module analyzes environmental changes and adaptively adjusts the working mode and parameters of the sensor by using deep learning and reinforcement learning algorithms, to generate optimized perception results. Through reinforcement learning, the system can optimize the perception efficiency and accuracy in dynamic environment. It includes the following submodules: Deep learning analysis submodule 241, the system analyzes data from different sensors through deep learning model, identifies and classifies dynamic objects such as obstacles, pedestrians and vehicles in the environment. Deep learning algorithm can extract complex spatial features from sensor data and optimize perception results in real time.

[0062] The reinforcement learning adaptive adjustment submodule 242, through a reinforcement learning algorithm, the system can adaptively adjust the configuration of the sensor according to the environmental feedback. For example, adjusting the exposure of the visual sensor, the scanning frequency of the laser radar, etc., to ensure efficient perception performance under different light, weather or obstacle changes.

[0063] The three-dimensional map construction and correction module 25 is used for generating a three-dimensional map based on a quantitative model and a perception result, and dynamically updating and error optimizing; The module uses a quantitative model and real-time perception data to construct and maintain a three-dimensional environment map, and dynamically updates and error optimizes according to environmental changes, ensuring the consistency and accuracy of the map in the time and space dimensions. Specifically, the following submodules are included: The three-dimensional reconstruction and topological relationship construction submodule 251 generates a three-dimensional environment map by combining a quantitative model and real-time perception data through a sparse point cloud reconstruction algorithm, and accurately describes the spatial structure by establishing a topological node relationship. At this time, the generated three-dimensional map not only contains the geometric shape of the objects in the environment, but also accurately represents the relative position and relationship between the objects.

[0064] The dynamic updating and error correction submodule 252 dynamically updates the map nodes and boundary information based on positioning correction feedback and environmental change data, and performs error correction and time series smoothing. By dynamically updating the map, the system ensures that the map is highly consistent with the real environment.

[0065] The path planning and execution module 26 includes a path generation unit and a control execution unit, which is used for path planning, evaluation, optimization and dynamic execution control based on a three-dimensional map and positioning information, to ensure efficient and safe completion of tasks.

[0066] The module is responsible for path planning, evaluation and optimization based on a three-dimensional map and positioning information, and realizes dynamic adjustment and execution of the path through a control execution unit, to ensure efficient and safe completion of tasks. Specifically, the following submodules are included: The path generation and evaluation submodule 261 generates a preliminary path through a heuristic search algorithm (such as A* algorithm, Dijkstra algorithm, etc.), and performs real-time optimization and evaluation based on multi-source perception data and environmental feedback. Path planning not only considers the shortest path, but also considers factors such as obstacles and traffic conditions, to ensure that the robot can choose the best path.

[0067] The control execution and local obstacle avoidance submodule 262 automatically triggers the local obstacle avoidance module to dynamically adjust the robot path by monitoring the deviation in the path execution process in real time. When obstacles or path blockages are detected, the system will re-plan a local path based on the real-time map, to ensure the continuity and safety of the task.

[0068] Corresponding to the above-mentioned embodiments, the embodiments of the present application provide a computer storage medium, comprising at least one memory and at least one processor. The memory is configured to store one or more program instructions. The processor is configured to execute the one or more program instructions to perform the method for high-precision positioning and map construction of an autonomous driving carrying robot.

[0069] Corresponding to the above-mentioned embodiments, the embodiments of the present application provide a computer readable storage medium, the computer readable storage medium comprising one or more program instructions, the one or more program instructions being configured to be executed by a processor to perform the method for high-precision positioning and map construction of an autonomous driving carrying robot.

[0070] The embodiments disclosed in the present application provide a computer readable storage medium, the computer readable storage medium storing computer program instructions, when the computer program instructions are executed on a computer, the computer executes the above-mentioned method for high-precision positioning and map construction of an autonomous driving carrying robot.

[0071] In the embodiments of the present application, the processor can be an integrated circuit chip with processing capability of signals. The processor can be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0072] The disclosed methods, steps and logic block diagrams in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can be any conventional processor. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as hardware code processor execution or executed by a combination of hardware and software modules in the code processor. The software module can be located in a random memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register or other mature storage medium in the art. The processor reads the information in the storage medium and combines the hardware to complete the steps of the above-mentioned method.

[0073] The storage medium can be a memory, for example, can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories.

[0074] The non-volatile memory can be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically EPROM (EEPROM), or a flash memory.

[0075] The volatile memory can be a Random Access Memory (RAM) used as an external cache. By way of example, and not limitation, many forms of RAM are available, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM).

[0076] The storage media described in the embodiments of the present application are intended to include, but are not limited to, these and any other suitable types of memory.

[0077] Those skilled in the art should be aware that the functions described in the embodiments of the present application can be implemented in combination of hardware and software in one or more of the above examples. When the software is applied, the corresponding functions can be stored in a computer readable medium or transmitted as one or more instructions or codes on the computer readable medium. The computer readable medium includes a computer storage medium and a communication medium, wherein the communication medium includes any medium that facilitates the transfer of computer programs from one place to another. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0078] The above detailed description of the specific embodiments of the present application is further intended to explain the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above detailed description is only a specific embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made on the basis of the technical solutions of the present application should be included in the protection scope of the present application.

Claims

1. A high-precision positioning and map construction method for an autonomous driving delivery robot, characterized in that, The method comprises the following steps: S10, collecting environment data through various sensors, using edge computing devices to pre-process the data in real time, generating efficient perception information and a preliminary environment model; S20, constructing a quantitative environment model based on the preliminary environment model, and improving the positioning accuracy of the robot using a correction algorithm, especially when the GPS signal is limited; S30, using deep learning and reinforcement learning algorithms to analyze environmental changes and adaptively adjust sensor parameters to generate optimized perception results; S40, based on the optimized perception results and the quantitative model, generating a three-dimensional map in real time, and dynamically updating and optimizing errors according to the positioning correction and environmental changes, ensuring the consistency and accuracy of the map in the time and space dimensions; S50, based on the dynamic three-dimensional map and positioning information, constructing a multi-level path planning structure to realize path generation, evaluation and real-time optimization; S60, executing control instructions according to the optimized path output by the path planning module, and dynamically adjusting the motion trajectory combined with real-time perception feedback to realize safe and efficient task execution. 2.The high-precision positioning and map construction method of the automatic driving carrier robot according to claim 1, characterized in that, Collecting environment data through various sensors, using edge computing devices to pre-process the data in real time, thereby generating efficient perception information and a preliminary environment model, including the following sub-steps: S101, collecting and synchronizing data from multiple sources such as vision, laser radar, ultrasonic and inertial measurement unit, performing time and space alignment to form a unified perception data set; S102, using edge computing devices to denoise, exclude outliers and extract features from the perception data set, generating a preliminary environment model containing spatial features and semantic information. 3.The high-precision positioning and map construction method of the automatic driving carrier robot according to claim 1, characterized in that, Based on the preliminary environment model, a quantitative environment model is constructed, and a correction algorithm is used to improve the positioning accuracy of the robot, especially when the GPS signal is limited, including the following sub-steps: S201, using a multi-sensor fusion algorithm to quantitatively model the preliminary environment model, converting obstacles and dynamic elements into calculable spatial feature units; S202, using a correction algorithm to correct positioning errors in real time, improving the positioning accuracy of the robot in complex environments. 4.The high-precision positioning and map construction method of the autonomous driving carrier robot according to claim 1, characterized in that, Using deep learning and reinforcement learning algorithms to analyze environmental changes and adaptively adjust sensor parameters to generate optimized perception results, including the following sub-steps: S301, analyzing perception data through a deep learning model to identify obstacles, pedestrians and environmental change elements; S302, based on a reinforcement learning algorithm, automatically adjusting the working mode and parameters of the sensor according to real-time environmental feedback to optimize the stability of data acquisition and feature extraction. 5.The high-precision positioning and map construction method of the autonomous driving carrier robot according to claim 1, characterized in that, Based on the optimized perception results and the quantitative model, a three-dimensional map is generated in real time, and dynamic updating and error optimization are performed according to the positioning correction and environmental changes, including the following sub-steps: S401, fusing the quantitative model and real-time perception data, generating a three-dimensional environment map through a sparse point cloud reconstruction algorithm, and establishing a topological node relationship to describe the spatial structure; S402, dynamically updating map nodes and boundary information according to positioning correction feedback and environmental change data, performing error correction and time series smoothing to ensure high consistency between the map and the actual environment. 6.The high-precision positioning and map construction method of an autonomous driving delivery robot according to claim 1, wherein, Based on dynamic three-dimensional map and positioning information, a multi-level path planning structure is constructed to realize path generation, evaluation and real-time optimization, including the following sub-steps: S501, constructing an initial path graph based on node connectivity on the three-dimensional map, and generating a preliminary driving path through a heuristic search algorithm; S502, combining perception feedback and risk factors, and using dynamic programming algorithm to adjust path node weight in real time to generate multi-objective optimization path. 7.The high-precision positioning and map construction method of the autonomous driving carrier robot according to claim 1, characterized in that, According to the output of the path planning module, the control instruction is executed, and the motion trajectory is dynamically adjusted combined with real-time perception feedback, including the following sub-steps: S601, converting the planned path into executable motion control instructions, and monitoring the pose deviation, obstacle change and power response parameters in the execution process; S602, when it is detected that the path is blocked or the deviation is out of limit, the local obstacle avoidance module is automatically triggered, the local path is re-planned based on the real-time map, and the control instruction is updated to maintain the task continuity.

8. A high-precision positioning and map construction system for an autonomous driving delivery robot, characterized by, It includes: a perception data acquisition module for collecting environmental data through various sensors and providing it to an edge computing device; an edge computing preprocessing module for denoising, correcting, synchronizing and feature extraction of sensor data to generate a preliminary environment model; an environment modeling and positioning correction module for constructing a quantitative environment model based on the preliminary environment model and improving positioning accuracy in real time through correction algorithms; an intelligent perception and sensor adjustment module for analyzing environmental changes using deep learning and reinforcement learning algorithms and adaptively adjusting the working mode and parameters of the sensors; a three-dimensional map construction and correction module for generating a three-dimensional map based on the quantitative model and perception results, and dynamically updating and optimizing errors; a path planning and execution module including a path generation unit and a control execution unit for path planning, evaluation, optimization and dynamic execution control based on three-dimensional map and positioning information to ensure efficient and safe completion of the task.

Images