A driving method and system for unmanned vending vehicles that combine community delivery and mobile vending functions

Abstract

This invention provides a driving method and system for unmanned vending vehicles that combine community delivery and mobile vending functions. The invention centers on "dynamic function switching" and constructs a five-layer architecture of "perception-decision-execution-interaction-energy" to achieve seamless switching between delivery and vending functions. Through dynamic function switching, intelligent algorithm collaboration, and other technologies, it achieves deep integration of delivery and vending functions, improving equipment utilization and operational revenue.

Description

Technical Field

[0001] This invention relates to the field of unmanned retail technology, and in particular to an unmanned vending vehicle driving method, an unmanned vending vehicle driving system, electronic equipment, and computer-readable storage medium that combine community delivery and mobile sales functions. Background Technology

[0002] With the development of unmanned technology, unmanned delivery vehicles and unmanned vending vehicles have been gradually applied in communities, parks and other scenarios, but existing technologies have significant limitations:

[0003] 1. Limited functionality: Traditional unmanned delivery vehicles focus solely on delivery tasks, with equipment utilization rates below 30% during idle periods (such as off-peak delivery times); traditional unmanned vending vehicles can only sell at fixed locations or through simple mobile operations, failing to respond to sudden delivery demands, resulting in resource waste.

[0004] 2. Rigid scheduling decisions: Existing systems mostly use fixed rules for scheduling (such as order-based delivery) without taking into account real-time pedestrian flow and traffic conditions to optimize task priorities, resulting in low delivery efficiency or failure to maximize sales revenue.

[0005] 3. Static route planning: Traditional route planning only considers the shortest distance and does not incorporate future pedestrian flow predictions (such as in sales scenarios) or dynamic traffic changes (such as in delivery scenarios), thus failing to balance efficiency and effectiveness.

[0006] 4. Fragmented application of algorithms: In the existing system, algorithms such as sensor fusion, path planning, and pedestrian flow prediction run independently without forming a collaborative optimization mechanism, resulting in decision lag or insufficient accuracy. Summary of the Invention

[0007] To address the technical problems existing in the prior art, the present invention provides the following technical solution:

[0008] On the one hand, a driving system for unmanned vending vehicles that combines community delivery and mobile vending functions is provided, adopting a five-layer architecture of perception-decision-execution-interaction-energy, including:

[0009] The perception layer, composed of the fusion of multiple sensors, is used to acquire vehicle status, environmental and task data;

[0010] The decision-making layer, based on data from the perception layer, enables task scheduling, dynamic function switching, and path planning.

[0011] The execution layer, based on instructions from the decision-making layer, controls vehicles to complete delivery, sales, or charging actions;

[0012] The interaction layer provides operation and monitoring interfaces for users and operations personnel.

[0013] The energy layer provides power to the system and manages energy.

[0014] The decision-making layer is configured to dynamically switch between delivery mode, sales mode, or charging mode based on delivery task queues, vehicle battery level, and external environmental information.

[0015] Preferably, the decision-making layer includes an intelligent task scheduling module, which uses a PPO algorithm based on deep reinforcement learning for decision-making, wherein:

[0016] The state space includes the delivery task queue (T), vehicle location (P), battery status (B), traffic conditions (C), and pedestrian density (H);

[0017] The action space includes receiving delivery tasks (a1), switching to sales mode (a2), and heading to a charging point (a3);

[0018] The reward function is R = α·R_del + β·R_sell - γ·C_bat - δ·C_time, where R_del is the delivery revenue, R_sell is the sales revenue, C_bat is the battery consumption cost, C_time is the time cost, and α, β, γ, and δ are weighting coefficients.

[0019] Preferably, the decision layer includes a dynamic path planning module, which uses the D Lite algorithm for real-time path planning and integrates the pedestrian density prediction results based on the LSTM model; the input features of the LSTM model include historical pedestrian density, weather data and time features, and the output is the predicted pedestrian density value of the target area for future periods.

[0020] Preferably, the execution layer includes:

[0021] The delivery box system has multiple compartments with electronic locks, each equipped with an RFID reader and a weight sensor for cargo identification and weight verification;

[0022] The vending machine system has multiple compartments with electronic locks, each equipped with an infrared sensor and a temperature sensor, and integrates an intelligent recommendation function based on a collaborative filtering algorithm;

[0023] The charging management system uses an extended Kalman filter algorithm to estimate the battery's state of charge (SOC) in real time and triggers the charging process when the SOC falls below a threshold.

[0024] Preferably, the energy layer includes an energy recovery system and a solar-assisted charging system; the energy recovery system converts kinetic energy into electrical energy and charges the battery when the vehicle brakes or decelerates; the solar-assisted charging system stores the electrical energy generated by the solar panels into the battery through an MPPT controller.

[0025] On the other hand, a method for driving an unmanned vending vehicle based on the system described above is provided, including the following steps:

[0026] Perception steps: Acquire vehicle status, external environment, and task-related data through multi-sensor fusion;

[0027] Decision-making steps: Based on the perceived data, intelligent algorithms are used to perform task scheduling, functional mode decision-making, and path planning;

[0028] Execution steps: Based on the decision-making instructions, control the vehicle to perform delivery, sales, or charging tasks;

[0029] The system dynamically switches between delivery mode, sales mode, and charging mode based on delivery task queues, vehicle battery level, and external environmental information.

[0030] Preferably, the task scheduling in the decision-making step employs the PPO algorithm based on deep reinforcement learning, including:

[0031] Construct the state space, action space, and reward function;

[0032] The policy network outputs an action probability distribution based on the current state, and selects the action with the highest probability to execute.

[0033] The reward value is calculated based on the execution result and fed back to the model for parameter updates.

[0034] Preferably, the path planning in the decision-making step employs the D Lite algorithm combined with LSTM pedestrian flow prediction, including:

[0035] Predict the future pedestrian density of a target area using an LSTM model;

[0036] The predicted value of pedestrian flow density is converted into a benefit cost, and a comprehensive cost map is constructed by combining the time cost of traffic conditions.

[0037] Incremental path planning is performed based on the comprehensive cost map using the D Lite algorithm.

[0038] Preferably, the delivery process in the execution steps includes:

[0039] When picking up goods, the RFID reader identifies the goods tag and the weight sensor verifies the weight of the goods; when delivering goods, the user unlocks the corresponding compartment with the pickup voucher and the weight sensor detects the pickup completion status.

[0040] The sales process in the execution steps includes:

[0041] The inventory is updated by detecting the status of goods being picked up using infrared sensors.

[0042] Based on collaborative filtering algorithms, products are recommended to users according to their profiles and the current context.

[0043] Preferably, the method further includes an energy management step:

[0044] Battery SOC is estimated in real time using the extended Kalman filter algorithm;

[0045] When the SOC is lower than the first threshold, the charging mode is triggered, and a path to the charging station is planned.

[0046] When the vehicle brakes or decelerates, kinetic energy is converted into electrical energy through an energy recovery system;

[0047] A solar-assisted charging system converts solar energy into electrical energy to charge the battery.

[0048] On the other hand, an electronic device is provided, comprising: a processor; and a memory storing computer-readable instructions, which, when executed by the processor, implement any of the above-described unmanned vending vehicle driving methods that combine community delivery and mobile vending functions.

[0049] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, the at least one instruction being loaded and executed by a processor to implement any of the above-described unmanned vending vehicle driving methods that combine community delivery and mobile vending functions.

[0050] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0051] By employing technologies such as multi-sensor fusion, intelligent decision-making algorithms, and dynamic function switching, the problems of "single function and low utilization rate" of traditional unmanned vending vehicles have been solved, realizing the dual value of "delivery + sales".

[0052] With "dynamic function switching" as its core, a five-layer architecture of "perception-decision-execution-interaction-energy" is constructed to achieve seamless switching between delivery and sales functions. Through technologies such as dynamic function switching and intelligent algorithm collaboration, the deep integration of delivery and sales functions is realized, improving equipment utilization and operational revenue. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1This is a block diagram of an unmanned vending vehicle drive system that combines community delivery and mobile vending functions, provided by an embodiment of the present invention.

[0055] Figure 2 This is a data fusion flowchart provided by an embodiment of the present invention;

[0056] Figure 3 This is a schematic diagram of a decision tree decision mechanism provided in an embodiment of the present invention;

[0057] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0058] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0059] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0060] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0061] In this embodiment of the invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0062] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0063] This invention provides a method for driving unmanned vending vehicles that combines community delivery and mobile vending functions. This method can be implemented by an electronic device, which can be a terminal or a server.

[0064] To address the aforementioned shortcomings, this system utilizes technologies such as dynamic function switching and intelligent algorithm collaboration to achieve deep integration of delivery and sales functions, thereby improving equipment utilization and operational revenue.

[0065] I. System Overall Architecture Design

[0066] This system, centered on "dynamic function switching," constructs a five-layer architecture of "perception-decision-execution-interaction-energy" to achieve seamless switching between delivery and sales functions. The system architecture is as follows: Figure 1 As shown:

[0067] (a) Perception layer: environmental and state perception

[0068] The perception layer is the "eyes" of the system. It acquires data on the vehicle's own status, the external environment, and task-related information through multi-sensor fusion, providing input to the decision-making layer.

[0069] 1. Sensor Configuration List

[0070] Sensor type Model / Parameters Function Description LiDAR Velodyne VLP-16 360° environmental perception, ranging accuracy ±2cm, used for SLAM mapping, obstacle avoidance and localization. High-definition camera Hikvision DS-2CD3T47FWD-LS 4K resolution, supports YOLOv5 object detection, used for pedestrian / vehicle recognition and cargo appearance verification. GPS / IMU combination UBLOX NEO-M8N + Bosch BMI088 GPS positioning accuracy is ±1m (±2cm after RTK correction), and the IMU provides acceleration / angular velocity data for integrated navigation. ultrasonic sensor Panasonic EVM2000 The ranging range is 0.1-2m, used for close-range obstacle avoidance (such as steps and low obstacles). Weight sensor Mettler Toledo MT1260 Accuracy ±0.1%, installed at the bottom of delivery boxes and vending machines for goods weight verification and inventory monitoring. RFID reader Zebra RFD8500 Reading distance 0-10cm, used for identifying delivery tags (to confirm correct pickup / delivery). Infrared sensor Sharp GP2Y0A21YK0F Installed on the vending machine shelves to detect whether goods have been taken (to prevent missing or mistaken delivery). Battery status sensor CATL BMS-100 Monitor battery voltage (0-1000V), current (-500A-+500A), and temperature (-40℃-+85℃) for SOC estimation.

[0071] 2. Sensing Data Fusion Process

[0072] The perception layer employs multi-sensor fusion (MSF) technology to spatiotemporally synchronize and correlate point cloud data from LiDAR, image data from cameras, positioning data from GPS / IMU, and other sensor data, outputting a unified environmental model (e.g., Figure 2 (As shown). The specific process is as follows:

[0073] Data preprocessing: Denoising (statistical filtering) and downsampling (voxel mesh) of LiDAR point clouds; distortion correction (based on camera intrinsic parameters) and target detection (YOLOv5 for pedestrian / vehicle identification) of camera images; time synchronization of GPS / IMU data (based on PTP protocol).

[0074] Spatiotemporal alignment: Based on GPS time, all sensor data are mapped to the same timestamp; point cloud data and image data are fused through camera extrinsic parameters (positional relationship between LiDAR and camera) to generate point clouds with semantic information (such as "point cloud + pedestrian label").

[0075] Environmental modeling: A local map is constructed using SLAM algorithms (such as LOAM), and combined with GPS global positioning to generate an environmental model of "global map + local obstacles"; at the same time, the status of goods is obtained through weight sensors and RFID readers (such as "delivery box No. 1 has goods, weight 2.5kg"), and the inventory status of vending machines is obtained through infrared sensors (such as "vending machine A1 is empty").

[0076] (II) Decision-making level: core logic and intelligent scheduling

[0077] The decision-making layer is the "brain" of the system. Based on the data from the perception layer, it realizes three core functions: task scheduling, function switching, and path planning. It incorporates intelligent algorithms such as reinforcement learning and deep learning to optimize decision-making results.

[0078] 1. Core Decision-Making Logic of the System

[0079] The core logic of the decision-making level is based on "task priority" and "resource utilization" as objectives, constructing a system such as... Figure 3 The decision tree shown:

[0080] Triggering conditions: The system monitors the delivery task queue (from platforms such as Meituan / Eleme), the current status of vehicles (battery level, location, cargo status), and the external environment (people density, traffic conditions) in real time.

[0081] Function switching logic:

[0082] When the delivery task queue is not empty, the delivery mode is executed first;

[0083] When the delivery task queue is empty and the battery level is ≥30%, switch to sales mode;

[0084] When the battery level is less than 20%, it will force a switch to charging mode (go to the nearest charging station).

[0085] In-pattern decision-making:

[0086] Delivery mode: Optimize delivery route (genetic algorithm), dynamic obstacle avoidance (DWA algorithm);

[0087] Sales model: Select high-traffic areas (LSTM prediction) and adjust sales strategies (such as promotional recommendations);

[0088] Charging mode: Plan the shortest charging path (A algorithm), predict charging time (based on battery model).

[0089] 2. Intelligent Task Scheduling Model (Optimized by Reinforcement Learning)

[0090] To address the problem of "dynamic balance between delivery tasks and sales opportunities," a task scheduling model is constructed using deep reinforcement learning (DRL) to maximize vehicle operating revenue.

[0091] (1) Model definition

[0092] State space:

[0093]

[0094] in:

[0095] (T): Delivery task queue (number of tasks, average distance, time requirement);

[0096] (P): Vehicle's current location (latitude and longitude);

[0097] (B): Battery status (SOC, remaining battery life);

[0098] (C): Real-time traffic conditions (current road congestion index, 0-10);

[0099] (H): Target area population density (predicted value, people / 100㎡).

[0100] Action space:

[0101]

[0102] in:

[0103] Receive new delivery assignments;

[0104] Switch to sales mode (go to high-traffic areas);

[0105] Head to a charging point (when battery is low).

[0106] Reward function:

[0107] The reward function is designed with the goals of "maximizing revenue" and "minimizing resource consumption", and the formula is as follows:

[0108]

[0109] in:

[0110] Delivery task revenue (RMB);

[0111] Sales revenue (RMB);

[0112] Battery consumption cost (RMB, based on power consumption × electricity price);

[0113] Time cost (RMB / minute, based on vehicle operating costs);

[0114] Weight coefficients (optimized through offline training, such as...) ).

[0115] (2) Model training and deployment

[0116] Training data: Collect 3 months of historical operational data (delivery tasks, sales records, power consumption, traffic conditions) to build a simulation environment (based on Gazebo simulation).

[0117] Training algorithm: Proximal policy optimization (PPO) algorithm is used to optimize the policy network through a "sampling-update" loop. The policy network structure is as follows: input layer (128-dimensional state vector) → hidden layer (2 layers, 64 neurons per layer, ReLU activation function) → output layer (3-dimensional action probability, Softmax activation function).

[0118] Deployment method: The trained PPO model is deployed on the vehicle's computing unit (NVIDIA Jetson AGXXavier) to receive perception layer data in real time and output action decisions (such as "switch to sales mode").

[0119] Specific application process: The PPO model executes the following steps in real time at a frequency of 1Hz during vehicle operation: 1. State acquisition: Obtain the current delivery task queue length (T), vehicle position (P), battery SOC (B), road congestion index (C), and predicted pedestrian flow in the target area (H) from the perception layer, forming a 128-dimensional state vector; 2. Action decision: The policy network outputs the probability distribution of three actions (e.g., =0.2、 =0.7、 =0.1), select the action with the highest probability (such as switching to sales mode); 3. Execution feedback: After the vehicle performs the action, the perception layer returns the execution result (such as the flow density of people after arriving at the sales area, sales revenue); 4. Reward update: Calculate the real-time reward according to the reward function (such as sales revenue of 30 yuan - battery consumption of 2 yuan - time cost of 5 yuan = 23 yuan), and feed the reward back to the model for parameter fine-tuning (the network weights are updated every 10 decisions). Example: When the delivery task queue is empty (T=0), battery SOC=80% (B≥30%), and the predicted flow of people in the target area H=90 people / 100㎡, the PPO model outputs an action probability of 0.85 for a2, triggering the sales mode switch.

[0120] 3. Dynamic path planning algorithm (combined with pedestrian flow prediction)

[0121] Route planning is a crucial aspect of delivery and sales models, requiring simultaneous consideration of both "efficiency" (delivery time) and "profitability" (customer traffic). This system employs a dynamic A(D Lite) algorithm combined with an LSTM crowd prediction model to achieve real-time route optimization.

[0122] (1) Principle of D Lite algorithm

[0123] D Lite is an incremental path planning algorithm that can quickly adjust the path when the environment changes (such as the sudden appearance of obstacles or traffic congestion). Its core idea is "searching from the goal point to the starting point." When the environment changes, it only updates the affected nodes, avoiding recalculating the entire path. The algorithm flow is as follows:

[0124] Initialization: Set the starting point (S) and the target point (G), and construct the cost map (based on the environment model of the perception layer);

[0125] Forward search: Starting from (G), calculate the cost (g value) from each node to (G);

[0126] Path generation: Starting from (S), generate a path along the direction of decreasing (g) value;

[0127] Environmental Update: When obstacles or traffic congestion are detected, update the corresponding nodes in the cost map;

[0128] Incremental update: Only recalculate the (g) value of nodes affected by environmental changes to quickly generate new paths.

[0129] Specific application process (delivery scenario example): When a vehicle is performing a community delivery task, it travels along the planned route to section XX. Suddenly, the LiDAR detects a temporary construction barrier 25 meters ahead. The system triggers the D Lite incremental update process: 1. Environment update: Set the cost of the node where the barrier is located to infinity; 2. Incremental calculation: Recalculate only the g-values ​​of the 5 nodes surrounding the barrier (derived backward from the target point); 3. Path generation: Generate a detour path (turning right onto an adjacent side road) within 100ms; the vehicle does not need to stop and directly switches routes, ensuring that the on-time delivery rate is not affected.

[0130] (2) LSTM pedestrian flow prediction model

[0131] To optimize the route selection for the sales model, it is necessary to predict the future pedestrian density in the target area. This system uses a Long Short-Term Memory (LSTM) model to predict the pedestrian density for the next hour based on historical data.

[0132] Input features: pedestrian density over the past 12 hours (once per hour, YOLOv5 pedestrian detection from cameras), weather data (temperature, humidity, whether it is raining, from Gaode Weather API), and time features (day of the week, whether it is a holiday, one-hot encoding).

[0133] Model structure: Input layer (32-dimensional feature vector) → LSTM layer (64 neurons, return_sequences=True) → LSTM layer (32 neurons) → Dropout layer (rate=0.5) → Fully connected layer (output 1-dimensional, predicted value of pedestrian flow density).

[0134] Training and Deployment: The model was trained using the PyTorch framework, with training data consisting of 100,000 surveillance cameras from 10 communities. The mean squared error (MSE) loss function was used. The trained model was deployed in the cloud (Alibaba Cloud ECS) to receive vehicle location data in real time and return predicted pedestrian flow values ​​for the target area (e.g., "the pedestrian density at the entrance of the community in the next hour is 80 people / 100㎡").

[0135] Specific application process (sales scenario example): In sales mode, the vehicle sends real-time data (current location, time, weather) to the cloud every 10 minutes. The cloud-based LSTM model performs the following steps: 1. Feature fusion: Encodes historical 12-hour pedestrian flow data, real-time weather (28℃ / sunny), and time features (Saturday 14:00) into a 32-dimensional vector; 2. Sequence prediction: Outputs the pedestrian flow density of 5 surrounding areas for the next hour through a two-layer LSTM network (e.g., 92 people / 100㎡ at the entrance of a residential area, 78 people / 100㎡ at the entrance of a supermarket); 3. Result feedback: The predicted value is returned to the vehicle's decision layer, which selects the area with the highest pedestrian flow as the target point to guide D Lite in planning the path; the entire process takes ≤2 seconds, ensuring that the sales strategy adapts to changes in pedestrian flow in real time.

[0136] (3) Path planning process (taking the sales model as an example)

[0137] Target area selection: Based on the prediction results of the LSTM model, the decision-making level selects the three areas with the highest population density in the next hour (such as "the entrance of the community, the subway entrance, and the park entrance").

[0138] Cost map construction: Convert the pedestrian density of the target area into "benefit cost" (e.g., the higher the pedestrian density, the lower the cost), and at the same time convert the traffic congestion situation into "time cost" (e.g., the higher the congestion index, the higher the cost).

[0139] DLite route calculation: Starting from the current vehicle position and ending at the target area, the optimal route is calculated by combining "benefit cost + time cost".

[0140] Real-time adjustment: If traffic congestion is detected during driving (e.g., LiDAR detects a queue of vehicles 100 meters ahead), D Lite will be triggered to update the route and adjust the path (e.g., take a detour).

[0141] (III) Execution Layer: Vehicle Control and Function Implementation

[0142] The execution layer is the "hands and feet" of the system. Based on the instructions of the decision layer, it controls the vehicles to complete actions such as delivery, sales, and charging. The core components include the vehicle drive system, delivery box system, vending machine system, and charging management system.

[0143] 1. Vehicle drive system

[0144] Hardware components:

[0145] Chassis: Customized wheeled chassis (4-wheel drive, load capacity 500kg);

[0146] Motor: Jingjin Electric permanent magnet synchronous motor (peak power 150kW, torque 400N·m);

[0147] Controller: Infineon TC397 microcontroller (supports CAN FD protocol);

[0148] Steering system: Electric power steering (EPS), steering accuracy ±0.5°;

[0149] Braking system: Brake-by-wire (BBW), braking distance ≤ 5m (vehicle speed 30km / h).

[0150] Control process:

[0151] The decision-making layer outputs path instructions (such as "drive along XX road, steering angle 30°"), which are sent to the motor controller and steering controller via the CAN bus. The motor controller adjusts the motor speed according to the instructions (PWM control), and the steering controller adjusts the steering angle (based on EPS current control). At the same time, the braking system monitors the vehicle status in real time (such as vehicle speed and distance to obstacles). When an emergency occurs (such as a pedestrian 5 meters ahead), emergency braking is triggered (BBW outputs maximum braking force).

[0152] 2. Delivery Box System

[0153] Hardware components:

[0154] Cargo box: Divided metal cargo box (6 compartments, each compartment measuring 50cm×50cm×50cm);

[0155] Locks: Electronic locks (1 per compartment, supporting Bluetooth unlocking);

[0156] Sensors: Each compartment is equipped with an RFID reader (to identify cargo tags) and a weight sensor (to verify cargo weight).

[0157] Delivery process (taking community group buying as an example):

[0158] Pick-up: When the vehicle arrives at the group leader's location, the group leader sends a pickup instruction via the APP, and the electronic lock on the cargo box automatically unlocks; the group leader places the goods into the cargo compartment, the RFID reader scans the goods label (to confirm the goods are correct), and the weight sensor verifies the weight of the goods (e.g., "The goods label shows a weight of 2kg, the actual weight is 1.98kg, the error is ≤1%, and the verification is passed"); after the pickup is completed, the electronic lock automatically locks.

[0159] Delivery: When the vehicle arrives at the user's building, the user receives a pickup notification (including a pickup code) via a mini-program; the user scans the QR code on the cargo box, enters the pickup code, and the corresponding electronic lock on the cargo compartment unlocks; after the user takes the goods, the weight sensor detects that the cargo compartment is empty and sends a "delivery complete" instruction to the decision-making level.

[0160] Exception handling:

[0161] If the weight error exceeds 1% when picking up the goods (e.g., "the goods label shows 2kg, but the actual weight is 1.8kg"), the system will trigger an alarm (the buzzer will sound) and notify the group leader and the operations department.

[0162] If the user does not pick up the goods upon delivery (more than 30 minutes later), the system will send a reminder SMS. If the goods are still not picked up after 2 hours, the vehicle will take the goods back to the group leader.

[0163] 3. Vending machine system

[0164] Hardware components:

[0165] Vending machine: Layered glass cabinet (4 layers, 8 compartments per layer, temperature controlled (0-25℃));

[0166] Interactive terminal: 10.1-inch touchscreen (supports QR code payment and voice interaction);

[0167] Sensors: Each compartment is equipped with an infrared sensor (to detect whether the goods have been taken) and a temperature sensor (to monitor the temperature of the compartment).

[0168] Sales process (taking the entrance of a residential community as an example):

[0169] Mode switching: After the decision-maker triggers the sales mode, the vending machine system is powered on, the touch screen displays a list of products (such as "mineral water 2 yuan, bread 5 yuan, fruit 10 yuan"), and the voice prompts "Welcome, please say 'Xiaodu Xiaodu' if you need help".

[0170] User purchase: Users select products via touchscreen (or voice command "I want to buy mineral water"), and the system displays the product price and inventory (e.g., "Mineral water inventory: 10 bottles"); users scan the QR code to pay (supports WeChat / Alipay), and after successful payment, the corresponding compartment's electronic lock unlocks (the electromagnetic lock is de-energized, and the spring pops out the product); the infrared sensor detects that the product has been taken (e.g., "The infrared beam of compartment A1 was blocked and then restored"), and sends a "sale completed" command to the decision-making level, updating the inventory (e.g., "Mineral water inventory reduced to 9 bottles").

[0171] Intelligent recommendation strategy:

[0172] The vending machine system integrates a collaborative filtering (CF) algorithm to recommend products (e.g., "User 1 has purchased mineral water and bread") based on the user's purchase history (e.g., "User 1 has purchased mineral water and bread") and the current scenario (e.g., "Summer, temperature 30℃"), such as "Iced tea recommended, discounted price 3 yuan". The recommendation process is as follows:

[0173] User profile building: Collect user data (such as age, gender, purchase history) through mini-programs to build user profiles (such as "25-year-old male, likes sports, often buys mineral water and energy bars").

[0174] Scene recognition: Identify the current scene (e.g., "Summer, at the entrance of the residential area, high population density") through perception layer data;

[0175] Recommendation generation: The collaborative filtering algorithm calculates the matching degree between products and user profiles and scenarios (e.g., "Iced tea has a matching degree of 0.8 with the summer scenario and a matching degree of 0.7 with user 1's profile"), and outputs the top 3 recommended products (e.g., "Iced tea, mineral water, watermelon").

[0176] Specific application process (taking user purchase as an example): When a user scans the code to enter the sales interface, the system executes the following steps: 1. User identification: Retrieve user profile through the mini-program ID (e.g., "25-year-old male, sports enthusiast"); 2. Scene acquisition: The perception layer acquires current scene data (temperature 30℃ / sunny / entrance to the community); 3. Matching degree calculation: The collaborative filtering algorithm calculates the product matching degree (Iced tea: scene matching 0.8 + user profile matching 0.7 = 1.5; Mineral water: scene matching 0.6 + user profile matching 0.9 = 1.5; Watermelon: scene matching 0.9 + user profile matching 0.5 = 1.4); 4. Recommended display: The touch screen prioritizes displaying the iced tea and mineral water with the highest matching degree and labels them with "Summer Recommendation"; The entire process takes ≤500ms, improving the product conversion rate by approximately 15%.

[0177] 4. Charging Management System

[0178] Hardware components:

[0179] Battery: Lithium iron phosphate battery (capacity 100kWh, voltage 72V, cycle life 2000 times).

[0180] BMS: CATL BMS-100 (supports CAN FD protocol, monitors battery status);

[0181] Charging station interface: Compliant with GB / T 20234.3-2015 standard (supports fast charging, maximum current 150A);

[0182] Solar panel (optional): LONGi Green Energy PERC solar panel (200W power, 22% efficiency) for auxiliary charging.

[0183] Charging process:

[0184] Battery power monitoring: The BMS estimates the battery's SOC (State of Charge) in real time using the Extended Kalman Filter (EKF) algorithm, as shown in the following formula:

[0185]

[0186] in:

[0187] State vector (SOC, battery internal resistance);

[0188] : State transition matrix;

[0189] Specific application process (charging scenario example): The BMS executes an EKF cycle every 100ms: 1. Prediction step: Calculate the SOC at the next moment using the state equation (e.g., current SOC=80%, discharge current 10A, sampling time 0.1s, coulomb efficiency 0.98 → predicted SOC=80% - (0.98×10×0.1) / 100 = 79.902%); 2. Observation step: Collect battery voltage (e.g., 3.65V), and combine it with the OCV-SOC curve (3.65V corresponds to SOC=79.8%); 3. Update step: Correct the predicted value with the observed value (Kalman gain K=0.8 → corrected SOC=79.902% + 0.8×(79.8% -79.902%)=79.82%); 4. Output: The corrected SOC (79.82%) is sent to the decision-making level to guide the switching of charging modes; this process ensures that the SOC estimation accuracy is ≥98% and avoids the risk of overcharging / over-discharging.

[0190] Input matrix ( For Coulomb efficiency, Sampling time, (Battery rated capacity).

[0191] Input (charging / discharging current);

[0192] Observed value (battery voltage);

[0193] Observation matrix ( (This refers to the open-circuit voltage, based on the battery OCV-SOC curve).

[0194] Process noise and observation noise (variance determined by offline testing).

[0195] EKF estimates SOC in real time through a "prediction-update" cycle, with an estimation accuracy of ≥98%.

[0196] Specific application process (discharge scenario example): When the vehicle discharges in delivery mode, the EKF algorithm executes a loop every 100ms: 1. Prediction phase: Calculate the predicted SOC value for the next moment based on the current SOC (e.g., 80%), discharge current (10A), and coulomb efficiency (0.98): Predicted SOC value = Current SOC - (Coulomb efficiency × discharge current × time change) / Rated capacity = 80% - (0.98 × 10 × 0.1) / 100 = 79.902%; 2. Observation phase: The BMS collects the battery voltage (e.g., 3.65V) and uses the OCV-SOC curve (3.65V corresponds to SOC = 79.8%) as the observed value; 3. Update phase: Calculate the Kalman gain (K = 0.8) and correct the predicted value: Corrected SOC = Predicted SOC + Kalman gain × (Observed SOC - Predicted SOC) = 79.902% + 0.8 × (79.8% - 79.902%) = 79.82%; 4. Output feedback: The corrected SOC (79.82%) is sent to the decision-making level to guide power management; this process ensures that the SOC estimation accuracy is ≥98% and avoids the risk of over-discharge.

[0197] Charging trigger: When SOC≤20%, the decision layer triggers the charging mode and plans the path to the nearest charging station (e.g., "the community charging station 1.2 kilometers away from the current location") using the D Lite algorithm.

[0198] Charging process: When the vehicle arrives at the charging station, it sends a charging request (including battery voltage and current requirements) via the CAN bus. The charging station responds and starts charging. The BMS monitors the battery temperature in real time (if the temperature is ≥50℃, the charging current is reduced). After charging is complete (SOC≥95%), a "charging complete" command is sent to the decision-making level.

[0199] (iv) Interaction layer: Interface between users and the operation side

[0200] The interaction layer serves as a bridge between the system and users and operations personnel. Its core objective is to improve user experience and reduce operating costs. It includes the user end (APP / Mini Program), the operations end (back-end management system), and the vehicle end (touchscreen / voice).

[0201] 1. User-side (APP / Mini Program)

[0202] Functional design:

[0203] Delivery service: Supports community group buying orders (select group leader and products), delivery progress tracking (real-time location, estimated arrival time), and pickup notification (SMS / push notification);

[0204] Sales services: Supports nearby vehicle search (displays vehicle location on map and a list of products for sale), online ordering (scan to pay, schedule pickup), and product reviews (ratings, comments);

[0205] Personal Center: Supports order history inquiry, coupon redemption (such as "first purchase gets a 5 yuan coupon"), and customer service contact (online chat / phone).

[0206] Interface design: Adopting the Material Design style, the interface is simple (such as the delivery progress page displaying "Current location: entrance of the community, estimated arrival time: 15:30"), and the operation process is short (such as purchasing goods only requires "scan code → select goods → pay" 3 steps).

[0207] 2. Operations side (back-end management system)

[0208] Functional design:

[0209] Vehicle monitoring: Real-time display of vehicle status (location, battery level, mode, cargo status), supports abnormal alarms (such as "Vehicle No. 1 battery temperature is too high");

[0210] Task scheduling: Supports manual task adjustment (e.g., "Switch vehicle No. 2 from sales mode to delivery mode") and viewing task statistics (e.g., "100 orders delivered today, sales revenue of 500 yuan").

[0211] Inventory Management: Real-time synchronization of vending machine inventory (e.g., "20 bottles of mineral water remaining"), and support for automatic replenishment reminders (e.g., "Notify the warehouse to replenish when mineral water inventory is ≤10 bottles").

[0212] Data statistics: Generate operational reports (such as "weekly delivery volume trend, popular selling products"), and support custom analysis (such as "sales revenue share of Community A").

[0213] Technical architecture: It adopts a Spring Boot + Vue.js front-end and back-end separation architecture. The back-end is deployed on Alibaba Cloud ECS (supporting high concurrency, such as 1,000 vehicles online at the same time), the front-end is deployed on Alibaba Cloud CDN (accelerating access), and the database uses MySQL (master-slave replication to ensure data reliability).

[0214] 3. Vehicle-side (touchscreen / voice)

[0215] Touchscreen functionality:

[0216] Display a list of products (with pictures, prices, and stock);

[0217] Supports QR code payment (WeChat / Alipay);

[0218] Display vehicle status (e.g., "Current mode: Selling, Battery level: 80%");

[0219] Supports fault reporting (such as "Vending machine A1 compartment cannot be opened").

[0220] Voice interaction function:

[0221] It uses Baidu's Automatic Speech Recognition (ASR) and Text-to-Speech (TTS) technologies to support natural language interaction (e.g., when a user says "I want to buy bottled water," the system replies "Okay, the bottled water costs 2 yuan, please scan the code to pay").

[0222] It supports voice navigation (e.g., if a user says "go to the entrance of the community", the system will reply "the route has been planned and you are expected to arrive in 5 minutes".).

[0223] (v) Energy layer: self-contained charging and energy management

[0224] The energy layer is the system's "power source," with the core objective of extending range and reducing charging costs. It adopts a hybrid energy solution of "battery + solar energy + energy recovery."

[0225] 1. Energy recovery system

[0226] Hardware components: motor controller (supports energy recovery), DC / DC converter (converts recovered electrical energy into battery voltage).

[0227] Working principle: When the vehicle brakes or decelerates, the motor is in regenerative braking mode, converting kinetic energy into electrical energy. The motor controller converts the generated current (AC) into DC current, adjusts the voltage (matching the battery voltage) through a DC / DC converter, and stores the electrical energy in the battery. The energy recovery efficiency is approximately 20% (e.g., approximately 0.5 kWh of electrical energy can be recovered when the vehicle decelerates from 30 km / h to 0).

[0228] 2. Solar-assisted charging system

[0229] Hardware components: solar panel (mounted on the vehicle roof, 200W power), MPPT controller (maximum power point tracking), battery.

[0230] Working principle: The solar panel converts sunlight into electrical energy. The MPPT controller tracks the solar panel's maximum power point (MPPT) (e.g., at 10 AM, the MPPT voltage is 36V and the current is 5.5A) and stores the electrical energy in the battery. On a sunny day, the solar panel can provide approximately 1.5 kWh of electricity per day (equivalent to increasing the range by 15 km).

[0231] (vi) Example of system application process

[0232] Taking the typical scenario of "morning peak delivery + midday sales" as an example, the complete application process of the system is demonstrated:

[0233] 1. Morning peak delivery mode (7:00-9:00)

[0234] Task Acquisition: At 6:50, the decision-makers obtained 10 delivery tasks from the Meituan Community Group Buying Platform (all of which were orders from the group leader of Community A).

[0235] Route planning: The decision-making layer optimizes the delivery route through a genetic algorithm (optimizing the pickup / delivery order of 10 orders to "Group Leader 1 → User 1 → User 2 → ... → User 10", reducing the total driving distance from 15 kilometers to 10 kilometers).

[0236] The specific application process of the genetic algorithm (delivery route optimization): The decision layer transforms the 10 delivery tasks into a Traveling Salesman Problem (TSP) and executes the following steps: 1. Encoding: Encode the order sequence into a gene sequence (e.g., [Group Leader 1 → User 3 → User 5 → User 2 → ...]); 2. Initialize the population: Generate 100 random gene sequences as the initial population; 3. Fitness calculation: Use the total travel distance as the fitness function (the shorter the distance, the higher the fitness); 4. Selection: Use roulette wheel selection to select the top 30% of individuals with the best fitness; 5. Crossover: Perform single-point crossover on the selected individuals (e.g., swap the 3rd to 5th gene sequences); 6. Mutation: Randomly change the position of one gene in 10% of the individuals; 7. Iteration: Repeat steps 3-6, iterate 50 times, and output the gene sequence with the highest fitness (i.e., the optimal delivery order); This process takes ≤2 seconds, and the total travel distance is shortened from 15 kilometers to 10 kilometers, improving efficiency by 33%.

[0237] Pick-up and delivery: 7:00, the vehicle drives to the group leader's location to pick up the goods (RFID verification, weight detection); 7:10, the vehicle drives to the user's location on the first floor, and the user scans the code to pick up the goods; 7:30, all 10 orders are delivered, and the task queue is empty;

[0238] Mode switching: At 7:30, the decision-making level triggers the switch to sales mode based on the decision of the PPO model (current power consumption is 80%, and the predicted population flow in community A is 70 people / 100㎡).

[0239] 2. Midday sales mode (9:00-12:00)

[0240] Target area selection: Based on the prediction results of the LSTM model, the decision-making level selected the entrance of community A (with a future hourly pedestrian density of 80 people / 100㎡) as the target area;

[0241] Route planning: The D Lite algorithm is used to plan a route (avoiding congested sections during the morning rush hour, with a driving distance of 1.5 kilometers and an estimated time of 10 minutes).

[0242] Sales process: 9:10, the vehicle arrives at the entrance of community A and starts the sales mode (the touchscreen displays a product list and a voice prompt says "Welcome"); 9:15, user 1 scans the code to purchase bottled water (payment successful, the vending machine displays the product); 10:30, the vending machine's inventory shows "5 bottles of bottled water remaining", and the decision-making level triggers a restocking reminder (notifying the warehouse); 12:00, sales revenue reaches 300 yuan, the task queue receives a new delivery task (from Ele.me), and the decision-making level switches to delivery mode.

[0243] 3. Charging mode (18:00-18:30)

[0244] Battery monitoring: At 17:50, the BMS estimated the SOC to be 18%, triggering charging mode;

[0245] Path planning: The decision-making layer plans the path to the community charging station (distance 1.2 kilometers) using the D Lite algorithm.

[0246] Charging process: At 18:00, the vehicle arrives at the charging station and begins charging; at 18:30, charging is complete (SOC≥95%), and the decision-making team switches to sales mode (heading to peak evening traffic hotspots).

[0247] (vii) System performance optimization and anomaly handling

[0248] 1. Performance optimization metrics

[0249] index target value Implementation On-time delivery rate ≥95% Dynamic route planning (D Lite) and real-time traffic forecasting Inventory accuracy ≥99% Fusion of infrared sensors and weight sensors Battery SOC estimation accuracy ≥98% Extended Kalman Filter (EKF) Vehicle positioning accuracy ±2cm (RTK corrected) GPS / IMU integrated navigation, SLAM Task scheduling efficiency ≤1 second (real-time decision) The PPO model is deployed in an edge computing unit (Jetson Xavier).

[0250] 2. Exception Handling Mechanism

[0251] Obstacle collision: The lidar detects an obstacle 5 meters ahead, triggering emergency braking (BBW outputs maximum braking force) and simultaneously sending an alarm message to the operations department (displaying "Vehicle No. 1 has collided with an obstacle at the entrance of the community").

[0252] Vending machine malfunction: The electronic lock cannot be opened (e.g., "Vending machine A1 compartment electronic lock is stuck"). The system switches to manual mode (the touch screen displays "Please contact customer service, phone: 400-XXX-XXXX") and notifies the operations department to arrange repairs.

[0253] Lost Goods: The weight sensor of the delivery box detects that "the weight of compartment 1 has changed from 2kg to 0kg, but no pickup instruction has been received". The system triggers an alarm (buzzer sounds) and notifies the operations department (displaying "Goods lost in compartment 1 of vehicle 2").

[0254] Network interruption: The network connection between the vehicle and the cloud is interrupted (e.g., weak 4G signal). The system switches to local mode (using stored map data for navigation, sales data is temporarily stored locally, and synchronized to the cloud after the network is restored).

[0255] (viii) System commercial value and promotion prospects

[0256] 1. Commercial Value

[0257] Improve equipment utilization: Traditional unmanned delivery vehicles work 8 hours a day (delivery), while this system can work 12 hours (8 hours of delivery + 4 hours of sales), increasing equipment utilization by 50%;

[0258] Increase income sources: In addition to delivery fees (about 1 yuan per order), sales revenue (about 500 yuan per day per vehicle) becomes a new source of income growth, and the monthly income per vehicle can be increased from 3,000 yuan to 6,000 yuan;

[0259] Reduced operating costs: Unmanned operation reduces labor costs (such as the need for drivers in traditional delivery, which this system does not require), and the monthly operating cost per vehicle is reduced from 5,000 yuan to 3,000 yuan;

[0260] Enhanced user experience: Features such as real-time delivery tracking and convenient local product purchase have improved user satisfaction (expected user retention rate to increase from 60% to 80%).

[0261] 2. Promotion Prospects

[0262] Application scenarios: Suitable for closed / semi-closed scenarios such as communities, parks, and scenic spots (e.g., "delivery and sales within the community", "delivery and sales within the park office").

[0263] Cooperation Model: Partner with community group buying platforms (such as Meituan Youxuan) and retail brands (such as Nongfu Spring) to achieve a win-win situation through a combination of "delivery + sales" services (the platform reduces delivery costs, and the brand increases sales channels).

[0264] scalability: The system adopts a modular design (the perception layer, decision layer, and execution layer are independent), which supports rapid replication (e.g., expanding from 10 vehicles to 100 vehicles only requires adding hardware and deployment models).

[0265] This system, through multi-sensor fusion, intelligent decision-making algorithms, and dynamic function switching technologies, solves the problems of "single function and low utilization rate" in traditional unmanned vending vehicles, realizing the dual value of "delivery + sales". The core advantage of the system is:

[0266] Intelligent scheduling: The PPO model optimizes task decisions and maximizes operational benefits;

[0267] Dynamic path planning: D Lite combines LSTM pedestrian flow prediction to balance efficiency and effectiveness;

[0268] High reliability: Multi-sensor fusion and anomaly handling mechanisms ensure stable system operation;

[0269] Excellent scalability: The modular design allows for rapid deployment to different scenarios.

[0270] In the future, with the further maturation of autonomous driving technology (such as Level 4 autonomous driving), this system can be extended to open roads (such as urban delivery and sales), further enhancing its commercial value.

[0271] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 4 As shown, optionally, electronic device 410 may include a first processor 2001.

[0272] Optionally, the electronic device 410 may also include a memory 2002 and a transceiver 2003.

[0273] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0274] The following is combined Figure 4 A detailed description of each component of electronic device 410 is provided below:

[0275] The first processor 2001 is the control center of the electronic device 410. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0276] Optionally, the first processor 2001 can perform various functions of the electronic device 410 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0277] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 4 CPU0 and CPU1 are shown in the diagram.

[0278] In a specific implementation, as one example, the electronic device 410 may also include multiple processors, for example... Figure 4 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0279] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0280] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently and be connected via the interface circuit of the electronic device 410. Figure 4 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0281] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0282] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 4 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0283] Optionally, the transceiver 2003 can be integrated with the first processor 2001, or it can exist independently and be connected via the interface circuit of the electronic device 410. Figure 4 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0284] It should be noted that, Figure 4 The structure of the electronic device 410 shown does not constitute a limitation on the router. Actual knowledge structure identification devices may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0285] Furthermore, the technical effects of the electronic device 410 can be referenced from the technical effects of the unmanned vending vehicle driving method with community delivery and mobile sales functions described in the above method embodiments, and will not be repeated here.

[0286] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0287] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0288] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0289] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0290] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0291] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0292] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0293] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, systems, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0294] In the embodiments provided by this invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0295] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0296] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0297] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0298] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A drive system for an unmanned vending vehicle that combines community delivery and mobile vending functions, characterized in that: The architecture adopts a five-layer structure: perception, decision-making, execution, interaction, and energy, including: The perception layer, composed of the fusion of multiple sensors, is used to acquire vehicle status, environmental and task data; The decision-making layer, based on data from the perception layer, enables task scheduling, dynamic function switching, and path planning. The execution layer, based on instructions from the decision-making layer, controls vehicles to complete delivery, sales, or charging actions; The interaction layer provides operation and monitoring interfaces for users and operations personnel. The energy layer provides power to the system and manages energy. The decision-making layer is configured to dynamically switch between delivery mode, sales mode, or charging mode based on delivery task queues, vehicle battery level, and external environmental information.

2. The system according to claim 1, characterized in that, The decision-making layer includes an intelligent task scheduling module, which uses the PPO algorithm based on deep reinforcement learning for decision-making, wherein: The state space includes the delivery task queue (T), vehicle location (P), battery status (B), traffic conditions (C), and pedestrian density (H); The action space includes receiving delivery tasks (a1), switching to sales mode (a2), and heading to a charging point (a3); The reward function is R = α·R_del + β·R_sell - γ·C_bat - δ·C_time, where R_del is the delivery revenue, R_sell is the sales revenue, C_bat is the battery consumption cost, C_time is the time cost, and α, β, γ, and δ are weighting coefficients.

3. The system according to claim 1, characterized in that, The decision-making layer includes a dynamic path planning module, which uses the D Lite algorithm for real-time path planning and integrates the pedestrian density prediction results based on the LSTM model. The input features of the LSTM model include historical pedestrian density, weather data, and time features, and the output is the predicted pedestrian density value of the target area for future periods.

4. The system according to claim 1, characterized in that, The execution layer includes: The delivery box system has multiple compartments with electronic locks, each equipped with an RFID reader and a weight sensor for cargo identification and weight verification; The vending machine system has multiple compartments with electronic locks, each equipped with an infrared sensor and a temperature sensor, and integrates an intelligent recommendation function based on a collaborative filtering algorithm; The charging management system uses an extended Kalman filter algorithm to estimate the battery's state of charge (SOC) in real time and triggers the charging process when the SOC falls below a threshold.

5. The system according to claim 1, characterized in that, The energy layer includes an energy recovery system and a solar-assisted charging system; the energy recovery system converts kinetic energy into electrical energy and charges the battery when the vehicle brakes or decelerates; the solar-assisted charging system stores the electrical energy generated by the solar panels into the battery through an MPPT controller.

6. A method for driving an unmanned vending vehicle based on the system described in any one of claims 1-5, characterized in that, Includes the following steps: Perception steps: Acquire vehicle status, external environment, and task-related data through multi-sensor fusion; Decision-making steps: Based on the perceived data, intelligent algorithms are used to perform task scheduling, functional mode decision-making, and path planning; Execution steps: Based on the decision-making instructions, control the vehicle to perform delivery, sales, or charging tasks; The system dynamically switches between delivery, sales, and charging modes based on delivery task queues, vehicle battery levels, and external environmental information.

7. The method according to claim 6, characterized in that, The task scheduling in the decision-making step adopts the PPO algorithm based on deep reinforcement learning, including: Construct the state space, action space, and reward function; The policy network outputs an action probability distribution based on the current state, and selects the action with the highest probability to execute. The reward value is calculated based on the execution result and fed back to the model for parameter updates.

8. The method according to claim 6, characterized in that, The path planning in the decision-making step uses the D Lite algorithm combined with LSTM pedestrian flow prediction, including: Predict the future pedestrian density of a target area using an LSTM model; The predicted value of pedestrian flow density is converted into a benefit cost, and a comprehensive cost map is constructed by combining the time cost of traffic conditions. Incremental path planning is performed based on the comprehensive cost map using the D Lite algorithm.

9. The method according to claim 6, characterized in that, The delivery process in the execution steps includes: When picking up goods, the RFID reader identifies the goods tag and the weight sensor verifies the weight of the goods. When delivering goods, users unlock the corresponding compartment with a pickup voucher, and a weight sensor detects the pickup completion status. The sales process in the execution steps includes: The inventory is updated by detecting the status of goods being picked up using infrared sensors. Based on collaborative filtering algorithms, products are recommended to users according to their profiles and the current context.

10. The method according to claim 6, characterized in that, The method also includes energy management steps: Battery SOC is estimated in real time using the extended Kalman filter algorithm; When the SOC is lower than the first threshold, the charging mode is triggered, and a path to the charging station is planned. When the vehicle brakes or decelerates, kinetic energy is converted into electrical energy through an energy recovery system; A solar-assisted charging system converts solar energy into electrical energy to charge the battery.

Images