Campus express air-ground integrated unmanned delivery system and use method

Abstract

The invention discloses a campus express air-ground integrated unmanned distribution system and a use method. The system comprises a front-end application service module, an express platform integrated service module, an unmanned vehicle scheduling and control service module, an unmanned aerial vehicle scheduling and control service module, a communication service module and a big data storage and analysis service module. The unmanned vehicle scheduling and control service module comprises unmanned vehicle intelligent scheduling and unmanned vehicle adaptive path adjustment, and is used for scheduling an unmanned vehicle to execute a delivery task according to the obtained express information and the matched address; the unmanned aerial vehicle scheduling and control service module comprises unmanned aerial vehicle adaptive path adjustment and is used for scheduling an unmanned aerial vehicle to handover with an unmanned vehicle and continuing to execute a distribution task; and the big data storage and analysis service module is used for storing and analyzing the distribution data so as to realize a dynamic adjustment mechanism. According to the invention, adaptive path adjustment of the unmanned vehicle and the unmanned aerial vehicle can be realized, and an efficient, safe and flexible campus express air-ground integrated distribution service is provided.

Description

Technical Field

[0001] The field of the present invention is the field of unmanned delivery technology, and specifically relates to a campus express air-ground integrated unmanned delivery system and a usage method thereof. Background Art

[0002] With the rapid development of the logistics and distribution industry, especially in special scenarios such as campuses, traditional logistics and distribution methods face many challenges. The campus environment is characterized by dense population, regular building layout but with restricted areas (such as non-passable areas like playgrounds and gardens), and large fluctuations in express delivery volume. These all pose higher requirements for the efficiency, safety, and flexibility of logistics and distribution.

[0003] Existing air-ground integrated logistics and distribution methods have received extensive attention, and their core lies in using the respective advantages of unmanned aerial vehicles (UAVs) and unmanned vehicles for collaborative operations. For example, Patent CN117172656A proposes an air-ground collaborative logistics planning method under road network constraints. By obtaining the real map of the target area (including building information and road network information), randomly sampling on the map to generate UAV delivery points and unmanned vehicle delivery points, then using the improved K-Means algorithm to cluster the UAV delivery points, determining the distance matrix based on the clustering center, and further using a heuristic algorithm to calculate the initial delivery path of the unmanned vehicle, and on this basis, adopting a local collaborative distribution planning algorithm for the collaborative distribution planning of UAVs and unmanned vehicles. Patent CN117436782A focuses on campus unmanned vehicle delivery, constructs a system by integrating digital twin technology, collects data through sensors, GPS, etc. and transmits it to the digital twin platform, uses a multiple linear regression model to predict traffic flow, and adopts the A* algorithm for path planning. Although the existing technologies have improved the efficiency and flexibility of logistics and distribution to a certain extent, in the special scenario of the campus, there are still many deficiencies:

[0004] 1. The delivery path planning is not accurate enough

[0005] The multiple linear regression model adopted by Patent CN117436782A has limitations in adaptability. This model is constructed based on specific input features and weights, and has limited fitting ability for the possible non-linear relationships in campus traffic (such as the complex changes in the sudden increase of people and vehicles during special periods). In complex traffic scenarios, it may not be able to effectively capture the dynamic change law of traffic flow, resulting in inaccurate delivery path planning.

[0006] 2. The dynamic adaptability of path planning is insufficient

[0007] The A of Patent CN117436782A *Although the algorithm takes into account the delivery location and predicted traffic flow during planning, the campus traffic situation changes rapidly. The initial route generated based on only this information may become suboptimal during the delivery process due to new traffic congestion or temporary traffic control. It lacks sufficient adaptive path adjustment ability and cannot respond to emergencies in a timely manner. The path planning of Patent CN117172656A relies on the real map information of the target area, but does not mention special optimization measures for unmanned delivery in complex environments. In actual logistics delivery, emergencies may occur, such as traffic accidents causing road network congestion and bad weather affecting the flight of drones, which will all affect the delivery efficiency.

[0008] 3. The accuracy of air-ground integrated collaborative scheduling is not high

[0009] The algorithm for air-ground integrated collaborative scheduling adopted by Patent CN117172656A has low accuracy and relies on the real map information of the target area. Once there are deviations in the map data, such as incorrect building positions, incomplete road network information or lagging updates, etc., it will not only greatly reduce the accuracy of sampling, clustering and path planning of delivery points, but also further exacerbate the delay in communication and coordination between unmanned vehicles and drones, ultimately leading to a reduction in delivery efficiency and even possibly causing the delivery task to fail. Summary of the Invention

[0010] The purpose of the present invention is to provide a campus express air-ground integrated unmanned delivery system and method. The present invention can achieve adaptive path adjustment of unmanned vehicles and drones, and provide efficient, safe and flexible campus express air-ground integrated delivery services.

[0011] The technical solution of the present invention: A campus express air-ground integrated unmanned delivery system includes a front-end application service module, an express platform integrated service module, an unmanned vehicle scheduling and control service module, a drone scheduling and control service module, a communication service module, and a big data storage and analysis service module; the front-end application service module is respectively connected to the express platform integrated service module, the unmanned vehicle scheduling and control service module, the drone scheduling and control service module, and the big data storage and analysis service module for data interaction; the express platform integrated service module is connected to the unmanned vehicle scheduling and control service module to transmit express-related data; the unmanned vehicle scheduling and control service module and the drone scheduling and control service module communicate and cooperate through the communication service module; the big data storage and analysis service module provides data support for the front-end application service module, the unmanned vehicle scheduling and control service module, and the drone scheduling and control service module;

[0012] Among them: The front-end application service module is used for user registration, login, order placement, querying order status, and receiving notifications;

[0013] The express delivery platform integration service module is used to interface with the express delivery service platform, obtain express delivery information and perform address matching, and synchronize the express delivery status information in real time;

[0014] The unmanned vehicle scheduling and control service module includes unmanned vehicle intelligent scheduling and unmanned vehicle adaptive path adjustment, and is used to schedule unmanned vehicles to perform delivery tasks according to the obtained express delivery information and the matched address;

[0015] The unmanned aerial vehicle scheduling and control service module includes unmanned aerial vehicle adaptive path adjustment, and is used to schedule unmanned aerial vehicles to hand over with unmanned vehicles and continue to perform delivery tasks;

[0016] The communication service module is used to realize the communication between the front-end application service module, the unmanned vehicle scheduling and control service module, and the unmanned aerial vehicle scheduling and control service module;

[0017] The big data storage and analysis service module is used to store and analyze delivery data to realize a dynamic adjustment mechanism for fluctuations in express delivery volume.

[0018] In the aforementioned campus express delivery integrated air and ground unmanned delivery system, the unmanned vehicle intelligent scheduling is responsible for receiving express delivery order information and obtaining the running status of all unmanned vehicles. According to the express delivery order information and the running status of unmanned vehicles, intelligent scheduling algorithms are used for order initialization allocation and order delivery execution. According to the running status of unmanned vehicles and road conditions, the order allocation and unmanned vehicle scheduling are dynamically adjusted.

[0019] In the aforementioned campus express delivery integrated air and ground unmanned delivery system, the specific process of the intelligent scheduling algorithm is to collect detailed express delivery order information, unmanned vehicle running status information, and road condition information, traverse all unmanned vehicles, construct a comprehensive cost function, and calculate the comprehensive cost of each unmanned vehicle executing an order through weighted summation of weights; the process of dynamic adjustment is to monitor the running status and road conditions of unmanned vehicles in real time during the task execution of unmanned vehicles. When a unmanned vehicle breaks down, has too low battery power, or is severely congested, its uncompleted orders are reallocated; at the same time, the regional order distribution is regularly checked, the unmanned vehicle resources and driving routes are adjusted, and urgent orders are given priority to ensure the efficient operation of the system.

[0020] In the aforementioned campus express delivery integrated air and ground unmanned delivery system, the unmanned vehicle adaptive path adjustment is based on real-time traffic conditions and real-time environmental information, analyzes and predicts real-time traffic conditions and real-time environmental information through existing large language models, and then uses reinforcement learning algorithms to plan the optimal path.

[0021] In the aforementioned integrated ground and air unmanned delivery system for campus express delivery, the specific process of the reinforcement learning algorithm for planning the optimal path is to define a state space containing multi-dimensional information, collect real-time traffic data, environmental information, campus layout, historical traffic patterns, and the current time; set the action set that the unmanned vehicle can choose in path planning, including driving direction, lane change, and speed adjustment; design a reward function to guide the unmanned vehicle to learn the optimal path decision; take the current state of the unmanned vehicle as the input, predict the estimated values of the long-term cumulative rewards of different actions, and select the action with the largest estimated value to execute. Update the neural network parameters through actual execution and feedback, and optimize the estimated value prediction to achieve rapid planning of the optimal path in real-time state.

[0022] In the aforementioned integrated ground and air unmanned delivery system for campus express delivery, the unmanned aerial vehicle's adaptive path adjustment is responsible for receiving the express delivery order information delivered by the unmanned vehicle and obtaining the operating state of the unmanned aerial vehicle. According to the express delivery order information and the operating state of the unmanned aerial vehicle, it allocates the unmanned aerial vehicle to conduct the handover of express delivery orders and execute order delivery, and then plans the optimal flight path for the unmanned aerial vehicle in combination with the three-dimensional building map and real-time environmental information, uses computer vision technology and deep learning algorithms for target detection and positioning, and assists in precise landing; during the working process, it monitors the operating state of the unmanned aerial vehicle in real-time, and feeds back the operating state of the unmanned aerial vehicle to the unmanned aerial vehicle scheduling and control service module and the front-end application service module to conduct flexible adjustment of order delivery and push delivery notifications.

[0023] In the aforementioned integrated ground and air unmanned delivery system for campus express delivery, the specific process of using computer vision technology and deep learning algorithms for target detection and positioning is to collect and preprocess the image data of the surrounding environment; establish a target detection model, use deep learning algorithms to train the target detection model, input the preprocessed image into the trained target detection model, detect the position and confidence of the target floor feature markers, combine the positioning system and three-dimensional map information of the unmanned aerial vehicle, calculate the actual spatial position of the target floor, provide a basis for flight path adjustment, continuously collect images and conduct target detection and positioning during the flight of the unmanned aerial vehicle; according to the change of the target position and the state of the unmanned aerial vehicle, adjust the flight attitude and speed through the flight control system to ensure that the unmanned aerial vehicle always flies towards the target floor and finally achieves precise landing.

[0024] In the aforementioned integrated ground and air unmanned delivery system for campus express delivery, the delivery data includes express delivery order information, the operating state of the unmanned vehicle, the operating state of the unmanned aerial vehicle, and traffic conditions; the dynamic adjustment mechanism monitors the delivery data and uses data mining and machine learning algorithms for analysis, thereby predicting the business volume of express delivery orders, and conducting scheduling, task allocation optimization, and dynamic adjustment of the working area of the unmanned vehicle and the unmanned aerial vehicle according to the prediction results.

[0025] In the aforementioned integrated ground and aerial unmanned delivery system for campus express delivery, the process of analysis by the data mining and machine learning algorithms is as follows: collect express order information, the operating status data of unmanned vehicles and drones, and traffic condition data, and use data cleaning technology for data missing, error, and duplication problems; perform standardization and normalization processing on the cleaned data, extract features from the original data to predict the express business volume and optimize scheduling; for the prediction of the express order business volume, use a time series prediction model, adjust the hyperparameters through training data, and minimize the prediction error; evaluate the performance of the time series prediction model and calculate the evaluation metrics.

[0026] In the usage method of the aforementioned integrated ground and aerial unmanned delivery system for campus express delivery, it includes the following steps: Step 1, the user registers, logs in, and places an order through the front-end application service module, and the system express platform integrated service module obtains the order information from the express platform;

[0027] Step 2, according to the order information and the operating status of the unmanned vehicle, the unmanned vehicle scheduling and control module schedules the unmanned vehicle to execute the delivery task according to the self-adaptive path adjustment of the unmanned vehicle to plan the path;

[0028] Step 3, the drone scheduling and control service module schedules the drone for express order handover, and the drone plans the flight path according to the self-adaptive path adjustment of the drone and executes the delivery;

[0029] Step 4, through the communication service module, realize the communication between the front-end application service module, the unmanned vehicle scheduling and control service module, and the drone scheduling and control service module, monitor the delivery status in real time and push it to the front-end application service module;

[0030] Step 5, utilize the big data storage and analysis service module to perform dynamic adjustment using the dynamic adjustment mechanism to optimize the unmanned vehicle scheduling and control service module and the drone scheduling and control service module.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] The present invention gives full play to the advantages of both unmanned vehicles and drones through the collaborative operation of unmanned vehicles. Unmanned vehicles can carry out large-scale ground delivery, while drones can quickly reach some special locations, such as high-rise dormitories or the top of teaching buildings. In the case of densely populated campuses, neatly laid out buildings and restricted areas, the system can dynamically adjust the delivery path according to real-time road conditions and environmental information to avoid congestion and restricted areas, thereby significantly reducing the delivery time. The present invention dynamically adjusts the delivery paths of unmanned vehicles and drones through intelligent scheduling of unmanned vehicles, adaptive path adjustment of unmanned vehicles and adaptive path adjustment of drones, avoids congestion and restricted areas, and efficiently completes the delivery task. The present invention accurately predicts traffic flow through a big data storage and analysis service module, provides a reliable basis for distribution path planning, and enhances the adaptability of the system in complex traffic scenarios. In addition, the unmanned vehicle scheduling of the present invention comprehensively considers multiple factors to calculate the task cost, and can dynamically adjust the task and vehicle allocation, and respectively uses reinforcement learning algorithms, large language models, computer vision technology and deep learning algorithms, etc., combined with campus real-time information and environmental factors, to achieve autonomous adaptive optimal path planning and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a system schematic diagram of the present invention;

[0034] Figure 2 Schematic diagram of the system architecture of the present invention. DETAILED DESCRIPTION

[0035] The present invention is further described below in conjunction with the accompanying drawings and embodiments, but they are not intended to limit the present invention.

[0036] Example

[0037] A campus express delivery air-ground integrated unmanned delivery system, such as Figure 1 and Figure 2As shown in the figure, the system architecture is divided into four layers: the access layer, the gateway layer, the service layer, and the basic service layer. The access layer includes the PC side, the mobile side, third-party applications, and the management background. These terminals communicate with the system through the HTTP protocol. The gateway layer is responsible for processing requests from the access layer and routing and forwarding them through the service gateway (campus network). The gateway layer is also responsible for authentication, authorization, and permission authentication to ensure that only verified users and applications can access system resources. In addition, the gateway layer provides a traffic limiting function to prevent system overload. The service layer is the core part of the system and is responsible for providing automated services and business services. The automated services include OCR intelligent recognition and UKEY device hosting; the business services include the front-end application service module, the express platform integration service module, the unmanned vehicle scheduling and control service module, the drone scheduling and control service module, the communication service module, the big data storage and analysis service module, the security and authentication service module, and the operation and maintenance management service module. OCR intelligent recognition uses optical character recognition to automatically identify and extract express information in images and transmit the express information to the business services. UKEY device hosting is used to manage and control UKEY devices and provide hosting services for devices in the business services. The basic service layer provides underlying support, including cache services, relational databases, blockchains, and distributed file systems. The cache service uses a Redis cluster (a distributed database solution that allows multiple Redis instances to work together to provide higher availability, scalability, and fault tolerance) to improve the system's response speed and performance. UKEY device hosting also manages the cache. The relational database uses a MySQL master-slave cluster (a database replication technology) to provide relational database services and support data storage and management. The blockchain uses blockchain infrastructure to provide blockchain services, and the business services enhance security, transparency, and traceability through the blockchain infrastructure. The distributed file system uses GlusterFS (an open-source distributed file system that allows multiple storage servers to be combined and used as a single, large-capacity file storage system) to provide distributed file system services and support large-scale file storage and sharing. This system adopts a microservices architecture to split each functional service module into independent services and realizes the entire distribution process through the collaboration between services.

[0038] In this embodiment, the front-end application service module is respectively connected to the integrated express platform service module, the unmanned vehicle scheduling and control service module, the unmanned aerial vehicle scheduling and control service module, and the big data storage and analysis service module for data interaction; the integrated express platform service module is connected to the unmanned vehicle scheduling and control service module to transmit express-related data; the unmanned vehicle scheduling and control service module and the unmanned aerial vehicle scheduling and control service module communicate and cooperate through the communication service module; the big data storage and analysis service module provides data support for the front-end application service module, the unmanned vehicle scheduling and control service module, and the unmanned aerial vehicle scheduling and control service module.

[0039] Among them: The front-end application service module is used for user registration, login, placing orders, querying order status, and receiving notifications. In this embodiment, in terms of technology selection, an open-source cross-platform mobile application development framework React Native or software development kit Flutter is used to achieve cross-platform mobile application development, covering iOS and Android systems to ensure consistent user experience. At the same time, GraphQL is used to optimize data query interaction, improve front-end performance, and reduce unnecessary data transmission. GraphQL is a query language and runtime environment for APIs (Application Programming Interfaces), allowing clients to precisely request the required data, thereby improving the efficiency and flexibility of data interaction. The functional implementation of the front-end application service module covers: User registration and login support mobile phone number registration and SMS verification code verification. After logging in, personal information such as address and dormitory building number can be improved for convenient delivery; when placing an order, the order number can be entered or information can be imported from the integrated express platform, the appointment delivery time can be filled in, and after submission, an unmanned vehicle or unmanned aerial vehicle will be automatically assigned for delivery; the express order status query can display the status and trajectory such as received, in transit, etc. in real time; it can also integrate push services (such as OneSignal or Firebase Cloud Messaging) to receive notifications such as express order status updates, unlock codes, and estimated arrival times to ensure that users can obtain express information in a timely manner. OneSignal is a powerful multi-channel message push service widely used in websites and mobile applications, supporting various forms such as push notifications, emails, SMS, and in-app messages. Firebase Cloud Messaging is a cross-platform message push service provided by Google for sending messages to mobile applications and web pages.

[0040] The express delivery platform integration service module is used to connect with the express delivery service platform, obtain express delivery information and perform address matching, and synchronize the express delivery status information in real time. In this embodiment, in terms of technology selection, the OpenAPI specification is used to connect with each express delivery service platform to ensure interface standardization and compatibility. At the same time, data mapping and conversion tools such as JSONata are introduced to unify the data formats of different platforms into a form available to the system. OpenAPI (Open Application Programming Interface) is a standard specification for describing RESTful APIs. It provides a standardized and programming language-independent way to define and describe HTTP API interfaces. JSONata is a lightweight query and transformation language dedicated to processing JSON data. The functional implementation of the express delivery platform integration service module covers: in terms of interface connection, it connects with mainstream platforms such as SF Express and China Post to obtain data such as express delivery numbers, recipient information (including mobile phone numbers), and express delivery status; when matching addresses, it associates the mobile phone number with the mobile phone number registered in the system according to the express delivery number, and completes the distribution address by supplementing the user's dormitory building number; it also realizes real-time data synchronization through timed or event-driven methods to ensure that the order status queried by users in this system is consistent with that of the express delivery platform.

[0041] The unmanned vehicle scheduling and control service module includes unmanned vehicle intelligent scheduling and unmanned vehicle adaptive path adjustment, and is used to schedule unmanned vehicles to perform delivery tasks according to the obtained express delivery information and the matched address.

[0042] Specifically, the unmanned vehicle intelligent scheduling is responsible for receiving express delivery order information and obtaining the running status of all unmanned vehicles. According to the express delivery order information and the running status of the unmanned vehicles, intelligent scheduling algorithms are used to perform order initialization allocation and execute order delivery, and dynamically adjust order allocation and unmanned vehicle scheduling according to the running status of the unmanned vehicles and road conditions.

[0043] In this embodiment, the specific process of the intelligent scheduling algorithm is as follows:

[0044] Step 1: Data Collection and Integration: After receiving a new express delivery order, the system immediately initiates the data collection process. On the one hand, it obtains the detailed information of the express delivery order, including not only the basic recipient address, weight, and scheduled delivery time, but also the type of the express (such as documents, packages, etc.) and special requirements (such as fragile items, urgent deliveries, etc.). On the other hand, it comprehensively collects the operating status information of all unmanned vehicles, covering the real-time location of the unmanned vehicle (obtaining longitude and latitude coordinates through a high-precision GPS positioning system), current battery level (accurate to percentage), load condition (weight and volume of the loaded express), driving speed, health status (judging whether there are potential faults by monitoring the operating parameters of vehicle components through on-vehicle sensors), and road condition information in the area where the vehicle is located, such as road congestion level (measured by the ratio of traffic flow on a road section to the maximum carrying capacity of the road based on real-time traffic data of the vehicle network), and whether there is construction or temporary control (obtained through data docking with the traffic management department or real-time monitoring by drones).

[0045] Step 2: Cost Function Construction and Calculation: For each express delivery order, it traverses all unmanned vehicles and constructs a comprehensive cost function to evaluate the suitability of an unmanned vehicle to execute this order. The cost function contains multiple weighted factors. The distance cost is determined based on the straight-line distance (calculated through map algorithms) between the current location of the unmanned vehicle and the shipping point, as well as the actual driving distance (considering road traffic conditions and restricted areas). The farther the distance, the higher the cost. The load matching cost considers the difference between the current load of the unmanned vehicle and the weight of the order. The larger the difference, the higher the cost. The time cost is calculated by combining the estimated driving time under the current road conditions (predicted through historical traffic data and real-time road conditions), and the difference between the order's scheduled delivery time and the current time. The closer or exceeding the estimated delivery time to the scheduled time, the higher the cost. The special situation cost is for fragile item orders. If there are many bumpy sections on the driving path of the unmanned vehicle, the cost is increased. For urgent delivery orders, the cost of unmanned vehicles far from the shipping point is correspondingly increased. These cost factors are weighted and summed according to a certain weight (optimally determined through machine learning algorithms such as genetic algorithms or SVM (support vector machine) based on actual business requirements and historical data) to obtain the comprehensive cost of each unmanned vehicle to execute this order.

[0046] Step 3: Initial Order Allocation: According to the calculated comprehensive cost, select the unmanned vehicle with the lowest cost as the initial allocated vehicle for this express delivery order, mark the order as allocated, and record the allocation information, including the correspondence between the order and the unmanned vehicle, the estimated departure time, etc.

[0047] Step 4, Dynamic Adjustment Mechanism: During the delivery task execution of the driverless vehicle, the system continuously monitors the running status of the driverless vehicle and the road conditions in real time. Once a driverless vehicle encounters a malfunction (such as abnormal detection of key components by on-vehicle sensors, like motor failure, braking system failure, etc.), low battery power, or severe congestion (traffic flow on the road exceeds 80% of the maximum carrying capacity of the road and lasts for a certain period of time) and is unable to complete the task on time, the system quickly marks the uncompleted orders of this driverless vehicle as reallocation status. Subsequently, the system traverses all available driverless vehicles again, recalculates the comprehensive costs of these driverless vehicles for executing the reallocated orders, and reallocates them in ascending order of cost. In addition, the system regularly checks the order distribution in each area, and judges whether the orders are overloaded by analyzing the ratio of the number of orders in the area to the current processing capacity of the driverless vehicles in this area (determined according to factors such as the number of driverless vehicles, average delivery efficiency, and remaining battery power). If the order volume in a certain area exceeds the current processing capacity of the driverless vehicles in this area, the system first transfers driverless vehicles from other areas with lighter loads (lower ratio of order volume to processing capacity) to this area, and at the same time adjusts the driving routes of the driverless vehicles in this area to prioritize the processing of urgent orders (such as orders with approaching appointment delivery times or orders marked as urgent) to ensure the efficient operation of the entire delivery system.

[0048] Specifically, the adaptive path adjustment of the driverless vehicle is based on real-time traffic conditions, real-time environmental information, and the analysis and prediction of real-time traffic conditions by existing large language models, and plans the optimal path through a reinforcement learning algorithm.

[0049] In this embodiment, in the adaptive path adjustment of the driverless vehicle, containerized deployment is first carried out based on Kubernetes, and ROS is used as the operating system framework. Kubernetes is an open-source container orchestration platform for automating the deployment, scaling, and management of containerized applications.

[0050] The reinforcement learning algorithm (such as Deep Q-Network) will use the sensors of the driverless vehicle itself to collect real-time traffic data and environmental information on campus, such as road congestion conditions, the positions and speeds of vehicles and pedestrians, the positions of restricted areas on campus, etc. as state inputs. Deep Q-Network (DQN) is an algorithm that combines deep learning and reinforcement learning, mainly used to solve problems with high-dimensional observation spaces.

[0051] The large language model (such as GPT-4) comprehensively analyzes factors such as the overall layout of the campus, historical traffic patterns, and the current time, weather, etc. based on its powerful language understanding and knowledge reasoning capabilities. For example, it can understand the changing patterns of the population density around teaching buildings, dormitory areas, and cafeterias at different time periods, and the impact of special events (such as sports meetings, exams, etc.) on traffic.

[0052] Based on these analysis results obtained from the large language model and the real-time information it collects itself, the reinforcement learning algorithm predicts the reward values for different path selections by continuously conducting experiments and learning in a simulated environment or during the actual delivery process. The setting of the reward value is related to factors such as delivery time, energy consumption, and whether to avoid congested and restricted areas. Through a large amount of training and iteration, the autonomous vehicle can autonomously learn the optimal path decision-making strategy in different situations, and thus quickly plan the optimal path from the starting point to the destination based on the current state information during the actual delivery task to achieve efficient delivery. The specific process of the reinforcement learning algorithm for planning the optimal path is as follows:

[0053] Step 1, State Space Definition and Information Acquisition: The reinforcement learning algorithm of the autonomous vehicle first defines the state space, which contains multi-dimensional information. Real-time traffic data is collected through on-vehicle sensors, such as the traffic flow on the road section (the number of vehicles passing through per unit time obtained through the vehicle networking), the vehicle speed distribution (distinguishing the vehicle speed ranges in different lanes), and the pedestrian density (statistical number of pedestrians per unit area using camera image recognition technology); environmental information includes the slope of the road (measured by an on-vehicle slope sensor), weather conditions (obtained by docking with meteorological department data, such as sunny, rainy, windy, etc.), and the locations of restricted areas on campus (such as the boundary coordinates of non-passable areas like playgrounds and gardens). The large language model conducts a comprehensive analysis based on the overall layout of the campus (such as the locations of buildings, the distribution of teaching buildings and dormitory areas), historical traffic patterns (analyzing the traffic flow change trends in different time periods and different regions over a past period), and the current time (accurate to minutes, with significant differences in the activities and traffic flow of people on campus at different time periods), weather, and other factors. For example, the large language model can understand that during the break time between classes, the pedestrian and vehicle flows on the roads around the teaching buildings will increase significantly, and in rainy weather, students are more inclined to move in the dormitory area, resulting in an increased probability of congestion on the roads around the dormitory area, etc.

[0054] Step 2, Action Space Setting: The action space is the set of actions that the autonomous vehicle can choose during the path planning process, including selecting different driving directions (such as turning left, turning right, going straight, making a U-turn, etc.), changing lanes (on multi-lane roads), and adjusting the driving speed (within the range permitted by safety and traffic rules).

[0055] Step 3, Reward Function Design: The reward function is used to guide the autonomous vehicle to learn the optimal path decision-making strategy. If the path selected by the autonomous vehicle can reduce the delivery time, lower the energy consumption, and successfully avoid congested sections and restricted areas, a positive reward is given; conversely, if the selected path leads to an increase in delivery time, an increase in energy consumption, or entering a congested or restricted area, a negative reward is given.

[0056] Step 4, Policy Learning and Iterative Optimization: The reinforcement learning algorithm (such as Deep Q-Network) takes the current state of the driverless vehicle (information in the state space) as input, and predicts the Q-values (estimations of the long-term cumulative rewards of the action in the current state) corresponding to different actions (actions in the action space) through a neural network. Select the action with the largest Q-value as the action decision of the driverless vehicle and execute this action during the actual delivery process. After executing the action, the driverless vehicle enters a new state. According to the difference between the new state and the expected state and the actual rewards obtained, update the parameters of the neural network through the backpropagation algorithm to adjust the prediction of the Q-value. This process is repeated continuously, and a large number of experiments and learning are carried out in the simulation environment (using historical data to construct a virtual campus scenario for training) and the actual delivery process. As the number of training increases, the driverless vehicle gradually learns the optimal path decision-making strategy in different situations and can quickly plan the optimal path from the starting point to the destination based on the current state information. For example, after multiple trainings, during the peak class time, the driverless vehicle can automatically avoid the congested roads around the teaching buildings and choose relatively unobstructed campus paths to drive in order to achieve efficient delivery.

[0057] The drone scheduling and control service module includes drone adaptive path adjustment for scheduling the handover of the drone to the driverless vehicle and continuing to execute the delivery task.

[0058] In this embodiment, when building a drone delivery system, multi-faceted collaborative operations are carried out to ensure efficient delivery. The cloud-native architecture (such as AWS Lambda or Azure Functions) is used to achieve a serverless deployment of the drone scheduling service, automatically scaling resources on demand and reducing costs. It is paired with open-source flight control systems such as PX4 (PX4 is an open-source autopilot flight stack that runs on the NuttX real-time operating system and is widely used in the autonomous control of drones and other unmanned vehicles) or ArduPilot (ArduPilot is an open-source autopilot system that can control various types of unmanned vehicles, including multi-rotor aircraft, traditional helicopters, fixed-wing aircraft, ground vehicles, submersibles, and antenna trackers. It provides a comprehensive set of tools suitable for almost any type of vehicle and application scenario) to ensure stable flight. AWS Lambda is a serverless computing service provided by Amazon Web Services that allows developers to run code in response to events without managing servers. AWS Lambda automatically manages the underlying computing resources, including the startup, scaling, and stopping of servers, enabling developers to focus on writing code without worrying about infrastructure maintenance. Azure Functions is a serverless computing service on the Microsoft Azure platform that allows developers to run code in response to events without managing servers. It provides comprehensive event-driven triggers and bindings, enabling functions to be connected to other services without writing additional code. When receiving a delivery order handover request from a delivery vehicle, the system first selects the most suitable drone to perform the delivery task based on the location, battery level, load, and target floor information of the drone. Then, in combination with the 3D building map and environmental information such as real-time wind direction and speed, it plans the optimal flight path from the delivery vehicle's location to the target floor. When planning the path, it uses its own positioning system and the 3D campus building map to calculate the precise position and height difference to avoid obstacles and no-fly zones. During the flight, the drone captures images of the surrounding environment using the on-board camera. First, computer vision technology (such as OpenCV) is used to denoise the images, enhance the contrast, and improve the quality. After that, deep learning algorithms (such as YOLOv5) identify landing targets such as feature markers of the target floor. The flight control system then controls the drone to gradually approach and land precisely based on this. During this period, computer vision and deep learning algorithms continuously monitor and correct to ensure the accuracy and safety of the landing. OpenCV (Open Source Computer Vision Library) is an open-source computer vision and machine learning software library widely used in fields such as image and video processing, object detection and recognition, feature extraction, image segmentation, and machine learning. YOLOv5 is an efficient object detection algorithm and is the fifth version of the YOLO (You Only Look Once) series.It is developed by the Ultralytics team and implemented based on the PyTorch framework. While maintaining high performance, YOLOv5 further reduces the complexity of the model, making it easier to deploy. Relying on 5G communication technology, high-speed and low-latency communication is achieved between the drone, the ground control center, and the unmanned vehicle. The flight states of the drone, such as its position, altitude, speed, battery level, and faults, are monitored in real time, and the data is immediately fed back to the drone scheduling and control service module and the front-end application service module to flexibly adjust the delivery plan according to the actual situation.

[0059] Specifically, the specific process of using computer vision technology and deep learning algorithms for object detection and positioning is as follows:

[0060] Step 1: Image acquisition and preprocessing: During the flight of the drone, the high-definition camera on board continuously acquires image data of the surrounding environment at a fixed frame rate. The acquired images first enter the preprocessing process of computer vision technology (such as OpenCV). In the grayscale processing stage, the color image is converted into a grayscale image to reduce the amount of data and computational complexity for subsequent processing. Then, a noise removal operation is performed, and the Gaussian filtering algorithm is used to smooth the image, effectively suppressing Gaussian noise in the image and making the image clearer. Subsequently, histogram equalization is used to enhance the contrast of the image, stretching the range of grayscale values of the image so that the target object (such as the feature sign of the target floor) is more prominent in the image, improving the accuracy of subsequent object detection.

[0061] Step 2: Training of the object detection model: Before performing object detection, deep learning algorithms (such as YOLOv5) need to be trained using a large amount of labeled data. An image dataset containing the feature signs of the target floor (such as specific building appearance patterns, floor signs, balcony features, etc.) is collected, and the targets in each image are accurately labeled. The labeling content includes the category of the target (such as the target floor) and the position (represented in the form of a bounding box, recording the upper left and lower right coordinates of the bounding box). The labeled dataset is divided into a training set, a validation set, and a test set. During the training process, the YOLOv5 model automatically learns the feature representation of the targets in the image through multiple convolutional layers and pooling layers. The convolutional layer slides the convolutional kernels of different sizes on the image for convolution to extract local features of the image, such as edges and textures; the pooling layer downsamples the output of the convolutional layer, reducing the amount of data while retaining key features. After multiple iterations of training, the model gradually learns the unique patterns of the feature signs of the target floor and can accurately identify whether there is a target floor in the image, as well as the position and category of the target.

[0062] Step 3, Object Detection and Localization: In practical applications, the preprocessed image is input into the trained YOLOv5 model. The YOLOv5 model quickly scans the image and classifies and predicts the bounding boxes for each region that may contain the object. If the feature signs of the target floor are detected, the model outputs the position (bounding box coordinates) of the target in the image and the corresponding confidence score (indicating the credibility of the detection result by the model, with a value range of 0 - 1). At the same time, combining the latitude and longitude, altitude information obtained from the UAV's own positioning system (such as GPS, inertial navigation system, etc.) and the pre-stored three-dimensional map information of campus buildings, the actual spatial position and orientation of the target floor relative to the UAV are calculated. For example, through the position of the target in the image, the attitude of the UAV (heading angle, pitch angle, roll angle), and the distance between the UAV and the target floor (calculated based on the positioning system and map information), the position of the target floor in the geographical coordinate system can be accurately determined, providing an accurate basis for adjusting the flight path of the UAV.

[0063] Step 4, Dynamic Tracking and Adjustment: During the process of the UAV flying towards the target floor, computer vision technology and deep learning algorithms continue to work to dynamically track the target. Due to the movement of the UAV and the influence of environmental factors (such as light changes, occlusions, etc.), the position and appearance of the target in the image may change. To ensure that the UAV can accurately reach the target floor, the system continuously updates the collected image data and re-performs object detection and localization. When the detected target position deviates, according to the deviation amount and the current state of the UAV, the flight control system adjusts the flight attitude and speed of the UAV so that the UAV always flies towards the target floor. For example, if it is found that the position of the target floor in the image deviates to the left, the flight control system controls the UAV to slightly adjust the heading to the left to maintain an accurate pointing at the target floor and finally achieve a precise landing.

[0064] The communication service module is used to implement the communication between the front-end application service module, the unmanned vehicle scheduling and control service module, and the unmanned aerial vehicle scheduling and control service module. The unmanned aerial vehicle scheduling and control service module is connected to the unmanned vehicle scheduling and control service module through 5G communication, enabling the unmanned aerial vehicle to transmit real-time operation status information such as its own battery power, component self-check status, and current location to the unmanned aerial vehicle scheduling and control service module. After receiving the information, the unmanned aerial vehicle scheduling and control service module issues a take-off command to the unmanned aerial vehicle according to the task requirements and overall scheduling arrangements, and simultaneously sends key data such as the location of the unmanned vehicle and the express delivery handover requirements, so that the unmanned aerial vehicle can clearly understand the subsequent delivery process in advance. The unmanned aerial vehicle continuously collects status data such as flight altitude, speed, heading, and ambient temperature and humidity of the surrounding environment using various sensors, and transmits it back to the unmanned aerial vehicle scheduling and control service module with low latency through the 5G network. The unmanned aerial vehicle scheduling and control service module uses professional algorithms to evaluate flight safety based on this. Once potential risks such as strong airflow interference or approaching a no-fly zone are detected, an adjustment command is immediately sent to the unmanned aerial vehicle, and the unmanned aerial vehicle quickly responds to adjust its flight attitude or heading to avoid danger. When the unmanned aerial vehicle approaches the unmanned vehicle, the unmanned aerial vehicle scheduling and control service module and the unmanned vehicle scheduling and control service module are accurately docked through 5G communication. The unmanned aerial vehicle sends its own location and landing requirements, and the unmanned vehicle feeds back information on the preparation of the express delivery order and the surrounding docking environment to ensure the smooth landing of the unmanned aerial vehicle to hand over the express delivery order. After the handover is completed, the unmanned aerial vehicle pushes information such as the delivered express delivery order, the estimated delivery route, and the remaining battery power to the unmanned aerial vehicle scheduling and control service module. The unmanned aerial vehicle scheduling and control service module optimizes the subsequent task allocation based on this and real-time tracks the delivery progress. When approaching the target floor, the ground control center provides landing guidance based on the three-dimensional map of the building, and the unmanned aerial vehicle combines its own vision and deep learning algorithm feedback to complete precise landing, and the key node information is pushed to the front-end application service module in real time for the user to easily grasp the delivery dynamics.

[0065] The big data storage and analysis service module is used to store and analyze the delivery data to implement a dynamic adjustment mechanism for fluctuations in the express delivery volume. The delivery data includes express delivery order information, the operation status of the unmanned vehicle, the operation status of the unmanned aerial vehicle, and the traffic conditions; the dynamic adjustment mechanism monitors the delivery data and uses data mining and machine learning algorithms for analysis to predict the express delivery order volume. When the volume fluctuation exceeds ±20%, the scheduling, task allocation optimization, and dynamic adjustment of the working area of the unmanned vehicle and the unmanned aerial vehicle are carried out according to the prediction results.

[0066] In this embodiment, a distributed file system is selected, such as Ceph (Ceph is an open source distributed storage system that aims to provide a high-performance, high-scalability and high-reliability unified storage solution for modern data centers. Ceph can simultaneously support block storage, object storage and file storage through a unified storage platform, and is widely used in cloud computing and big data processing fields) or GlusterFS (GlusterFS is an open source distributed file system that aims to provide high-performance, high-scalability and high-reliability storage solutions for large-scale data centers through software-defined storage) to classify and store massive express order information, unmanned vehicle operating status, drone operating status and traffic conditions to ensure their high availability and durability.

[0067] Specifically, the big data storage and analysis service module regularly analyzes historical data in depth to mine key information such as user behavior patterns, traffic conditions, and the operating status of unmanned vehicles and drones. For example, it analyzes the changing trends of express delivery business volume in different time periods (such as the start of school, shopping festivals, daily breaks and after-school breaks), the order distribution patterns in different areas (dormitory areas, teaching areas, etc.), and the delivery efficiency and resource utilization of unmanned vehicles and drones, providing data basis for the unmanned vehicle dispatching and control service module and the drone dispatching and control service module.

[0068] Specifically, the big data storage and analysis service module uses data mining and machine learning algorithm libraries, such as Scikit-learn (Scikit-learn is an open source machine learning library for Python programming language. It is built on scientific computing libraries such as NumPy, SciPy and matplotlib, and provides simple and effective data mining and data analysis tools) and TensorFlow (TensorFlow is an open source machine learning framework developed by the Google Brain team, widely used to build and train various machine learning models, especially deep neural networks) to clean, standardize and normalize various types of data, and carry out feature engineering. After extracting features such as order weight, road conditions, and user ordering behavior, the module selects appropriate neural network algorithms (such as long short-term memory networks (LSTM) or gated recurrent units (GRU)) to build delivery time prediction models, traffic flow prediction models, etc. Based on these models, the business volume is predicted, and more unmanned vehicles and drone resources are deployed in advance before the peak of business volume. For example, if it is predicted that the volume of express delivery will increase significantly within a week after the start of school, the system will dispatch idle unmanned vehicles from the equipment hosting area to the express centralized storage point in advance, and arrange for drones to charge batteries and perform equipment maintenance inspections to ensure sufficient capacity to cope with the peak. The process of data mining and machine learning algorithm analysis is as follows:

[0069] Step 1: Data Collection and Cleaning: The big data storage and analysis service module collects express order information (including order number, order placement time, recipient information, recipient address, express weight, express volume, express type, scheduled delivery time, order status, etc.), unmanned vehicle operation status data (such as location, speed, battery level, load, driving mileage, fault records, etc.), unmanned aerial vehicle operation status data (location, altitude, speed, battery level, flight attitude, fault records, etc.), and traffic condition data (road traffic flow, congestion situation, impact of weather on traffic, impact of special events on traffic, etc.) from various data sources of the system. During the data collection process, problems such as missing data, data errors, and duplicate data may occur. To address these issues, data cleaning techniques are used. For missing data, if the missing proportion is small, it can be filled using the mean, median, or machine learning algorithms (such as the K-nearest neighbor algorithm); if the missing proportion is large, the corresponding data records are deleted. For incorrect data, it is identified and corrected by setting reasonable data ranges and logical rules (such as the order of change of the order status should conform to the business process). For duplicate data, duplicate records are removed through a data deduplication algorithm (such as the deduplication method based on the hash function) to ensure the accuracy and integrity of the data.

[0070] Step 2: Data Preprocessing and Feature Engineering: The cleaned data is preprocessed, including data standardization and normalization. Z-score standardization method is used for data standardization, adjusting the mean of the data to 0 and the standard deviation to 1, so that data with different features have the same dimension, facilitating subsequent machine learning algorithm processing. Data normalization maps the data to the [0,1] interval, and common methods include min-max normalization. In the feature engineering stage, features meaningful for analysis and prediction are extracted from the original data. For predicting the business volume of express orders, the extracted features may include time features (such as day of the week, time period, holidays, etc.), campus activity features (such as the start of school season, exam week, campus sports meeting, etc.), weather features (such as temperature, precipitation, wind power, etc.), and historical order volume features (such as the change trend of order volume in the past week or month). For optimizing the scheduling of unmanned vehicles and unmanned aerial vehicles, the extracted features may include the performance features of vehicles and unmanned aerial vehicles (such as maximum load, maximum speed, endurance mileage, etc.), location features (such as the distance between the current location and the express concentration area, the order density in the area where it is located, etc.), and traffic condition features (such as road congestion index, average driving speed, etc.). Through feature combination and transformation, feature vectors that can better reflect the internal laws of the data are generated. For example, time features and historical order volume features are combined to generate time series features for predicting the future change trend of order volume.

[0071] Step 3, Model Selection and Training: According to the specific analysis task, select appropriate algorithms and model architectures from data mining and machine learning algorithm libraries (such as Scikit-learn, TensorFlow). For predicting the business volume of express delivery orders, select a time series prediction model, such as Long Short-Term Memory Network (LSTM) or Gated Recurrent Unit (GRU). These models can effectively handle the long-term dependencies in time series data and capture the changing trends of order volumes. Taking LSTM as an example, use the preprocessed and feature-engineered time series data (such as the order volumes at different time periods every day in the past month) as input, set the hyperparameters of the model (such as the number of hidden layer nodes, the number of layers, the learning rate, etc.), and minimize the loss function (such as the mean squared error loss function) between the predicted value and the actual value through the backpropagation algorithm to train the model. During the training process, use the validation set to evaluate the model to prevent overfitting. If the performance of the model on the validation set no longer improves, stop the training. For optimizing the scheduling of unmanned vehicles and drones, adopt a reinforcement learning algorithm (such as Deep Deterministic Policy Gradient algorithm DDPG). The reinforcement learning algorithm learns the optimal scheduling strategy through the interaction between the agent (unmanned vehicle or drone) and the environment (campus delivery scenario). During the training process, continuously adjust the parameters of the model according to the rewards (such as improved delivery efficiency, reduced costs, etc.) obtained from different scheduling decisions (such as which vehicle to choose to execute an order, which path to plan, etc.), so that the agent gradually learns the optimal strategy.

[0072] Step 4, Model Evaluation and Application: Use the test set to evaluate the trained model and calculate evaluation metrics to measure the performance of the model. For prediction models, common evaluation metrics include Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Coefficient of Determination (R 2 ) etc. For models that optimize scheduling, the evaluation metrics can be the percentage increase in delivery efficiency, the percentage reduction in cost, the order completion rate, etc. If the evaluation metrics of the model meet the expected requirements, deploy the model to the actual system for application. During the actual application process, continuously collect new data and regularly update and retrain the model to adapt to the changes in data distribution and the adjustment of business requirements. For example, with the seasonal changes in the express delivery business volume on campus or the emergence of new express delivery service demands, retrain the model so that the model can more accurately predict the business volume and optimize the scheduling strategy to ensure the efficient operation of campus express delivery.

[0073] Specifically, during peak business hours, the system quickly adjusts the task allocation strategy based on the real-time express order volume and the operating status of the unmanned vehicles. When calculating the comprehensive cost of each unmanned vehicle to execute order tasks, more emphasis is placed on considering the urgency of orders in the current area and the vehicle's load capacity. For example, for urgent and light-weight express orders, they are preferentially assigned to unmanned vehicles that are close to the shipping point, have sufficient power, and have a light load; for batch orders with slightly lower timeliness requirements, they are assigned to large unmanned vehicles or coordinated with drones for collaborative distribution. At the same time, urgent orders are processed first to ensure that key materials or urgently needed express deliveries can be delivered in a timely manner.

[0074] Specifically, the system dynamically adjusts the working areas of unmanned vehicles and drones based on real-time data. By analyzing the order distribution in each area, resources are concentrated in areas with greater express delivery demand. For example, during the break between classes, the order volume in the dormitory area increases, and the system guides the unmanned vehicles to transfer from the teaching area or other idle areas to the dormitory area and re-plans the delivery routes; if there is a sudden increase in express delivery demand due to an event held in a certain teaching area, the surrounding unmanned vehicles and drones are promptly deployed for service. Taking advantage of the characteristics that campus express delivery recipients are relatively concentrated in the dormitory area and the teaching area, the resource utilization rate is improved, waste of distribution resources is reduced, and efficient and orderly campus express delivery is ensured.

[0075] Embodiment 2: Based on the system in Embodiment 1, this embodiment provides a method for using a campus express air-ground integrated unmanned distribution system, including the following steps:

[0076] Step 1: The user registers, logs in, and places an order through the front-end application service module, and the system express platform integrated service module obtains order information from the express platform;

[0077] Step 2: According to the order information and the operating status of the unmanned vehicle, the unmanned vehicle scheduling and control module schedules the unmanned vehicle to execute the delivery task according to the self-adaptive path adjustment of the unmanned vehicle to plan the path;

[0078] Step 3: The drone scheduling and control service module schedules the drone for express order handover, and the drone adjusts the planned flight path according to the self-adaptive path of the drone and executes the delivery;

[0079] Step 4: Through the communication service module, communication is achieved between the front-end application service module, the unmanned vehicle scheduling and control service module, and the drone scheduling and control service module, and the delivery status is monitored in real time and pushed to the front-end application service module;

[0080] Step 5: The big data storage and analysis service module uses the dynamic adjustment mechanism for dynamic adjustment to optimize the unmanned vehicle scheduling and control service module and the drone scheduling and control service module.

[0081] Air-ground integrated logistics distribution has significant advantages in the campus scenario and can effectively meet the complex needs of campus express delivery.

[0082] 1. Campus express delivery: With a dense population, unique building layouts and road planning, and large fluctuations in express delivery volume on campus, air-ground integrated distribution can give full play to the advantages of drones and unmanned vehicles. For example, during peak student activity periods such as between classes or during lunch breaks, drones can avoid crowds and quickly deliver express packages to the rooftops of teaching buildings or open spaces, and then unmanned vehicles can transport them short distances to each classroom or office. In the dormitory area, unmanned vehicles can transport express packages in large quantities to the downstairs, and drones can assist in delivering them to high-rise dormitories, improving the delivery efficiency and solving the delivery problems during peak express delivery volumes.

[0083] 2. Campus material transportation: For tasks such as the allocation of a large number of books in the school library, the handling of equipment in laboratories, and the supply of food ingredients in the cafeteria, the air-ground integrated system can plan the transportation method according to the weight, volume, and urgency of the materials. Urgent small experimental reagents can be quickly transported by drones, while a large number of books or food ingredients can be transported in batches by unmanned vehicles to ensure the timely supply of materials and optimize the internal logistics process on campus.

[0084] 3. Campus emergency rescue: In case of emergencies such as fires or earthquakes on campus, the air-ground integrated distribution system can be used to transport emergency medical supplies, small rescue equipment, etc. to the rescue site or temporary shelter area, buying time for rescue work and ensuring the safety of teachers and students. For example, when a fire breaks out, drones can quickly reach near the fire floor with small equipment such as fire extinguishers or breathing masks to provide support for initial rescue.

[0085] The present invention gives full play to the advantages of both through the collaborative operation of unmanned vehicles and drones. Unmanned vehicles can carry out large-scale ground delivery, while drones can quickly reach some special locations, such as high-rise dormitories or the rooftops of teaching buildings. In the case of a dense population, regular building layouts, and restricted areas on campus, the system can dynamically adjust the delivery route according to real-time road conditions and environmental information, avoiding congestion and restricted areas, thus significantly reducing the delivery time. The present invention dynamically adjusts the delivery routes of unmanned vehicles and drones through intelligent scheduling of unmanned vehicles, adaptive path adjustment of unmanned vehicles, and adaptive path adjustment of drones, avoiding congestion and restricted areas, and efficiently completing the delivery task. The present invention accurately predicts traffic flow through the big data storage and analysis service module, providing a reliable basis for delivery route planning and enhancing the adaptability of the system in complex traffic scenarios. In addition, the unmanned vehicle scheduling of the present invention comprehensively considers various factors to calculate the task cost, can dynamically adjust tasks and vehicle allocation, and respectively uses reinforcement learning algorithms, large language models, computer vision technologies, and deep learning algorithms, etc., combined with real-time campus information and environmental factors, to achieve autonomous adaptive optimal path planning and decision-making.

[0086] In summary, the present invention can achieve the adaptive path adjustment of driverless vehicles and drones, providing an efficient, safe and flexible integrated air-ground distribution service for campus express delivery.

Claims

1. A campus express delivery air-ground integrated unmanned delivery system, characterized by: It includes a front-end application service module, an express platform integrated service module, an unmanned vehicle dispatching and control service module, an unmanned aerial vehicle dispatching and control service module, a communication service module and a big data storage and analysis service module; the front-end application service module is respectively connected with the express platform integrated service module, the unmanned vehicle dispatching and control service module, the unmanned aerial vehicle dispatching and control service module and the big data storage and analysis service module for data exchange; the express platform integrated service module is connected with the unmanned vehicle dispatching and control service module and transmits express-related data; the unmanned vehicle dispatching and control service module and the unmanned aerial vehicle dispatching and control service module communicate and work together through the communication service module; the big data storage and analysis service module provides data support for the front-end application service module, the unmanned vehicle dispatching and control service module and the unmanned aerial vehicle dispatching and control service module; Wherein: the front-end application service module is used for user registration, login, order placement, order status query and notification reception; The express platform integrated service module is used to connect with the express service platform, obtain express information and perform address matching, and synchronize express status information in real time; The unmanned vehicle dispatching and control service module includes unmanned vehicle intelligent dispatching and unmanned vehicle adaptive path adjustment, which is used to dispatch unmanned vehicles to perform delivery tasks according to the acquired express information and the matched address; The drone dispatching and control service module includes drone adaptive path adjustment, which is used to dispatch drones to take over from unmanned vehicles and continue to perform delivery tasks; The communication service module is used to realize the communication between the front-end application service module, the unmanned vehicle dispatching and control service module and the unmanned vehicle dispatching and control service module; The big data storage and analysis service module is used to store and analyze delivery data to implement a dynamic adjustment mechanism for express delivery business volume fluctuations.

2. The campus express air-ground integrated unmanned delivery system according to claim 1 is characterized by: The unmanned vehicle intelligent scheduling is responsible for receiving express order information and obtaining the operating status of all unmanned vehicles. According to the express order information and the operating status of the unmanned vehicles, an intelligent scheduling algorithm is used to initialize order allocation and execute order delivery. According to the operating status of the unmanned vehicle and road conditions, order allocation and unmanned vehicle scheduling are dynamically adjusted.

3. The campus express air-ground integrated unmanned delivery system according to claim 2 is characterized by: The specific process of the intelligent scheduling algorithm is to collect detailed information on express orders, unmanned vehicle operating status information and road condition information, traverse all unmanned vehicles, build a comprehensive cost function, and calculate the comprehensive cost of each unmanned vehicle executing an order through weighted summation; the dynamic adjustment process is to monitor the operating status and road conditions of the unmanned vehicle in real time during the execution of the task, and reallocate its unfinished orders when the unmanned vehicle fails, the battery is too low or there is severe congestion; at the same time, regularly check the regional order distribution, adjust the unmanned vehicle resources and driving routes, give priority to emergency orders, and ensure the efficient operation of the system.

4. The campus express air-ground integrated unmanned delivery system according to claim 1 is characterized by: The self-adaptive path adjustment of the unmanned vehicle is based on real-time traffic conditions and real-time environmental information. The real-time traffic conditions and real-time environmental information are analyzed and predicted through the existing large language model, and then the optimal path is planned using the reinforcement learning algorithm.

5. The campus express air-ground integrated unmanned delivery system according to claim 4 is characterized by: The specific process of the reinforcement learning algorithm for planning the optimal path is to define a state space containing multi-dimensional information, collect real-time traffic data, environmental information, campus layout, historical traffic patterns and the current time; set a set of actions that the unmanned vehicle can choose in path planning, including driving direction, lane change and speed adjustment; design a reward function to guide the unmanned vehicle to learn the optimal path decision; use the current state of the unmanned vehicle as input, predict the estimated value of the long-term cumulative reward of different actions, and select the action with the largest estimated value to execute, update the neural network parameters through actual execution and feedback, optimize the estimated value prediction, so as to realize the real-time state to quickly plan the optimal path.

6. The campus express air-ground integrated unmanned delivery system according to claim 1 is characterized by: The drone adaptive path adjustment is responsible for receiving express order information delivered by unmanned vehicles and obtaining the operating status of the drone. According to the express order information and the operating status of the drone, the drone is assigned to take over the express order and execute the order delivery. Then, the optimal flight path is planned for the drone in combination with the three-dimensional map of the building and the real-time environmental information. Computer vision technology and deep learning algorithms are used for target detection and positioning to assist in precise landing. During the working process, the operating status of the drone is monitored in real time, and the operating status of the drone is fed back to the drone scheduling and control service module and the front-end application service module to flexibly adjust the order delivery and push delivery notifications.

7. The campus express air-ground integrated unmanned delivery system according to claim 6 is characterized by: The specific process of using computer vision technology and deep learning algorithms to detect and locate targets is to collect and preprocess image data of the surrounding environment; Establish a target detection model, use a deep learning algorithm to train the target detection model, input the preprocessed image into the trained target detection model, detect the position and confidence of the characteristic marks of the target floor, combine the UAV's positioning system and three-dimensional map information, calculate the actual spatial position of the target floor, and provide a basis for adjusting the flight path. During the flight of the UAV, continuously collect images and perform target detection and positioning; according to the changes in the target position and the status of the UAV, adjust the flight attitude and speed through the flight control system to ensure that the UAV always flies towards the target floor and finally achieves precise landing.

8. The campus express air-ground integrated unmanned delivery system according to claim 1 is characterized by: The distribution data includes express order information, unmanned vehicle operating status, drone operating status and traffic conditions; the dynamic adjustment mechanism monitors the distribution data and uses data mining and machine learning algorithms to analyze it, so as to predict the express order business volume, and dispatch unmanned vehicles and drones, optimize task allocation and dynamically adjust work areas based on the prediction results.

9. The campus express air-ground integrated unmanned delivery system according to claim 8 is characterized by: The process of data mining and machine learning algorithm analysis is to collect express order information, unmanned vehicle and drone operating status data, and traffic condition data, and use data cleaning technology to address data missing, errors, and duplication problems; standardize and normalize the cleaned data, extract features from the original data to predict express business volume and optimize scheduling; for express order business volume prediction, use a time series prediction model, adjust hyperparameters through training data, and minimize prediction errors; evaluate the performance of the time series prediction model and calculate evaluation indicators.

10. The method for using the campus express air-ground integrated unmanned delivery system according to any one of claims 1 to 9, characterized in that: The following steps are involved: Step 1: The user registers, logs in and places an order through the front-end application service module, and the system express platform integration service module obtains the order information from the express platform; Step 2: According to the order information and the operating status of the unmanned vehicle, the unmanned vehicle dispatches and controls the unmanned vehicle to perform the delivery task according to the unmanned vehicle adaptive path adjustment planning path; Step 3: The drone dispatching and control service module dispatches the drone to deliver the express order. The drone adjusts the flight path according to the drone adaptive path and executes the delivery. Step 4: Realize the communication between the front-end application service module, the unmanned vehicle dispatching and control service module, and the drone dispatching and control service module through the communication service module, monitor the delivery status in real time and push it to the front-end application service module; Step 5: Use the big data storage and analysis service module to use the dynamic adjustment mechanism to dynamically adjust and optimize the unmanned vehicle scheduling and control service module and the drone scheduling and control service module.

Images