Park unmanned driving path control method and system combined with multimodal perception

Abstract

The present invention provides a method and system for controlling unmanned driving paths in a park combined with multimodal perception, which involves the field of multi-vehicle collaborative path planning. The method includes: monitoring and obtaining a number of commodity consumption data sequences and a number of passenger flow density sequences, analyzing and obtaining a number of predicted commodity shortage times; enumerating deliverable paths and generating a number of initial delivery plans; using a number of predicted commodity shortage times as delivery constraints, minimizing the time difference in distribution and minimizing the distance between pickup and delivery as comprehensive optimization goals, evaluating the pros and cons of a number of initial delivery plans, and outputting the optimal delivery plan for control. The method aims to solve the problems of untimely scheduling response, inaccurate paths, and low vehicle resource utilization in the actual application of traditional path planning methods; it can significantly improve the real-time performance of collaborative scheduling of unmanned delivery vehicles, the accuracy of path planning, and the overall delivery efficiency, and achieve intelligent, efficient, and safe park-level unmanned delivery control effects.

Description

Technical Field

[0001] The present invention relates to the field of multi-vehicle collaborative path planning, and in particular to a method and system for controlling unmanned driving paths in a park combined with multimodal perception. Background Art

[0002] With the development of smart scenic spots and unmanned delivery technology, tourist attractions have gradually introduced unmanned delivery vehicles for material transportation, commodity delivery and tourist services to improve operational efficiency and tourist experience.

[0003] However, existing unmanned delivery route planning methods are mostly based on static maps and preset routes, and lack the ability to dynamically perceive the actual operating status of scenic spots. Especially in scenic areas with dense passenger flow and frequent demand fluctuations, traditional route planning methods cannot flexibly adjust scheduling strategies based on real-time passenger flow, material consumption and vehicle distribution, resulting in uneven distribution of multi-vehicle delivery tasks, frequent route conflicts, and delayed responses. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for controlling unmanned driving paths in a park combined with multimodal perception, so as to solve the problems existing in traditional unmanned delivery vehicle path planning methods for tourist attractions, such as the inability to dynamically adjust the coordinated scheduling of multiple vehicles according to the actual operating status of the scenic area, resulting in untimely scheduling response, inaccurate paths, and low vehicle resource utilization in actual applications. The method includes:

[0005] In the first aspect, the present invention provides a method for controlling unmanned driving paths in a park combined with multimodal perception, comprising: building a three-dimensional simulation space of the park based on a network distribution map and a line structure map of the target park, wherein the network points include multiple storage warehouses and several supply points; monitoring and obtaining a plurality of commodity consumption data sequences of the several supply points in a preset historical time zone, and a plurality of passenger flow density sequences of a plurality of preset monitoring points, and analyzing and obtaining a plurality of predicted commodity shortage times; within the three-dimensional simulation space of the park, enumerating deliverable paths based on the real-time position coordinates of a plurality of idle unmanned delivery vehicles, and generating a plurality of initial delivery plans; using the several predicted commodity shortage times as delivery constraints, and minimizing the distribution time difference and minimizing the pick-up and delivery distance as the comprehensive optimization goals, evaluating the advantages and disadvantages of the multiple initial delivery plans according to the multiple passenger flow density sequences, outputting the optimal delivery plan, and controlling the driving paths of the multiple idle unmanned delivery vehicles.

[0006] Preferably, the method for controlling unmanned driving paths in a park combined with multimodal perception also includes: within the preset historical time zone, continuously monitoring and obtaining the commodity consumption ratio of the supply points, constructing a commodity consumption ratio sequence as a commodity consumption data sequence, and obtaining several commodity consumption data sequences; continuously monitoring and obtaining the passenger flow density of the preset monitoring points, and obtaining multiple passenger flow density sequences; performing correlation analysis on the supply points and the preset monitoring points according to the preset coverage range, determining several associated monitoring point sets of the several supply points, and mapping to obtain several associated passenger flow density sequence sets; constructing a commodity shortage predictor based on a generative adversarial network, and performing commodity shortage prediction based on several real-time remaining commodity ratios of the several supply points, combined with the several commodity consumption data sequences and the several associated passenger flow density sequence sets, and outputting several predicted commodity shortage times.

[0007] Preferably, the unmanned driving path control method for a park combined with multimodal perception also includes: collecting a sample remaining commodity ratio set, a sample commodity consumption data sequence set and multiple sample associated passenger flow density sequence sets based on the historical operation monitoring records of the target park, and obtaining the historical commodity shortage duration as the sample commodity shortage time to obtain a sample commodity shortage time set; using the sample remaining commodity ratio set, the sample commodity consumption data sequence set and multiple sample associated passenger flow density sequence sets as input, and using the sample commodity shortage time set as supervision, the generator and discriminator of the generative adversarial network are supervised trained until convergence to obtain the commodity shortage predictor.

[0008] Preferably, the unmanned driving path control method for a park combined with multimodal perception also includes: randomly selecting a first distribution plan from the multiple initial distribution plans, and randomly selecting a first unmanned delivery vehicle, obtaining the first distribution path and the first distribution commodity weight of the first unmanned delivery vehicle in the first distribution plan; obtaining first path structure information based on the first distribution path mapping, wherein the first path structure information includes first pickup path structure information and first delivery path structure information, and the path structure information includes at least line length, road type, slope and curvature, and the delivery path includes one or more delivery supply points; obtaining a first covering passenger flow density sequence set based on the first delivery path mapping, wherein the first covering passenger flow density sequence set includes a first pickup coverage passenger flow density sequence set and a first delivery coverage passenger flow density sequence set; in the three-dimensional simulation space of the park, when picking up goods according to the first pickup path structure information and the first pickup coverage passenger flow density sequence set. long prediction, output the first predicted pickup time; predict the delivery time according to the first delivery route structure information, the first delivery coverage passenger flow density sequence set and the first delivery commodity weight, and output the first predicted delivery time; determine the first predicted loading time based on the first delivery commodity weight analysis, and obtain the first arrival time by combining the first predicted pickup time and the first predicted delivery time, wherein the first arrival time includes the delivery arrival time of one or more delivery supply points; analyze and obtain multiple arrival times of multiple delivery routes in the first delivery plan in sequence, and determine multiple delivery arrival times of multiple supply points; judge whether the multiple delivery arrival times meet the multiple predicted commodity shortage times, and if so, set the first delivery plan as an optional delivery plan, and analyze and obtain multiple optional delivery plans in sequence; with minimizing the distribution time difference and minimizing the pickup and delivery distance as the comprehensive optimization goal, evaluate the advantages and disadvantages of the multiple optional delivery plans and output the optimal delivery plan.

[0009] Preferably, the method for controlling unmanned driving paths in a park combined with multimodal perception also includes: taking the several predicted commodity shortage times as a benchmark, performing deviation calculations on the multiple delivery arrival time sets of the multiple optional distribution plans respectively, determining multiple time deviation sets, and summing up to obtain multiple delivery time deviations; taking the several predicted commodity shortage times as a benchmark, performing deviation amplitude analysis based on the multiple time deviation sets, and calculating multiple delivery time deviation amplitude means; determining multiple distribution time difference coefficients based on the multiple delivery time deviations and the multiple delivery time deviation amplitude means, wherein the distribution time difference coefficient is negatively correlated with the delivery time deviation and positively correlated with the delivery time deviation amplitude mean; evaluating the advantages and disadvantages of the multiple optional distribution plans based on multiple delivery distance coefficients and the multiple distribution time difference coefficients, and outputting the optimal distribution plan.

[0010] Preferably, the method for controlling unmanned driving paths in a park combined with multimodal perception also includes: configuring a delivery distance weight according to the weight of the goods, wherein the delivery distance weight is positively correlated with the weight of the goods, the delivery distance weight is greater than or equal to 1, and when the weight of the goods is 0, the delivery distance weight is 1; randomly selecting a first optional delivery plan to obtain a first pickup distance and a first delivery distance of the first optional delivery plan; compensating the first delivery distance according to the delivery distance weight to obtain a first compensated delivery distance, and calculating a first pickup and delivery distance in combination with the first pickup distance, setting it as a first pickup and delivery distance coefficient, and analyzing in sequence to obtain multiple pickup and delivery distance coefficients.

[0011] Preferably, the method for controlling unmanned driving paths in a park combined with multimodal perception also includes: performing dimensionless processing on the multiple pick-up and delivery distance coefficients and the multiple distribution time difference coefficients to obtain multiple standard pick-up and delivery distance coefficients and multiple standard distribution time difference coefficients; performing scheme fitness evaluation based on the multiple standard pick-up and delivery distance coefficients and the multiple standard distribution time difference coefficients, outputting multiple fitnesses, and selecting the optional distribution scheme with the maximum fitness as the optimal distribution scheme, wherein the fitness is negatively correlated with the standard pick-up and delivery distance coefficient and the standard distribution time difference coefficient.

[0012] In a second aspect, the present invention further provides a campus unmanned driving path control system combined with multimodal perception, which is used to execute a campus unmanned driving path control method combined with multimodal perception as described in the first aspect, including: a campus simulation space construction module, which is used to build a three-dimensional simulation space of the campus based on the network distribution map and line structure map of the target campus, wherein the network includes multiple storage warehouses and several supply points; a commodity shortage time prediction module, which is used to monitor and obtain multiple commodity consumption data sequences of the several supply points in a preset historical time zone, and multiple passenger flow density sequences of multiple preset monitoring points, and analyze and obtain multiple predicted commodity shortage times; a delivery path enumeration module, which is used to enumerate deliverable paths based on the real-time position coordinates of multiple idle unmanned delivery vehicles in the three-dimensional simulation space of the campus, and generate multiple initial delivery plans; a driving path control module, which is used to use the several predicted commodity shortage times as delivery constraints, minimize the distribution time difference and minimize the pick-up and delivery distance as the comprehensive optimization goals, evaluate the advantages and disadvantages of the multiple initial delivery plans according to the multiple passenger flow density sequences, output the optimal delivery plan, and control the driving paths of the multiple idle unmanned delivery vehicles.

[0013] The embodiments of the present invention include the following advantages:

[0014] Based on the network distribution map and line structure diagram of the target park, a three-dimensional simulation space of the park is built, wherein the network includes multiple storage warehouses and several supply points; then, several commodity consumption data sequences of the several supply points in a preset historical time zone and multiple passenger flow density sequences of multiple preset monitoring points are monitored and obtained, and several predicted commodity shortage times are analyzed and obtained; then, within the three-dimensional simulation space of the park, the deliverable paths are enumerated based on the real-time position coordinates of multiple idle unmanned delivery vehicles, and multiple initial delivery plans are generated; finally, with the several predicted commodity shortage times as delivery constraints and minimizing the delivery time difference and minimizing the pickup and delivery distance as the comprehensive optimization goals, the multiple initial delivery plans are evaluated according to the multiple passenger flow density sequences, the optimal delivery plan is output, and the driving path control of the multiple idle unmanned delivery vehicles is performed. In other words, by integrating multimodal information such as passenger flow perception, commodity consumption prediction and vehicle status, constructing a three-dimensional simulation space for the scenic area, analyzing the shortage time of supply points in real time, and generating and optimizing multi-vehicle delivery routes, it is possible to significantly improve the real-time collaborative scheduling of unmanned delivery vehicles, the accuracy of route planning and the overall delivery efficiency, and achieve intelligent, efficient and safe park-level unmanned delivery control effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flowchart of the steps of a path control method for unmanned driving in a park combined with multimodal perception of the present invention;

[0016] Figure 2 This is a structural schematic diagram of a campus unmanned driving path control system combined with multimodal perception in the present invention.

[0017] Description of reference numerals:

[0018] Park simulation space construction module 11, commodity shortage time prediction module 12, distribution path enumeration module 13, driving path control module 14. DETAILED DESCRIPTION

[0019] By providing a method and system for controlling unmanned driving paths in a park that incorporates multimodal sensing, this invention addresses the problems inherent in traditional route planning methods for unmanned delivery vehicles in tourist attractions, which suffer from untimely dispatch responses, inaccurate routes, and low vehicle resource utilization, due to the inability to dynamically adjust the coordinated dispatch of multiple vehicles based on the actual operating status of the scenic area. By integrating multimodal information such as passenger flow perception, commodity consumption prediction, and vehicle status, a three-dimensional simulation space for the scenic area is constructed, supply point material shortages are analyzed in real time, and multi-vehicle delivery routes are generated and optimized. This significantly improves the real-time coordinated dispatch of unmanned delivery vehicles, the accuracy of route planning, and overall delivery efficiency, achieving intelligent, efficient, and secure park-level unmanned delivery control.

[0020] Below, the technical solutions of the present invention will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments described herein. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. It should also be noted that, for the convenience of description, only the parts related to the present invention, rather than all, are shown in the accompanying drawings.

[0021] For example, see the attached Figure 1 The present invention provides a path control method for unmanned driving in a park combined with multimodal perception, which is applied to a path control system for unmanned driving in a park combined with multimodal perception, and specifically includes the following steps:

[0022] S10: Based on the network distribution map and line structure map of the target park, a three-dimensional simulation space of the park is constructed, wherein the network includes multiple storage warehouses and several supply points.

[0023] Specifically, first, obtain the network distribution map and line structure map of the target park. The target park is a tourist attraction. The network refers to the important functional nodes in the park, which are used to store materials, make supplies or provide services to tourists. According to different scenarios, the network includes multiple storage warehouses and several supply points. The storage warehouse is used to store and distribute materials (such as beverages, souvenirs, snacks, etc.); the supply point refers to the place where tourists may need material supplies, which are usually distributed in tourist-concentrated areas (such as near scenic spots, rest areas, catering spots, etc.). These points need to be set according to actual needs. The route structure diagram is an image drawn based on the road network of the park. It is mainly used to define the driving path of unmanned delivery vehicles and evaluate the transportation efficiency of different paths. The route structure diagram contains the length information of the internal roads of the park, road type (such as pedestrian path, driveway), curvature, slope and other attribute information, which is used to truly restore the traffic terrain conditions of the park. Among them, curvature refers to the degree of twists and turns of the road. Roads with large curves may affect the driving speed of unmanned delivery vehicles, so their impact on driving efficiency needs to be considered during planning; the slope of the road (uphill or downhill) will affect the power consumption and speed of the unmanned delivery vehicle, especially electric delivery vehicles may require more power when going uphill, affecting the overall travel time.

[0024] Next, based on the network distribution map and route structure diagram, computer graphics and spatial modeling techniques are used to convert the park's map, roads, buildings, and facilities into a three-dimensional model. This model accurately recreates the road's various types, curvatures, and slopes in a virtual environment. The 3D simulation space is more than just a static display; it also simulates the driving state of unmanned delivery vehicles. By inputting information such as delivery requirements, vehicle locations, and road conditions, the simulation system simulates the vehicle's actual driving path and provides support for subsequent route optimization and scheduling decisions. The 3D simulation space is generated to realistically simulate the park's traffic, road conditions, supply points, warehouses, and other real-world conditions, enabling the system to perform various simulations and optimizations. Building a 3D simulation space based on the target park's network distribution map and route structure diagram enables accurate route planning and real-time scheduling support for the park's unmanned delivery system. By simulating the park's actual operating environment (such as road type, slope, and passenger density), this simulation space not only provides data support for delivery vehicles but also provides a powerful tool for subsequent multi-vehicle coordinated scheduling, route optimization, and system optimization.

[0025] S20: monitoring and obtaining a plurality of commodity consumption data sequences of the plurality of supply points within a preset historical time zone, and a plurality of passenger flow density sequences of a plurality of preset monitoring points, and analyzing and obtaining a plurality of predicted commodity shortage times.

[0026] Furthermore, step S20 of the present invention further includes:

[0027] S21: Within the preset historical time zone, continuously monitor and obtain the commodity consumption ratio of the supply points, construct a commodity consumption ratio sequence as a commodity consumption data sequence, and obtain several commodity consumption data sequences; S22: continuously monitor and obtain the passenger flow density of the preset monitoring points, and obtain multiple passenger flow density sequences; S23: perform correlation analysis on the supply points and the preset monitoring points according to the preset coverage range, determine several associated monitoring point sets of the several supply points, and map and obtain several associated passenger flow density sequence sets.

[0028] Specifically, in actual operations, each supply point in a scenic area (such as beverage counters and souvenir sales points) experiences a process of product consumption. To understand material demand in real time, it is necessary to continuously monitor the product consumption at each supply point and generate a data series on product consumption. First, a historical time zone (such as the last hour) is selected as the monitoring range, based on which the consumption patterns of the products are analyzed. The selection of a monitoring time zone helps the system understand the cyclical nature of material consumption and predict future demand based on this pattern. Data is collected regularly within this historical time zone, for example, every 5 minutes (or other appropriate intervals) to obtain dynamic, real-time product consumption data. Regular monitoring helps the system capture changing trends in product consumption. For each supply point, the ratio of recorded product consumption to the maximum capacity of that supply point is defined as the product consumption ratio. This ratio represents the speed of product consumption at each supply point and can reflect changes in demand. Within the entire monitoring time zone, the product consumption ratios monitored at each time are arranged in chronological order to form a product consumption ratio series for multiple supply points. These series can be used for subsequent product shortage prediction.

[0029] On the other hand, in tourist attractions, the dynamic changes in passenger flow are a significant factor influencing delivery routes and scheduling decisions. To improve delivery efficiency and avoid conflicts with tourists, it is necessary to continuously monitor passenger flow density at each pre-set monitoring point and generate corresponding passenger flow density series. First, multiple monitoring points are set up according to the park's layout. These points are usually located in areas with high or critical tourist traffic, such as the scenic area entrance, popular attractions, dining areas, and rest areas. Each monitoring point is responsible for collecting passenger flow information within that area. Then, at each monitoring point, sensors (such as infrared sensors, video surveillance, and Wi-Fi positioning) are used to obtain real-time passenger flow data for that area. Passenger flow density refers to the number of tourists passing through a certain area per unit time, usually expressed as the number of tourists per square meter. The passenger flow density value reflects the degree of congestion in the area. High-density areas may affect the speed of delivery vehicles or cause traffic jams. Passenger flow density data at each monitoring point is regularly monitored and recorded to form a time series, similar to a commodity consumption data series. These passenger flow density series can reveal the fluctuations in passenger flow in various areas of the park over different time periods.

[0030] Then, an association analysis is performed between the supply points and the preset monitoring points according to the preset coverage range. The coverage range refers to the area associated with each supply point, typically represented by a circular area with a radius set according to actual needs (e.g., 500 meters). This range determines which monitoring points' passenger flow density data should be associated with a specific supply point. Each supply point has a preset coverage range, typically a circular area centered at the supply point with a radius of 500 meters (or other suitable value). For each supply point, the actual distances between all preset monitoring points and the supply point are calculated (typically using Euclidean distance). The calculated distances are then compared with the preset coverage range. If the distance of a monitoring point falls within the coverage range of the supply point (e.g., within a 500-meter radius), the monitoring point is considered associated with the supply point. Through the aforementioned association analysis, a number of associated monitoring points are determined for each supply point, and corresponding passenger flow density sequences are obtained based on these monitoring points, resulting in a number of associated passenger flow density sequence sets.

[0031] S24: Constructing a product shortage predictor based on a generative adversarial network, performing product shortage prediction based on the real-time remaining product ratios of the supply points, combining the product consumption data sequences and the associated customer flow density sequence sets, and outputting a number of predicted product shortage times.

[0032] Furthermore, step S24 of the present invention further includes:

[0033] S241: Based on the historical operation monitoring records of the target park, a sample remaining commodity ratio set, a sample commodity consumption data sequence set, and multiple sample associated passenger flow density sequence sets are collected, and the historical commodity shortage duration is obtained and set as the sample commodity shortage time to obtain the sample commodity shortage time set; S242: Using the sample remaining commodity ratio set, the sample commodity consumption data sequence set, and multiple sample associated passenger flow density sequence sets as input, and using the sample commodity shortage time set as supervision, the generator and discriminator of the generative adversarial network are supervised trained until convergence to obtain the commodity shortage predictor.

[0034] Specifically, based on the historical operational monitoring records of the target park, we first collect a set of sample remaining product ratios, a set of sample product consumption data series, and multiple sets of sample associated passenger flow density series. For each supply point, the system records its remaining product ratio during the historical period—that is, the ratio of the number of remaining products at the supply point to its maximum capacity. These data reflect the temporal trend of product inventory and can be used to predict shortage risks. Furthermore, we need to collect a series of product consumption data, which shows how the product consumption at each supply point changes over time. These data help analyze the speed of product consumption and consumption trends. The passenger flow density series of the surrounding monitoring points of each supply point also need to be used as input data. The passenger flow density series can provide important information about tourist flow and density, helping to predict demand changes in specific areas. Then, we obtain the historical product shortage duration and set it as the sample product shortage time. The product shortage time refers to the specific time when a supply point experienced product shortages during a certain historical period in the past, reflecting the imbalance between product consumption and supply. This results in a sample product shortage time set.

[0035] Next, supervised training of the generator and discriminator of a generative adversarial network is performed using the sample remaining product ratio set, sample product consumption data sequence set, and multiple sample associated customer flow density sequence sets as input, and the sample product shortage time set as supervision. A generative adversarial network consists of two parts: a generator and a discriminator. The generator receives input data (such as remaining product ratio, product consumption data, and customer flow density data) and attempts to generate a predicted product shortage time. This is the generative process of the generative adversarial network, which attempts to infer the product shortage time from the input data. The discriminator's task is to evaluate the accuracy of the generator's prediction. It is trained based on a historical sample product shortage time set, learning how to judge the difference between the generator's predicted time and the actual shortage time. The discriminator outputs a probability value indicating whether the generator's prediction is consistent with the actual situation. During the training process, the sample product shortage time set serves as a supervision signal to guide the training of the generator and discriminator. The generator learns from sample data to generate predictions that are as close as possible to the actual shortage time, while the discriminator evaluates the generator's output and the true labels. The generator and discriminator are continuously optimized through adversarial training, with the generator attempting to improve its generated product shortage time predictions and the discriminator striving to improve its ability to distinguish between actual shortage times and generated predictions. Training continues until the generator and discriminator reach convergence, meaning the generator can generate predictions that are very close to the actual shortage time, while the discriminator cannot significantly distinguish between real data and generated data. After multiple rounds of adversarial training, the generative adversarial network converges, and the parameters of the generator and discriminator become stable. At this point, the generator can accurately predict product shortage times based on historical input data (such as product consumption, remaining product ratio, and customer flow density). The trained generator is an efficient product shortage predictor that can predict future product shortage risks based on real-time data within the park, providing proactive decision support for the distribution system and ensuring sufficient supply at supply points.

[0036] Then, the commodity shortage predictor is used to predict commodity shortages based on the real-time remaining commodity ratios of the supply points, the commodity consumption data sequences and the associated customer flow density sequence sets, and outputs a plurality of predicted commodity shortage times.

[0037] S30: In the three-dimensional simulation space of the park, enumerate possible delivery paths based on the real-time position coordinates of multiple idle unmanned delivery vehicles to generate multiple initial delivery plans.

[0038] Specifically, first, the coordinate information of each unmanned delivery vehicle is acquired in real time through a positioning system (such as GPS, RTK, LiDAR, and visual recognition). The coordinates of each unmanned delivery vehicle are continuously updated to reflect its current position and speed. Based on this real-time location data, the status and availability of each delivery vehicle are determined, resulting in multiple real-time location coordinates of multiple idle unmanned delivery vehicles. Next, within the three-dimensional simulation space of the park, possible delivery paths are enumerated based on these multiple real-time location coordinates. By analyzing the location of the unmanned delivery vehicle and the three-dimensional structure of the park, possible paths for the delivery vehicle to reach various destinations from its current location are calculated. This path enumeration generates multiple possible delivery paths that can cover different areas and supply points within the entire park, providing alternative options for subsequent delivery task selection.

[0039] S40: Taking the predicted product shortage times as delivery constraints, minimizing the delivery time difference and minimizing the pickup and delivery distance as the comprehensive optimization goal, the multiple initial delivery plans are evaluated according to the multiple passenger flow density sequences, the optimal delivery plan is output, and the driving path control of the multiple idle unmanned delivery vehicles is performed.

[0040] Furthermore, step S40 of the present invention further includes:

[0041] S41: randomly selecting a first delivery plan from the multiple initial delivery plans, and randomly selecting a first unmanned delivery vehicle, and obtaining a first delivery path and a first delivery commodity weight of the first unmanned delivery vehicle in the first delivery plan; S42: obtaining first path structure information based on the first delivery path mapping, wherein the first path structure information includes first pickup path structure information and first delivery path structure information, and the path structure information includes at least line length, road type, slope and curvature, and the delivery path includes one or more delivery supply points; S43: obtaining a first coverage passenger flow density sequence set based on the first delivery path mapping, wherein the first coverage passenger flow density sequence set includes a first pickup coverage passenger flow density sequence set and a first delivery coverage passenger flow density sequence set; S44: in the three-dimensional simulation space of the park, according to the first pickup path structure information and the first pickup coverage passenger flow density sequence set, The first delivery route structure information, the first delivery coverage passenger flow density sequence set and the first delivery commodity weight are used to predict the pickup time, and the first predicted pickup time is output; S45: the delivery time is predicted according to the first delivery route structure information, the first delivery coverage passenger flow density sequence set and the first delivery commodity weight, and the first predicted delivery time is output; S46: the first predicted loading time is determined based on the first delivery commodity weight analysis, and the first arrival time is obtained by summing the first predicted pickup time and the first predicted delivery time, wherein the first arrival time includes the delivery arrival time of one or more delivery supply points; S47: the multiple arrival times of the multiple delivery routes in the first delivery plan are analyzed in turn, and the multiple delivery arrival times of the multiple supply points are determined; S48: whether the multiple delivery arrival times meet the multiple predicted commodity shortage times, if so, the first delivery plan is set as an optional delivery plan, and multiple optional delivery plans are obtained by analysis in turn.

[0042] Specifically, first, any one of the multiple initial delivery plans is randomly selected as the first delivery plan, and a first unmanned delivery vehicle is randomly selected from the multiple unmanned delivery vehicles in the first delivery plan. The first delivery path and the first delivery product weight of the first unmanned delivery vehicle in the first delivery plan are obtained. Each delivery vehicle performs different delivery tasks. Therefore, after selecting the first delivery vehicle, an analysis can be performed based on the specific task of the vehicle. The first delivery product weight refers to the total weight of the goods required to be delivered by the selected unmanned delivery vehicle. This information is crucial for subsequent path optimization, vehicle scheduling, energy consumption calculation, etc. Next, based on the first delivery path mapping, first path structure information is obtained. The first path structure information includes first pickup path structure information and first delivery path structure information. The path structure information includes at least route length, road type, slope, and curvature. The delivery path includes one or more delivery supply points. The pickup path structure information refers to the path information from the warehouse or supply point to the area where the delivery vehicle needs to pick up the goods. This part of the path mainly involves the driving path from the starting point to the pickup point. The delivery path structure information refers to the path information from the pickup point or warehouse to the destination supply point. This part of the path involves the driving route from the starting point to each delivery destination. A delivery route can contain multiple supply points, that is, a route may need to pass through multiple delivery points to complete the delivery task.

[0043] In the park, the impact of passenger flow density on the delivery task is very important, because areas with large passenger flow will increase the driving difficulty of unmanned delivery vehicles or prolong the delivery time; then, based on the first delivery path mapping, a first covering passenger flow density sequence set is obtained, wherein the first covering passenger flow density sequence set includes a first pickup covering passenger flow density sequence set and a first delivery covering passenger flow density sequence set. The first pickup covering passenger flow density sequence set refers to the passenger flow density sequence of all areas passed by the delivery vehicle starting from the warehouse or supply point. By recording the passenger flow density of each area passed from the starting point to the pickup point, its impact on the path is evaluated; the first delivery covering passenger flow density sequence set refers to the passenger flow density sequence of each area passed by the delivery vehicle on the path from the pickup point to each delivery supply point. The passenger flow density of different delivery points will affect the driving speed and path selection of the delivery vehicle, especially in scenic areas with dense tourists.

[0044] Then, in the three-dimensional simulation space of the park, the pickup time is predicted based on the first pickup path structure information and the first pickup coverage passenger flow density sequence set. The pickup time can be predicted through a machine learning model or a simulation model. For example, based on historical data and path structure parameters (such as road type, slope, curvature, etc.) and passenger flow density sequence, a regression model is trained to predict the driving time; through the calculation of the model, the system will output the first predicted pickup time, which refers to the time required for the delivery vehicle to start from the starting point (warehouse or supply point), arrive at the pickup point and complete the pickup task. On the other hand, delivery time is predicted based on the first delivery route structure information, the first delivery coverage passenger flow density sequence set, and the first delivered product weight. That is, the delivery time is predicted using a machine learning model or a physical modeling method based on multiple factors such as the delivery route structure information, the passenger flow density sequence, and the product weight. For example, a regression model is trained based on historical data and route structure parameters (such as road type, slope, curvature, etc.) and the passenger flow density sequence to predict the delivery time. If there are multiple delivery points, the total delivery time is calculated by comprehensively considering the following factors: first, the time from the starting point to the first delivery point is predicted, taking into account factors such as path length, road type, slope, curvature, and passenger flow density. After each delivery, the delivery vehicle needs to unload the goods. The unloading time is usually a known fixed time. Since the weight of the delivery vehicle is reduced after unloading, the speed of subsequent deliveries will be relatively faster. Therefore, the time of subsequent deliveries will be calculated based on the arrival time and unloading time of the previous delivery point, as well as the complexity of the delivery route. For each subsequent delivery point, the time can be calculated recursively, taking into account the impact of the gradual reduction in vehicle load. Through these steps, the system ultimately outputs the first predicted delivery time, which is the time it will take for the autonomous delivery vehicle to complete all deliveries. This time factored in dynamic factors like route structure, customer flow density, and product weight, and is cumulative for multiple delivery points.

[0045] Then, a first predicted loading time is determined based on the weight analysis of the first delivered goods, and the predicted loading time based on the goods weight and loading efficiency is output through calculation of the model; then, the first predicted loading time, the first predicted pickup time and the first predicted delivery time are summed to obtain a first arrival time, wherein the first arrival time includes the delivery arrival time of one or more delivery supply points, that is, if there are multiple delivery points, the delivery time of each delivery point will be calculated in sequence.

[0046] Using the same method used to calculate the first arrival time, multiple arrival times for multiple delivery routes in the first delivery plan are sequentially analyzed and obtained, and multiple delivery arrival times for multiple supply points are determined. A determination is then made as to whether the multiple delivery arrival times satisfy the multiple predicted product shortage times. If any of the delivery arrival times is less than the corresponding predicted product shortage time, the multiple predicted product shortage times are satisfied, and the first delivery plan is designated as an optional delivery plan. Multiple optional delivery plans that satisfy the multiple predicted product shortage times are then sequentially analyzed.

[0047] S49: Taking minimizing the time difference in delivery and minimizing the distance between pickup and delivery as the comprehensive optimization goal, the multiple optional delivery plans are evaluated for their pros and cons, and the optimal delivery plan is output.

[0048] Furthermore, step S49 of the present invention further includes:

[0049] S491: Based on the several predicted commodity shortage times, deviation calculations are performed on the multiple delivery arrival time sets of the multiple optional distribution plans respectively, multiple time deviation sets are determined, and the sum is used to obtain multiple delivery time deviations; S492: Based on the several predicted commodity shortage times, deviation amplitude analysis is performed based on the multiple time deviation sets, and multiple delivery time deviation amplitude means are calculated; S493: Based on the multiple delivery time deviations and the multiple delivery time deviation amplitude means, multiple distribution time difference coefficients are determined, wherein the distribution time difference coefficient is negatively correlated with the delivery time deviation, and positively correlated with the delivery time deviation amplitude mean.

[0050] Specifically, first, using the predicted product shortage times as a benchmark, deviations are calculated for each of the multiple delivery arrival time sets for the multiple optional distribution plans. This involves subtracting the delivery arrival time of the corresponding supply point from the predicted product shortage time, and setting the difference between the two as the time deviation to determine multiple time deviation sets. These deviations are then summed to obtain the total delivery time deviation for the plan. Next, using the predicted product shortage times as a benchmark, a deviation amplitude analysis is performed on the multiple time deviation sets. The deviation amplitude is the ratio of the time deviation to the time interval between the predicted product shortage time and the current time interval. The average of the multiple delivery time deviation amplitudes is calculated to measure the delivery consistency and uniformity within each distribution plan. Furthermore, multiple distribution time difference coefficients are determined based on the evaluation of the multiple delivery time deviations and the multiple delivery time deviation amplitude averages, wherein the distribution time difference coefficient is negatively correlated with the delivery time deviation, and positively correlated with the delivery time deviation amplitude average, that is, the smaller the delivery time deviation, the closer it is to the shortage time point, and the more timely the delivery; the smaller the delivery time deviation amplitude average, the more coordinated and balanced the distribution plan is.

[0051] S494: Evaluate the multiple optional delivery plans based on the multiple pickup and delivery distance coefficients and the multiple delivery time difference coefficients, and output the optimal delivery plan.

[0052] Furthermore, step S494 of the present invention further includes:

[0053] S4941: Configure the delivery distance weight according to the weight of the goods, wherein the delivery distance weight is positively correlated with the weight of the goods, and the delivery distance weight is greater than or equal to 1. When the weight of the goods is 0, the delivery distance weight is 1; S4942: Randomly select the first optional delivery plan, and obtain the first pickup distance and the first delivery distance of the first optional delivery plan; S4943: Compensate the first delivery distance according to the delivery distance weight to obtain the first compensated delivery distance, and calculate the first pickup and delivery distance in combination with the first pickup distance, set it as the first pickup and delivery distance coefficient, and analyze in sequence to obtain multiple pickup and delivery distance coefficients.

[0054] Specifically, first, a delivery distance weight is assigned based on the weight of the goods. The delivery distance weight is positively correlated with the weight of the goods. This means that the cost of the delivery distance depends not only on the distance itself but also on the weight of the goods being transported. Heavier goods increase energy consumption (higher power consumption), driving speed (lower speed, increased time), and safety (more complex braking, turning, and climbing). When the weight of the goods is 0, the delivery distance weight is 1; heavier goods increase the delivery distance weight. Next, a first optional delivery option is randomly selected from multiple delivery options. The first pickup distance and first delivery distance for this first optional delivery option are obtained. If the delivery route involves multiple supply points (for example, delivering A first, then B), each segment is independently weighted based on the weight of the goods carried at the start of that segment. The final first compensated delivery distance is the sum of the compensated segments. The first delivery distance is then compensated according to the delivery distance weight, and the product of the two is set as the first compensated delivery distance; the first compensated delivery distance and the first pickup distance are then added together to obtain the first pickup delivery distance, which is set as the first pickup delivery distance coefficient. Multiple pickup delivery distance coefficients are then analyzed in turn. This coefficient reflects the "weighted path cost" required to complete this delivery task, which is closer to the time, energy consumption, and scheduling burden during actual execution.

[0055] Furthermore, step S494 of the present invention further includes:

[0056] S4944: Perform dimensionless processing on the multiple pick-up and delivery distance coefficients and the multiple distribution time difference coefficients to obtain multiple standard pick-up and delivery distance coefficients and multiple standard distribution time difference coefficients; S4945: Evaluate the fitness of the scheme based on the multiple standard pick-up and delivery distance coefficients and the multiple standard distribution time difference coefficients, output multiple fitnesses, and select the optional delivery scheme with the maximum fitness as the optimal delivery scheme, wherein the fitness is negatively correlated with the standard pick-up and delivery distance coefficient and the standard distribution time difference coefficient.

[0057] Specifically, first, the multiple pick-up and delivery distance coefficients and the multiple distribution time difference coefficients are dimensionlessly processed, that is, the two types of data are standardized, the dimensionality effect is eliminated, and they are converted into a unified measurement to obtain multiple standard pick-up and delivery distance coefficients and multiple standard distribution time difference coefficients; then, the fitness of the scheme is evaluated based on the multiple standard pick-up and delivery distance coefficients and the multiple standard distribution time difference coefficients, wherein the fitness is negatively correlated with the standard pick-up and delivery distance coefficient and the standard distribution time difference coefficient, that is, the smaller the better, and the smaller the fitness, the higher the fitness. The weight can be adjusted according to the different emphasis on distance cost or time coordination; finally, the optional delivery scheme with the maximum fitness is selected as the optimal delivery scheme.

[0058] In summary, the present invention provides a method for controlling the path of an unmanned driving system in a park combined with multimodal perception, which has the following technical effects:

[0059] Based on the network distribution map and line structure diagram of the target park, a three-dimensional simulation space of the park is built, wherein the network includes multiple storage warehouses and several supply points; then, several commodity consumption data sequences of the several supply points in a preset historical time zone and multiple passenger flow density sequences of multiple preset monitoring points are monitored and obtained, and several predicted commodity shortage times are analyzed and obtained; then, within the three-dimensional simulation space of the park, the deliverable paths are enumerated based on the real-time position coordinates of multiple idle unmanned delivery vehicles, and multiple initial delivery plans are generated; finally, with the several predicted commodity shortage times as delivery constraints and minimizing the delivery time difference and minimizing the pickup and delivery distance as the comprehensive optimization goals, the multiple initial delivery plans are evaluated according to the multiple passenger flow density sequences, the optimal delivery plan is output, and the driving path control of the multiple idle unmanned delivery vehicles is performed. In other words, by integrating multimodal information such as passenger flow perception, commodity consumption prediction and vehicle status, constructing a three-dimensional simulation space for the scenic area, analyzing the shortage time of supply points in real time, and generating and optimizing multi-vehicle delivery routes, it is possible to significantly improve the real-time collaborative scheduling of unmanned delivery vehicles, the accuracy of route planning and the overall delivery efficiency, and achieve intelligent, efficient and safe park-level unmanned delivery control effects.

[0060] In the second embodiment, based on the same inventive concept as the method for controlling a path for an unmanned driving vehicle in a park combined with multimodal sensing in the aforementioned embodiment, the present invention further provides a path control system for an unmanned driving vehicle in a park combined with multimodal sensing, as shown in the attached figure. Figure 2 , including: a park simulation space construction module 11, which is used to build a three-dimensional simulation space of the park based on the network distribution map and line structure map of the target park, wherein the network points include multiple storage warehouses and several supply points; a commodity shortage time prediction module 12, which is used to monitor and obtain multiple commodity consumption data sequences of the several supply points in a preset historical time zone, and multiple passenger flow density sequences of multiple preset monitoring points, and analyze and obtain multiple predicted commodity shortage times; a delivery path enumeration module 13, which is used to enumerate deliverable paths based on the real-time position coordinates of multiple idle unmanned delivery vehicles in the three-dimensional simulation space of the park, and generate multiple initial delivery plans; a driving path control module 14, which is used to use the several predicted commodity shortage times as delivery constraints, minimize the distribution time difference and minimize the pick-up and delivery distance as the comprehensive optimization goals, evaluate the advantages and disadvantages of the multiple initial delivery plans according to the multiple passenger flow density sequences, output the optimal delivery plan, and control the driving paths of the multiple idle unmanned delivery vehicles.

[0061] Furthermore, the park unmanned driving path control system combined with multimodal perception is also used to: continuously monitor and obtain the commodity consumption ratio of the supply point within the preset historical time zone, construct a commodity consumption ratio sequence as a commodity consumption data sequence, and obtain several commodity consumption data sequences; continuously monitor and obtain the passenger flow density of the preset monitoring points, and obtain multiple passenger flow density sequences; perform association analysis on the supply points and the preset monitoring points according to the preset coverage range, determine several associated monitoring point sets of the several supply points, and map to obtain several associated passenger flow density sequence sets; construct a commodity shortage predictor based on a generative adversarial network, and perform commodity shortage prediction based on several real-time remaining commodity ratios of the several supply points, combined with the several commodity consumption data sequences and the several associated passenger flow density sequence sets, and output several predicted commodity shortage times.

[0062] Furthermore, the park unmanned driving path control system combined with multimodal perception is also used to: collect a sample remaining commodity ratio set, a sample commodity consumption data sequence set and multiple sample associated passenger flow density sequence sets based on the historical operation monitoring records of the target park, and obtain the historical commodity shortage duration as the sample commodity shortage time to obtain a sample commodity shortage time set; use the sample remaining commodity ratio set, the sample commodity consumption data sequence set and multiple sample associated passenger flow density sequence sets as input, and use the sample commodity shortage time set as supervision to perform supervised training on the generator and discriminator of the generative adversarial network until convergence, so as to obtain the commodity shortage predictor.

[0063] Furthermore, the park unmanned driving path control system combined with multimodal perception is also used to: randomly select a first distribution plan from the multiple initial distribution plans, and randomly select a first unmanned delivery vehicle, to obtain the first distribution path and the first distribution commodity weight of the first unmanned delivery vehicle in the first distribution plan; obtain first path structure information based on the first distribution path mapping, wherein the first path structure information includes first pickup path structure information and first delivery path structure information, and the path structure information includes at least line length, road type, slope and curvature, and the delivery path contains one or more delivery supply points; obtain a first covering passenger flow density sequence set based on the first delivery path mapping, wherein the first covering passenger flow density sequence set includes a first pickup coverage passenger flow density sequence set and a first delivery coverage passenger flow density sequence set; in the three-dimensional simulation space of the park, the pickup time is calculated according to the first pickup path structure information and the first pickup coverage passenger flow density sequence set. Prediction, output a first predicted pickup time; predict the delivery time according to the first delivery route structure information, the first delivery coverage passenger flow density sequence set and the first delivery commodity weight, and output the first predicted delivery time; determine the first predicted loading time based on the first delivery commodity weight analysis, and obtain the first arrival time by combining the first predicted pickup time and the first predicted delivery time, wherein the first arrival time includes the delivery arrival time of one or more delivery supply points; sequentially analyze and obtain multiple arrival times of multiple delivery routes in the first delivery plan, and determine multiple delivery arrival times of multiple supply points; determine whether the multiple delivery arrival times meet the multiple predicted commodity shortage times, and if so, set the first delivery plan as an optional delivery plan, and sequentially analyze to obtain multiple optional delivery plans; with minimizing the distribution time difference and minimizing the pickup and delivery distance as the comprehensive optimization goal, evaluate the advantages and disadvantages of the multiple optional delivery plans and output the optimal delivery plan.

[0064] Furthermore, the campus unmanned driving path control system combined with multimodal perception is also used to: based on the several predicted commodity shortage times, perform deviation calculations on the multiple delivery arrival time sets of the multiple optional distribution plans, determine multiple time deviation sets, and sum up to obtain multiple delivery time deviations; based on the several predicted commodity shortage times, perform deviation amplitude analysis based on the multiple time deviation sets, and calculate multiple delivery time deviation amplitude means; determine multiple distribution time difference coefficients based on the multiple delivery time deviations and the multiple delivery time deviation amplitude means, wherein the distribution time difference coefficient is negatively correlated with the delivery time deviation and positively correlated with the delivery time deviation amplitude mean; evaluate the advantages and disadvantages of the multiple optional distribution plans based on multiple delivery distance coefficients and the multiple distribution time difference coefficients, and output the optimal distribution plan.

[0065] Furthermore, the campus unmanned driving path control system combined with multimodal perception is also used to: configure a delivery distance weight according to the weight of the goods, wherein the delivery distance weight is positively correlated with the weight of the goods, the delivery distance weight is greater than or equal to 1, and when the weight of the goods is 0, the delivery distance weight is 1; randomly select a first optional delivery plan to obtain a first pickup distance and a first delivery distance of the first optional delivery plan; compensate the first delivery distance according to the delivery distance weight to obtain a first compensated delivery distance, calculate the first pickup and delivery distance in combination with the first pickup distance, set it as the first pickup and delivery distance coefficient, and analyze in sequence to obtain multiple pickup and delivery distance coefficients.

[0066] Furthermore, the campus unmanned driving path control system combined with multimodal perception is also used to: perform dimensionless processing on the multiple pick-up and delivery distance coefficients and the multiple distribution time difference coefficients to obtain multiple standard pick-up and delivery distance coefficients and multiple standard distribution time difference coefficients; perform scheme fitness evaluation based on the multiple standard pick-up and delivery distance coefficients and the multiple standard distribution time difference coefficients, output multiple fitnesses, and select the optional distribution scheme with the maximum fitness as the optimal distribution scheme, wherein the fitness is negatively correlated with the standard pick-up and delivery distance coefficient and the standard distribution time difference coefficient.

[0067] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The method and specific examples of the path control method for an unmanned driving path in a park combined with multimodal perception in the aforementioned embodiment 1 are also applicable to the path control system for an unmanned driving path in a park combined with multimodal perception in this embodiment. Through the aforementioned detailed description of the path control method for an unmanned driving path in a park combined with multimodal perception, those skilled in the art can clearly understand the path control system for an unmanned driving path in a park combined with multimodal perception in this embodiment, so for the sake of brevity of the specification, it will not be described in detail here. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0068] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

[0069] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalents, the present invention is intended to include these modifications and variations.

Claims

1. The path control method for unmanned driving in a park combined with multimodal perception is characterized by: Methods include: Based on the network distribution map and line structure diagram of the target park, a three-dimensional simulation space of the park is built. The network includes multiple storage warehouses and several supply points. Monitoring and obtaining a plurality of commodity consumption data sequences of the plurality of supply points within a preset historical time zone, as well as a plurality of customer flow density sequences of a plurality of preset monitoring points, and analyzing and obtaining a plurality of predicted commodity shortage times; In the three-dimensional simulation space of the park, possible delivery paths are enumerated based on the real-time position coordinates of multiple idle unmanned delivery vehicles to generate multiple initial delivery plans; Taking the predicted product shortage times as delivery constraints and minimizing the delivery time difference and the pickup and delivery distance as the comprehensive optimization goal, the multiple initial delivery plans are evaluated based on the multiple passenger flow density sequences, the optimal delivery plan is output, and the driving paths of the multiple idle unmanned delivery vehicles are controlled; The method uses the predicted product shortage times as delivery constraints, minimizes the delivery time difference and minimizes the pickup and delivery distance as comprehensive optimization goals, evaluates the multiple initial delivery plans based on the multiple customer flow density sequences, and outputs the optimal delivery plan, including: Randomly selecting a first delivery plan from the multiple initial delivery plans, and randomly selecting a first unmanned delivery vehicle, and obtaining a first delivery route and a first delivery commodity weight of the first unmanned delivery vehicle in the first delivery plan; Acquiring first route structure information based on the first delivery route mapping, wherein the first route structure information includes first pickup route structure information and first delivery route structure information, the route structure information including at least route length, road type, slope, and curvature, and the delivery route includes one or more delivery supply points; Acquire a first covering passenger flow density sequence set based on the first delivery path mapping, wherein the first covering passenger flow density sequence set includes a first pickup covering passenger flow density sequence set and a first delivery covering passenger flow density sequence set; In the three-dimensional simulation space of the park, predicting the pickup time according to the first pickup path structure information and the first pickup coverage passenger flow density sequence set, and outputting a first predicted pickup time; Predicting the delivery time based on the first delivery route structure information, the first delivery coverage passenger flow density sequence set, and the weight of the first delivery product, and outputting a first predicted delivery time; Determining a first predicted loading time based on the weight of the first delivered product, and summing the first predicted pickup time and the first predicted delivery time to obtain a first arrival time, wherein the first arrival time includes the delivery arrival time of one or more delivery supply points; Analyze and obtain multiple arrival times of multiple delivery routes in the first delivery plan in sequence, and determine multiple delivery arrival times of multiple supply points; determining whether the plurality of delivery arrival times satisfy the plurality of predicted product shortage times; if so, setting the first delivery plan as an optional delivery plan, and sequentially analyzing to obtain a plurality of optional delivery plans; With the comprehensive optimization goal of minimizing the time difference in distribution and minimizing the distance between pick-up and delivery, the multiple optional delivery plans are evaluated and the optimal delivery plan is output.

2. The method for controlling the path of an unmanned driving vehicle in a park combined with multimodal perception according to claim 1 is characterized in that: Monitoring and obtaining a plurality of commodity consumption data sequences of the plurality of supply points within a preset historical time zone, as well as a plurality of customer flow density sequences of a plurality of preset monitoring points, and analyzing and obtaining a plurality of predicted commodity shortage times, including: In the preset historical time zone, continuously monitoring and obtaining the commodity consumption ratio of the supply point, constructing a commodity consumption ratio sequence as a commodity consumption data sequence, and obtaining a plurality of commodity consumption data sequences; Continuously monitor and obtain the passenger flow density of preset monitoring points to obtain multiple passenger flow density sequences; Performing correlation analysis on the supply points and the preset monitoring points according to the preset coverage range, determining a plurality of associated monitoring point sets of the plurality of supply points, and mapping to obtain a plurality of associated passenger flow density sequence sets; A product shortage predictor is constructed based on a generative adversarial network. Product shortage prediction is performed based on the real-time remaining product ratios of the supply points, the product consumption data sequences, and the associated customer flow density sequence sets, and a plurality of predicted product shortage times are output.

3. The method for controlling the path of an unmanned driving vehicle in a park combined with multimodal perception according to claim 2 is characterized in that: Build a product shortage predictor based on a generative adversarial network, including: According to the historical operation monitoring records of the target park, a sample remaining commodity ratio set, a sample commodity consumption data sequence set, and multiple sample associated passenger flow density sequence sets are collected, and the historical commodity shortage duration is set as the sample commodity shortage time to obtain the sample commodity shortage time set; The sample remaining commodity ratio set, the sample commodity consumption data sequence set and multiple sample associated passenger flow density sequence sets are used as input, and the sample commodity shortage time set is used as supervision. The generator and discriminator of the generative adversarial network are supervised trained until convergence to obtain the commodity shortage predictor.

4. The method for controlling the path of an unmanned driving system in a park combined with multimodal perception according to claim 1, characterized in that: With the comprehensive optimization goal of minimizing the time difference between delivery and pickup and delivery, the advantages and disadvantages of the multiple optional delivery solutions are evaluated, including: Based on the plurality of predicted product shortage times, performing deviation calculations on the plurality of delivery arrival time sets of the plurality of optional delivery plans, determining a plurality of time deviation sets, and summing the sums to obtain a plurality of delivery time deviations; Based on the plurality of predicted product shortage times, performing deviation analysis based on the plurality of delivery time deviation sets, and calculating a plurality of delivery time deviation amplitude means; Determining a plurality of delivery time difference coefficients based on the plurality of delivery time deviations and the plurality of delivery time deviation amplitude averages, wherein the delivery time difference coefficient is negatively correlated with the delivery time deviation and positively correlated with the delivery time deviation amplitude average; The plurality of optional delivery plans are evaluated according to the plurality of pickup and delivery distance coefficients and the plurality of delivery time difference coefficients, and an optimal delivery plan is output.

5. The method for controlling path of unmanned driving in a park combined with multimodal perception according to claim 4 is characterized in that: The method for calculating the pickup and delivery distance coefficient includes: Configure the delivery distance weight based on the product weight. The delivery distance weight is positively correlated with the product weight. The delivery distance weight is greater than or equal to 1. When the product weight is 0, the delivery distance weight is 1. Randomly selecting a first optional delivery plan, and obtaining a first pickup distance and a first delivery distance of the first optional delivery plan; The first delivery distance is compensated according to the delivery distance weight to obtain a first compensated delivery distance, and the first pickup distance is calculated to obtain a first pickup delivery distance, which is set as the first pickup delivery distance coefficient. Multiple pickup delivery distance coefficients are analyzed in sequence.

6. The method for controlling path of unmanned driving in a park combined with multimodal perception according to claim 4 is characterized in that: Evaluating the multiple optional delivery plans based on the multiple pickup and delivery distance coefficients and the multiple delivery time difference coefficients, and outputting the optimal delivery plan, including: Performing dimensionless processing on the multiple pick-up and delivery distance coefficients and the multiple delivery time difference coefficients to obtain multiple standard pick-up and delivery distance coefficients and multiple standard delivery time difference coefficients; The solution fitness is evaluated based on the multiple standard pickup and delivery distance coefficients and the multiple standard delivery time difference coefficients, multiple fitnesses are output, and the optional delivery solution with the maximum fitness is selected as the optimal delivery solution, wherein the fitness is negatively correlated with the standard pickup and delivery distance coefficient and the standard delivery time difference coefficient.

7. The unmanned driving path control system in the park combined with multimodal perception is characterized by: The steps for implementing the method for controlling path of an unmanned driving system in a park combined with multimodal perception as described in any one of claims 1 to 6 include: The park simulation space construction module is used to build a three-dimensional park simulation space based on the network distribution map and line structure map of the target park. The network includes multiple storage warehouses and several supply points. A commodity shortage time prediction module is used to monitor and obtain a plurality of commodity consumption data sequences of the plurality of supply points within a preset historical time zone, as well as a plurality of customer flow density sequences of a plurality of preset monitoring points, and analyze and obtain a plurality of predicted commodity shortage times; A delivery path enumeration module is used to enumerate possible delivery paths within the three-dimensional simulation space of the park based on the real-time position coordinates of multiple idle unmanned delivery vehicles and generate multiple initial delivery plans; The driving path control module is used to use the multiple predicted product shortage times as delivery constraints, minimize the distribution time difference and minimize the pickup and delivery distance as the comprehensive optimization goal, evaluate the advantages and disadvantages of the multiple initial delivery plans based on the multiple passenger flow density sequences, output the optimal delivery plan, and control the driving paths of the multiple idle unmanned delivery vehicles.

Images