A method and system for scheduling sanitation vehicles

Abstract

This invention provides a method and system for dispatching sanitation vehicles. By spatially gridding task points in sanitation areas and integrating multi-dimensional state parameters such as historical overflow rate, garbage composition heat, and pedestrian density, semantic task units are generated. Based on behavioral intent tags, lightweight protocol stack data packets of up to 16 bits are generated using hash encoding. Role types are determined based on dynamic operational indicators such as vehicle remaining load, battery power, task completion rate, and communication quality. For different roles such as dispatch agent, task bearer, and environmental messenger, conflict-avoidance path planning, energy-consumption-aware random tree, and local event reporting mechanisms are invoked respectively to achieve intelligent path allocation and dynamic obstacle collaborative handling. Through distributed consistency scoring and path replanning mechanisms, the semantic matching degree of task execution and overall operational stability are effectively guaranteed, improving the intelligence level of sanitation vehicle dispatching.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling and route optimization technology for sanitation vehicles, and in particular to a method and system for scheduling sanitation vehicles. Background Technology

[0002] With the acceleration of smart city construction and the popularization of digital management of environmental service vehicles, sanitation vehicle dispatching and route optimization systems have become an important technological direction for improving urban management efficiency and resource utilization. Existing sanitation vehicle route allocation and dispatching methods are mainly based on centralized route planning, game theory models, or payoff function optimization. Mainstream technical solutions typically employ distributed route planning, vehicle task allocation, and cooperative navigation. Some studies introduce Nash equilibrium, swarm intelligence, and genetic algorithms to maximize task coverage, minimize empty-running rates, and reduce energy consumption through multi-vehicle optimization iterations. Currently, industry development trends focus on multi-vehicle collaborative decision-making, information synchronization in a vehicle-to-everything (V2X) environment, and edge intelligent path adjustment. Typical application scenarios include urban garbage collection, road sweeping, and regional environmental services, which place higher demands on vehicle autonomous decision-making, dynamic task response, path conflict avoidance, and efficient task coverage. Some existing technologies have attempted to use game theory models or centralized command collaboration, but they still rely on methods such as central node scheduling, local solution of vehicle path optimality, and hard coding of task points, making it difficult to fully utilize vehicle field data and real-time environmental information for global and efficient optimization. Existing representative technologies, such as centralized control scheduling methods and multi-vehicle cooperative optimization algorithms based on game theory Nash equilibrium, are suitable for road network scheduling in conventional task allocation and stable traffic environments. These methods mainly rely on explicit game modeling between vehicles, path payoff function calculation, centralized command issuance, and iterative equilibrium solving, making them suitable for small-scale collaborative scenarios with no more than ten vehicles or areas with regularly distributed task points. However, in dynamically changing urban environments, where task point behavior intentions cannot be reported in real time, and where the vehicle's own operating status is complex and changeable, these methods suffer from insufficient execution efficiency and limited scheduling flexibility. They often result in local optimal path selection leading to global resource scheduling imbalances, task overlaps, or omissions. Summary of the Invention

[0003] In order to solve the above-mentioned technical problems, the present invention provides a method and system for dispatching sanitation vehicles.

[0004] The technical solution of this invention is implemented as follows: a method for dispatching sanitation vehicles, comprising: S1: Obtain the original task point set coordinate data and historical overflow rate, garbage composition heat map, and surrounding pedestrian density multi-dimensional state parameters within the sanitation operation area. Perform weighted fusion processing on the multi-dimensional state parameters to generate task semantic topology units with behavioral intent tags. S2: Based on the behavioral intent label of the task semantic topology unit, use the hierarchical hash encoding rule to serialize and map the task unit type code, time decay coefficient and collaborative constraint identifier to generate a lightweight semantic protocol stack data packet with a length strictly controlled within 16 bits. S3: Real-time collection of four operational indicators of the current vehicle: remaining load, battery balance, deviation rate of the last three task completions, and communication quality with neighboring vehicles. Based on preset threshold logic, the four operational indicators are judged to generate the current dynamic role type identifier of the vehicle. S4: If the dynamic role type is identified as a scheduling agent role, then the semantic priority order in the semantic protocol stack data packet is used as the guiding factor to perform a conflict-avoiding path planning operation and generate a conflict-avoiding driving path containing a connection buffer segment between semantic units. S5: If the dynamic role type is identified as a task carrier role, then the center coordinates of the task semantic topology unit are used as the sampling guide point to perform an energy consumption-aware path planning operation to generate a task carrier type driving path with the goal of minimizing energy consumption. S6: If the dynamic role type is identified as an environmental messenger role, then extract the local road condition disturbance event features and perform lightweight encoding, update the dynamic obstacle confidence parameters in the neighbor vehicle path prediction model, and do not perform the local driving path regeneration operation. S7: The generated conflict-avoidance driving path or task-carrying body driving path is sliced ​​according to the task semantic topology unit, the task intent satisfaction score covered by each path segment is calculated, and the average score of the feedback from neighboring vehicles is aggregated to generate a distributed semantic consistency verification result. S8: Determine whether the distributed semantic consistency verification result is greater than the preset threshold. If the condition is met, lock and send the corresponding driving path to the vehicle actuator. If the condition is not met, trigger the path replanning instruction and return to step S3 to re-execute the dynamic role type identification determination.

[0005] The present invention also provides a sanitation vehicle dispatching system, which uses the above-mentioned sanitation vehicle dispatching method to dispatch sanitation vehicles.

[0006] The sanitation vehicle dispatching method and system provided by this invention have the following beneficial effects: (1) By constructing a task semantic compression model and introducing an intent-driven role classification mechanism, this invention effectively overcomes the problems of static task representation and slow response of path planning in traditional sanitation scheduling, realizes a paradigm shift from "point response" to "intent understanding", significantly improves the task matching accuracy and response flexibility of multi-vehicle collaboration, and shows stronger robustness and adaptability, especially in the scenario of sudden congestion or temporary addition of tasks. (2) This invention uses a lightweight semantic protocol stack and a role-specific path generation strategy to significantly reduce communication bandwidth usage and computing resource consumption while ensuring collaborative efficiency. This enables edge devices to complete local real-time decision-making without strong cloud dependency, effectively solving the technical bottlenecks of high single-point failure risk, poor scalability, and high deployment cost in the existing centralized architecture. (3) By introducing a distributed semantic consistency verification mechanism, each vehicle can spontaneously approach the optimal behavior mode of the group without explicitly modeling the global objective function. This invention achieves the organic unity of individual action accuracy and group coordination, forming an interpretable, verifiable and scalable native collaborative control system, providing a new technical path support for complex dynamic scenarios such as smart sanitation and urban governance. Attached Figure Description

[0007] Figure 1 This is a flowchart of a sanitation vehicle dispatching method according to the present invention; Figure 2 This is a sub-flowchart of a sanitation vehicle dispatching method according to the present invention; Figure 3 This is another sub-flowchart of a sanitation vehicle dispatching method according to the present invention. Detailed Implementation

[0008] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0009] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0010] like Figure 1 As shown, the present invention provides a method for dispatching sanitation vehicles, specifically including: S1: Obtain the original task point set coordinate data and historical overflow rate, garbage composition heat map, and surrounding pedestrian density multi-dimensional state parameters within the sanitation operation area. Perform weighted fusion processing on the multi-dimensional state parameters to generate task semantic topology units with behavioral intent tags. S2: Based on the behavioral intent label of the task semantic topology unit, use the hierarchical hash encoding rule to serialize and map the task unit type code, time decay coefficient and collaborative constraint identifier to generate a lightweight semantic protocol stack data packet with a length strictly controlled within 16 bits. S3: Real-time collection of four operational indicators of the current vehicle: remaining load, battery balance, deviation rate of the last three task completions, and communication quality with neighboring vehicles. Based on preset threshold logic, the four operational indicators are judged to generate the current dynamic role type identifier of the vehicle. S4: If the dynamic role type is identified as a scheduling agent role, then the semantic priority order in the semantic protocol stack data packet is used as the guiding factor to perform a conflict-avoiding path planning operation and generate a conflict-avoiding driving path containing a connection buffer segment between semantic units. S5: If the dynamic role type is identified as a task carrier role, then the center coordinates of the task semantic topology unit are used as the sampling guide point to perform an energy consumption-aware path planning operation to generate a task carrier type driving path with the goal of minimizing energy consumption. S6: If the dynamic role type is identified as an environmental messenger role, then extract the local road condition disturbance event features and perform lightweight encoding, update the dynamic obstacle confidence parameters in the neighbor vehicle path prediction model, and do not perform the local driving path regeneration operation. S7: The generated conflict-avoidance driving path or task-carrying body driving path is sliced ​​according to the task semantic topology unit, the task intent satisfaction score covered by each path segment is calculated, and the average score of the feedback from neighboring vehicles is aggregated to generate a distributed semantic consistency verification result. S8: Determine whether the distributed semantic consistency verification result is greater than the preset threshold. If the condition is met, lock and send the corresponding driving path to the vehicle actuator. If the condition is not met, trigger the path replanning instruction and return to step S3 to re-execute the dynamic role type identification determination.

[0011] Step S1: Obtain the original task point set coordinate data and historical overflow rate, garbage composition heat map, and surrounding pedestrian density multi-dimensional state parameters within the sanitation operation area. Perform weighted fusion processing on these multi-dimensional state parameters to generate task semantic topology units with behavioral intent tags. Specifically, this includes: S1.1: Spatial gridding is performed on the original task point coordinate data within the sanitation operation area to generate a set of standard geographic raster units with unique spatial indexes, which serve as the basic spatial carrier for the fusion of multi-dimensional state parameters. Spatial data preprocessing is performed on the original task point coordinate data within the sanitation operation area. The system receives a set of two-dimensional latitude and longitude coordinates and the associated unique identifier of the task point from the geographic information acquisition terminal. The input coordinate dataset is processed to unify the coordinate system, and coordinate data from different sources are converted to the unified national geodetic coordinate system (CGCS2000) to ensure the consistency of spatial location description. The boundary range of coordinate data under a unified coordinate system is calculated. The maximum coverage boundary of the sanitation operation area is determined by the minimum bounding rectangle generation step. Based on the boundary range and the preset raster resolution parameters, a rectangular grid generation operation is performed to construct a standardized raster segmentation template. The standardized grid segmentation template is mapped to the sanitation operation area. A spatial index allocation algorithm is used to generate a unique spatial index code for each grid cell. This index code is distinguished by a combination of eastward offset coordinates and northward offset coordinates, and ensures that the index code is globally unique within the area. For each task point coordinate, the point falls into the grid determination function to map the task point to its unique spatial index grid cell, and record the identifier and coordinate position of the task point in the grid cell spatial attribute table to form a two-way mapping relationship between the task point and the grid cell. For all grid cells, the number of task points is counted and their location centers are calculated. The center point of all task point coordinates within a cell is obtained using the arithmetic mean method and is taken as the spatial representative location of that grid cell. The calculation formula is: in and These are the east-west and north coordinates of the mission point, respectively. For the number of task points, and The coordinate values ​​in all directions are summed. Through spatial gridding, the original task point set is transformed into a set of standard geographic raster units with unique spatial indexes, thus constructing the basic spatial carrier required for subsequent multi-dimensional state parameter fusion calculations. For example, within a sanitation operation area with a coverage area offset 1500 meters eastward and 2000 meters northward, coordinate data containing 320 task points is received. After being uniformly converted to the CGCS2000 coordinate system, 30×40 rectangular grid cell templates are generated at a resolution of 50 meters eastward and 50 meters northward. A spatial indexing allocation algorithm is used to label each cell as IDX(i,j), where i is the eastward index and j is the northward index. Using a grid cell placement function, the 320 task points are distributed across different grid cells. The number of task points in each cell is counted, and the coordinates of the center point are calculated. For example, a cell may contain 4 task points with east-west coordinates of 1020, 1040, 1060, and 1080 meters, and northward coordinates of 500, 520, 520, and 540 meters, with the cumulative sums being... and The center point coordinates are calculated to be (1050, 520), which can be used as the spatial representative position of this grid cell. This process establishes a stable mapping between environmental task points and grid cells, ensuring the accuracy and consistency of spatial localization during subsequent feature vector extraction and multi-dimensional state fusion. S1.2: Based on the standard set of geographic grid cells, extract historical overflow rate time series data, waste composition heat map distribution data and surrounding pedestrian density real-time monitoring data to generate a three-dimensional feature vector group corresponding to each grid cell; S1.3: The historical overflow rate, garbage component heat value and pedestrian density value in the three-dimensional feature vector group are weighted and fused using an adaptive weight allocation algorithm to generate a comprehensive situation evaluation value that characterizes the urgency of the operation of each grid unit. Based on the three-dimensional feature vector group generated in step S1.2, a structured data matrix containing historical overflow rate values, garbage component heat values, and pedestrian density values ​​is loaded as the input condition for the fusion operation. An initialization module of an adaptive weight allocation algorithm is used to apply this algorithm to each component in the three-dimensional feature vector group. Based on a preset sensitivity function, the contribution weight of each component to the urgency of the task is calculated to form an initial weight vector. The weight vector is input into the dynamic adjustment module, and the initial weight is iteratively corrected using real-time task status feedback parameters to construct an adaptive weight coefficient set that adapts to the current grid cell state. A weighted fusion calculator was used to perform a multidimensional linear combination calculation on historical overflow rate values, waste composition heat values, and pedestrian density values. The calculation formula is as follows: in This is a comprehensive situation assessment value. This represents the historical overflow rate. The calorific value of waste components. The value of human flow density. , , These are adaptive weighting coefficients; The fusion results are input into the normalization processing module, and the comprehensive situation assessment value is mapped to a unified dimension range using the minimum-maximum scaling function to ensure the consistency of subsequent classification mapping processing. Through the above calculations and normalization, the three-dimensional feature vector group from the previous step is transformed into a comprehensive situational assessment value that characterizes the urgency of operations for each grid cell, thereby achieving a unified quantitative expression of multi-dimensional state parameters. For example, for a standard geographic raster unit in a sanitation operation area, its historical overflow rate is 0.78, waste composition heat value is 0.65, and pedestrian density value is 0.52. The adaptive weight allocation algorithm initializes the weight vector to [0.5, 0.3, 0.2], and updates it to [0.45, 0.35, 0.20] after adjustment based on task status feedback. Substituting the above parameters into the formula: Calculations yielded After normalization, the comprehensive situation assessment value is mapped to 0.70 within the [0,1] interval. This value is determined as a high-urgency category in the subsequent behavior intent classification mapping, directly affecting the priority scheduling of operation strategies. Under the input conditions of different grid cells, the adaptive weight coefficients are adjusted according to real-time feedback. The comprehensive situation assessment value output by fusion calculation can support accurate classification and topology construction after normalization, achieving a significant improvement in the accuracy of multi-grid operation urgency determination and scheduling response capability. S1.4: Perform behavioral intent classification mapping processing based on the interval distribution characteristics of the comprehensive situation assessment value, convert the continuous numerical comprehensive situation assessment value into a discrete behavioral intent label type, so as to generate intent-marked raster data with clear operational strategy direction; S1.5: Perform a topology connection construction operation based on the spatial adjacency relationship of the intent-labeled raster data, and aggregate adjacent raster units with the same or compatible behavioral intent label types into connected regions to generate the final task semantic topology unit.

[0012] Step S2: Based on the behavioral intent label of the task semantic topology unit, a hierarchical hash encoding rule is used to serialize and map the task unit type code, time-decrease coefficient, and collaborative constraint identifier to generate a lightweight semantic protocol stack data packet with a length strictly controlled within 16 bits. Specifically, this includes: S2.1: Obtain the behavioral intent label data of the task semantic topology unit, and use the discrete state mapping algorithm to perform binary encoding processing on semantic categories such as high overflow risk area, low frequency stagnation area and continuous sweeping zone to generate a task unit type code with a fixed length of 2 bits; S2.2: Based on the historical overflow rate change trend of the task semantic topology unit and the difference between the current timestamp, a linear decay quantization algorithm is used to dynamically calculate the urgency of the task in order to generate a 4-bit timeliness decay coefficient that reflects the timeliness characteristics. S2.3: Receive the topology connection attribute data of the task semantic topology unit, and apply the logical bitmap construction algorithm to perform feature extraction processing on the cooperation rules such as the requirement for indivisible jobs and the constraint of continuous jobs, so as to generate a 3-bit cooperation constraint identifier containing specific cooperation restrictions; S2.4: Integrate the task unit type code, the time-attenuation coefficient and the collaborative constraint identifier, and execute the hierarchical hash concatenation algorithm to serially combine the above three coded data according to the preset bit order to generate an initialized unverified semantic protocol stack original frame; The input conditions include the task unit type code (fixed binary length 2 bits), the time decay coefficient (fixed binary length 4 bits), and the cooperative constraint identifier (fixed binary length 3 bits) obtained through the previous sub-steps, as well as the preset bit order rules. The execution object is the above three encoded data items and the bit order rules; For this execution object, the bit order mapping module is first called to locate the three encoded data according to the preset bit position index, and to clarify the start and end bits of each encoded field in the protocol frame; Next, the initial processing unit using the hierarchical hash concatenation algorithm places the task unit type code field in the first layer bit segment of the protocol frame according to the bit order rule, places the time-attenuation coefficient field in the second layer bit segment, and places the collaborative constraint identifier field in the third layer bit segment, forming a logical combination model of three layers bit segments. Subsequently, hash calculation is introduced during the splicing process. The binary values ​​of each field are XORed and weighted shifted bitwise to generate an intermediate hash value for obfuscation and anti-collision. This hash value is then filled into the high-bit reserved segment of the protocol frame. Then, the spliced ​​three-layer bit segments and the high-bit reserved segments are arranged in sequence according to the serial combination rules. During the combination process, bit width alignment is performed, and zero values ​​are filled in for fields with insufficient bit width to ensure the overall continuity of the protocol frame. Finally, all bit segments are merged to generate an initialized, unverified semantic protocol stack raw frame, which contains task category encoding, timeliness encoding, cooperative constraint encoding, and obfuscation hash field; Through the above-described layered hash concatenation process, the discrete encoding result of the previous step is transformed into structured, continuous, and transmissible protocol stack raw frame data, enabling the role determination and path planning module to directly parse the signaling input.

[0013] For example, in a sanitation operation scenario, the task unit type code is... The time-related decay coefficient is The collaborative constraint identifier is The positional rules stipulate that the type code occupies positions 1-2, the timeliness coefficient occupies positions 3-6, the collaborative constraint occupies positions 7-9, and the high-order reserved segment occupies positions 10-16. After the positional mapping module locates the start and end positions of each field, the hierarchical hash concatenation algorithm first places the type code. Place the time factor in the first or second position. Cooperative constraints at positions 3-6 At positions 7-9. Perform hash operation: ,in This represents a bitwise XOR operation, and the result is then weighted and shifted to obtain the high-order bit fill value. Fill in positions 10-16. This will ultimately form the original frame. In local link testing, the hash value with high-bit padding significantly improves the anti-collision capability of packet parsing. When the path planning module parses the protocol frame, it can directly extract each field and achieve efficient intent synchronization in real-time scheduling. S2.5: Implement a total length truncation verification algorithm on the initialized unverified semantic protocol stack raw frame, forcibly discard redundant information exceeding the 16-bit upper limit and fill in the high-bit zero values ​​to generate a lightweight semantic protocol stack data packet that finally meets the transmission requirements of the vehicle communication channel. Perform bit width statistics on the initialized, unverified semantic protocol stack raw frame, and obtain the actual total bit value of the raw frame as the input benchmark for length detection; The actual total bit value is compared with the preset maximum allowed bit length threshold. If the comparison result shows that the original frame length exceeds the threshold, the total length truncation verification algorithm module is called to perform high-bit redundancy removal operation, and the high-bit encoding of the excess part is deleted to retain the low-bit key fields. After performing the high-bit redundancy removal operation, the difference between the remaining coding length and the maximum allowed bit length is calculated to obtain the padding bit width parameter; Based on the padding bit width parameter, the zero-value padding function is called to perform zero-value bit padding in the high-bit segment of the original frame so that its total length precisely matches the maximum allowed bit length. Perform structural verification on the semantic protocol stack after length adjustment to verify the start position and bit order integrity of the encoded fields of each layer, and ensure that the truncation and completion operations do not destroy the correct interpretation of the hierarchical hash structure. The validated and adjusted semantic protocol stack is marked as the final lightweight semantic protocol stack data packet and output to the transmit buffer of the vehicle communication module. By strictly controlling the length and maintaining the bit order, the hierarchical hash concatenation result of the previous step is transformed into protocol data that meets the requirements of V2X low bandwidth channel for immediate transmission and use, so as to achieve controllable latency and guaranteed message integrity for cross-vehicle semantic synchronization. For example, for a length of The initialization of bits in the unverified semantic protocol stack raw frame has a system-preset maximum allowed bit length threshold of [value missing]. The number of bits exceeded after performing a length comparison operation is 1. The truncation algorithm is invoked to remove the high-order bits of the original frame. Bit redundancy encoding retains only the lower 16 bits of the core field. The padding bit width parameter is calculated after length adjustment. The result is No additional padding is required. The structural verification shows that the reserved low-order fields start at bit 1 and end at bit 16. The hierarchical hash encoding uses bits 1-2 for the first-layer task unit type code, bits 3-6 for the second-layer time-attenuation coefficient, and bits 7-9 for the third-layer cooperative constraint identifier. The high-order padding bits (bits 10-16) are empty, ensuring the communication module's protocol decoder correctly parses the data according to the preset bit order. After the data packet is sent to the vehicle communication module's transmit buffer, in actual vehicle V2X communication tests, the data packet latency remains within... Within milliseconds, and with a significant improvement in the decoding consistency score of neighboring vehicles, it is evident that the length truncation and bit padding processing in this step effectively improves the stability and real-time performance of cross-vehicle collaborative semantic synchronization.

[0014] like Figure 2 As shown, step S3 involves: real-time collection of four operational indicators for the current vehicle: remaining load, battery charge, deviation rate of the last three task completions, and communication quality with neighboring vehicles. These four operational indicators are then judged based on preset threshold logic to generate a dynamic role type identifier for the vehicle. Specifically, this includes: S3.1: Acquire the raw data of remaining load collected by the vehicle-mounted sensors, the raw data of battery balance output by the battery management system, the raw data of the deviation rate of the three most recent tasks stored in the task log, and the raw data of the communication quality of neighboring vehicles received by the vehicle network communication module. Perform timestamp alignment and outlier removal on the above four types of raw data to generate a standardized four-dimensional operation index vector. S3.2: Based on the four-dimensional operation index vector, the sliding window filtering algorithm is used to perform noise suppression and trend smoothing on each index, and the remaining load smoothing value, battery balance smoothing value, task completion deviation rate smoothing value and neighbor vehicle communication quality smoothing value reflecting the instantaneous operation capability of the vehicle are extracted to construct a highly reliable dynamic state feature set. S3.3: Based on the preset role determination rule base, compare the remaining load smooth value in the dynamic state feature set with the first load threshold, compare the battery balance smooth value with the first power threshold, compare the task completion deviation rate smooth value with the deviation tolerance threshold, and compare the neighbor vehicle communication quality smooth value with the communication delay threshold to generate a preliminary role determination intermediate variable containing four Boolean logic results. The high-confidence dynamic state feature set output in step S3.2 is used as the input object, and the numerical threshold comparison module in the role determination rule base is called in turn to perform a quantitative comparison operation between the remaining load smooth value and the first load threshold. A Boolean mapping function based on the difference sign is used to convert the comparison result into a Boolean value of the vehicle load state, forming the first item of the preliminary Boolean logic sequence; The battery remaining smooth value is quantitatively compared with the first power threshold. The battery state determination result is encapsulated into a Boolean value using the same symbol mapping mechanism, forming the second item of the preliminary logic sequence. The smoothed value of the task completion deviation rate is compared with the deviation tolerance threshold. A threshold inversion mechanism is used to map results below the tolerance threshold as true and results above the tolerance threshold as false, forming the third item of the logical sequence. A quantitative comparison is performed between the smoothed value of the communication quality of neighboring vehicles and the communication delay threshold. Results with a delay below the threshold are mapped to true, and results with a delay above the threshold are mapped to false, generating the fourth item of the logical sequence. The four Boolean results are combined in a predetermined order to form intermediate variables for preliminary role determination, realizing the structured transformation from state features to logical determination. Through Boolean logic processing, the feature values ​​of the previous step are transformed into determination variables that can be input into arbitration rules, thus realizing the establishment of the preconditions for dynamic role type determination. For example, in a sanitation vehicle dispatching scenario, the remaining load smoothing value is set to... The first load threshold is The battery balance smoothing value is set to The first power threshold is The task completion deviation rate smoothing value is set to The deviation tolerance threshold is The smoothing value for neighbor vehicle communication quality is set to (Unit: seconds), the communication delay threshold is set to... (Unit: seconds). Perform a comparison operation on the remaining load: The result is true; perform a comparison operation on the remaining battery capacity: The result is true; perform a comparison operation on the task completion deviation rate: The result is true; perform a comparison operation on the communication delay: The result is true. The combined intermediate variables for the initial role determination are [true, true, true, true]. In this scenario, the rule base arbitration maps the roles that simultaneously meet the conditions of high load and low latency to scheduling agent roles. To meet the high power condition, it can be used as a redundant attribute to enhance role stability. Execution effect verification shows that when all four inputs are true, the accuracy of role allocation is significantly improved, and the global collaboration efficiency is greatly enhanced. S3.4: Based on the intermediate variables of the preliminary role determination, execute the multi-condition priority arbitration logic. If the high load and low latency conditions are met, it is mapped to the scheduling agent role code. If the low load and high power conditions are met, it is mapped to the task carrier role code. If the path intersection risk and communication fluctuation conditions are met, it is mapped to the environmental messenger role code. Output a unique dynamic role type identifier. Based on the preliminary role determination intermediate variable containing four Boolean logic results, the multi-condition priority arbitration logic module is called to load the synchronized set of load condition Boolean value, power condition Boolean value, task deviation condition Boolean value and communication delay condition Boolean value obtained by the preceding threshold comparison at the data input end. In the rule matching unit, a condition combination matrix is ​​constructed. The load condition Boolean value and the communication delay condition Boolean value are combined to form a high load and low delay condition group. The power condition Boolean value and the load condition Boolean value are combined to form a low load and high power condition group. The path cross risk identifier and the communication quality fluctuation Boolean value are combined to form a path cross communication fluctuation condition group. Within the priority arbitrator, arbitration priorities are assigned to condition groups according to preset weights. The high-load, low-delay condition group is assigned the highest priority, the low-load, high-power condition group is assigned the second-highest priority, and the path-crossing communication fluctuation condition group is assigned the lowest priority. Boolean AND operations are performed within the same priority group to determine whether the condition group is fully satisfied. If the high load and low latency condition group is met, the role mapping table retrieval operation is performed, and the matching index is mapped to the scheduling agent role code; if the high load and low latency condition group is not met but the low load and high power condition group is met, it is mapped to the task carrier role code; if neither of the first two condition groups is met but the path cross-communication fluctuation condition group is met, it is mapped to the environmental messenger role code. The above mapping results are processed by a uniqueness extraction module to remove conflicts of multiple satisfactions, retaining only the single role identifier with higher priority, thus forming unique dynamic role type identifier data; By using a multi-condition priority arbitration process, the Boolean logic result of the previous step is transformed into a single and clear dynamic role type identifier, enabling real-time role determination and task matching of vehicles based on their operating status in multi-vehicle collaboration. For example, in a sanitation vehicle with a rated load of 1000kg and a battery capacity of 50kWh, the first load threshold is set to 850kg, the first battery capacity threshold to 30kWh, the communication delay threshold to 0.2 seconds, and the communication fluctuation threshold to 30%. The vehicle's real-time smoothed remaining load value is 900kg, the smoothed remaining battery capacity value is 28kWh, the smoothed communication delay value is 0.15 seconds, the smoothed communication fluctuation value is 25%, and the path intersection risk flag is true. The load condition Boolean value is true, the battery capacity condition Boolean value is false, the communication delay condition Boolean value is true, and the communication quality fluctuation Boolean value is false. According to the multi-condition priority arbitration logic, the high load low delay condition group Boolean AND result is true, directly mapped to the scheduling agent role code. When using the path intersection communication fluctuation condition group as input, because the communication quality fluctuation Boolean value is false, the condition group is not satisfied and has a low priority, so it is not selected. The mapping result is output as a unique scheduling agent role type identifier. When applied to subsequent conflict-avoidance path planning algorithms, this vehicle can significantly improve the efficiency of task connection within the same area and reduce the probability of conflict in actual operation. S3.5: Perform validity verification and status maintenance on the generated dynamic role type identifier. If the verification passes, write the dynamic role type identifier into the role configuration bit of the vehicle's local control register and trigger the role response interrupt signal of the downstream path planning module to complete the final establishment and issuance of the vehicle's current dynamic role type identifier.

[0015] like Figure 3 As shown, step S4: If the dynamic role type identifier is a scheduling agent role, then using the semantic priority order in the semantic protocol stack data packet as the guiding factor, a conflict-avoiding path planning operation is performed to generate a conflict-avoiding travel path containing a connection buffer segment between semantic units. Specifically, this includes: S4.1: Based on the lightweight semantic protocol stack data packet generated in the previous steps, extract the task unit type code and time decay coefficient, and perform dynamic weight assignment on the task semantic topology unit in the current sanitation operation area to generate a semantic priority sorting sequence with real-time urgency indicators, which serves as the spatial guidance basis for path planning. In the preceding steps, a lightweight semantic protocol stack data packet has been obtained as an input condition. The data packet contains a task unit type code and a time decay coefficient generated by hierarchical hash encoding. Both are fixed-length binary fields and conform to the 16-bit length constraint. The task unit type code is parsed in binary form, and each type code is mapped to the category index of the corresponding task semantic topology unit to establish a direct index relationship from the encoded field to the task category. Numerical decoding is performed on the timeliness attenuation coefficient to map the 4-bit encoded value to the actual attenuation weight factor. The mapping is completed according to the preset timeliness level table to ensure that the attenuation coefficient and the urgency of the task conform to a linear or piecewise linear correspondence. The category index and the decay weight factor are associated with the spatial coordinates of the task semantic topology unit to construct a basic task weight mapping table, in which each unit has two fields: initial category weight and dynamic time-effect weight. A dynamic weighting algorithm is used to generate a comprehensive task priority weight by multiplying and fusing the initial category weights and dynamic time-sensitive weights. The algorithm formula is as follows: in As the weight of task category, The product result is a weighting factor for time decay. This is a comprehensive task priority weight; Normalization is performed on the priority weights of the comprehensive tasks, mapping the weight range to the closed interval [0,1] to eliminate the difference in weight dimensions between different task categories; Based on the normalization results, a semantic priority sorting sequence is generated, the weights are arranged from high to low, and the spatial index of the corresponding task semantic topology unit is recorded to form a spatial guidance sequence that can be used for subsequent path planning. Through the above multi-level parsing, mapping, fusion and normalization processing, the result of the previous step is transformed into a semantic priority sorting sequence with real-time urgency indicators, realizing the spatial guidance function of the scheduling agent in path planning. For example, in a lightweight semantic protocol stack data packet, the task unit type code is binary "10", corresponding to a high overflow risk area. The category weight is set to 0.85, and the timeliness attenuation coefficient is binary "0110", corresponding to an attenuation weight factor of 0.75. A basic task weight mapping table is established through mapping, where the initial category weight of this task unit is 0.85, and the dynamic timeliness weight is 0.75. According to the formula... The overall task priority weight is calculated to be 0.6375. After normalizing the overall weight of all task units, the normalized value of this unit is 0.92, ranking second in priority during sorting, and its index coordinates in the spatial guidance sequence are (245, 178). In this scenario, the generated semantic priority sorting sequence enables the path planning engine to significantly improve the coverage of high-value, high-urgency task units during global search, reduce the path resources occupied by low-value task units, and effectively improve the global collaborative efficiency of scheduling agent paths. S4.2: Based on the spatial distribution characteristics of adjacent high-priority task semantic topology units in the semantic priority sorting sequence, calculate the theoretical minimum safe distance between unit boundaries, and perform buffer segment expansion operation in combination with vehicle kinematic constraint parameters to generate a geometric model of the inter-semantic unit connection buffer segment used to isolate potential path conflicts. S4.3: Map the geometric model of the connection buffer segment between the semantic units to a dynamic obstacle layer, and integrate the dynamic obstacle confidence parameters updated by the environmental messenger role of the neighboring vehicle broadcast to construct a multi-layer conflict avoidance grid map containing static road network topology and dynamic game-theoretic no-entry zone as the physical constraint environment for path search. A coordinate system transformation is performed on the spatial data of the geometric model of the connecting buffer section, and the boundary path of the buffer section is mapped to a global road coordinate system consistent with the static road network topology to ensure accurate matching of the matrix index of the subsequent dynamic obstacle layer. A dynamic obstacle layer generation algorithm is executed on the unified buffer segment geometric model. The method of batch filling according to grid resolution is used to map the continuous coordinate sequence within the boundary of each buffer segment into a two-dimensional grid matrix containing obstacle marker bits, forming the first layer of dynamic obstacle data that can be superimposed on the static road network data. The system receives dynamic obstacle confidence parameter data packets generated by the environmental messenger role broadcast by a neighboring vehicle, performs protocol parsing on the data packets, extracts the spatial location index, confidence value, and category label of the corresponding dynamic obstacle, and maps them to the corresponding cells of the same two-dimensional raster matrix. A confidence fusion model is used to calculate the fusion confidence of the dynamic obstacle layer generated in the buffer section and the dynamic obstacle confidence of the environmental messenger broadcast. The superposition is achieved using a confidence-weighted average, and the fusion formula is as follows: in For the confidence level of the buffer section obstacle layer, For environmental messenger broadcast obstacle confidence, and These are the corresponding weights; The fused dynamic obstacle layer and static road network topology data are combined in multiple layers to construct a multi-layer conflict avoidance grid map containing a static road network layer and a fused dynamic obstacle layer. Through the above processing method, the buffer section geometric model of the previous step is transformed into a multi-layer conflict avoidance grid map that can be directly used for path search physical constraints, so as to realize the construction of a path planning environment that simultaneously considers the fixed road network structure and the dynamic game information of neighboring vehicles. For example, in a certain sanitation operation area, the geometric model of the buffer section consists of rectangular path segments with a boundary length of 15 meters and a width of 3 meters. The dynamic obstacle layer raster resolution generated after mapping is 0.5 meters, and each buffer section occupies 90 raster cells. The environmental messenger data packet broadcast by neighboring vehicles contains three dynamic obstacles with location indices of (12,8), (15,8), and (18,9), confidence levels of 0.6, 0.8, and 0.7, respectively, and the category label of all three being temporary construction barriers. Assuming the base confidence level C of the buffer section obstacle layer is 0.9, the weight w is 0.7, and the neighboring vehicle confidence weight wen is 0.3, then the fusion calculation formula is: The fusion result is This represents the overall confidence level of dynamic obstacles at that location. After the dynamic obstacle layer and static road topology are synthesized through fusion processing, the multi-layer raster map shows high-confidence dynamic obstacle markers in areas containing buffer segment data. The path search algorithm has a significantly improved avoidance probability in these areas, ensuring the safety of the planned path and the scheduling coordination effect. S4.4: Based on the multi-layer conflict-avoidance grid map, using the center coordinates of the high-priority task semantic topology unit in the semantic priority sorting sequence as the heuristic target point, the improved conflict-avoidance A-star variant algorithm is invoked to perform global path search, so as to generate a conflict-free rough driving trajectory that initially avoids the dynamic restricted area and covers the key task nodes. Based on the static road network topology layer and dynamic game-theoretic forbidden zone layer of the multi-layer conflict-avoidance grid map, the path search module is called to initialize the heuristic target set. The center coordinates of the semantic topology unit of the high-priority task in the semantic priority sorting sequence are used as the target point input to generate a target point index set. By traversing the target point index set, the coordinate transformation algorithm is called to convert the center coordinates into a raster map index number, and the priority coefficient of the node where the target point is located is marked in the multi-layer raster map; An improved conflict-avoidance variant of the A* algorithm is adopted, taking the current grid node of the vehicle as the starting node, using the dynamic obstacle layer parameters of the multi-layer grid map as the node prohibition condition, and setting the heuristic evaluation function as the product of the squared Euclidean distance and the priority coefficient. By calculating the following cost function at each node expansion, forbidden zones are filtered out and high-priority nodes are expanded preferentially: in, The total cost of the node. This represents the cumulative path cost from the starting node to the current node. For heuristic valuation, Weighting coefficients for node priority; The entry prohibition logic of the dynamic obstacle layer is integrated into the extension process. During the adjacent node generation stage, the movement direction legality check module is called to filter out nodes that conflict with the dynamic obstacle layer and generate a candidate node set. Perform a priority queue enqueue operation on the candidate node set, using the cost function value as the sorting basis, and repeat the expansion until all target points are covered, generating a conflict-free coarse driving trajectory that includes key task nodes; Through the above processing method, the multi-layer conflict avoidance grid map constructed in the previous step is transformed into preliminary trajectory data driven by high-priority heuristic target points, so as to achieve the expected effect of vehicle paths covering key task nodes while avoiding dynamic restricted areas. For example, in a sanitation operation area, the grid map size is 200×200 units. The static road network layer consists of road nodes, and the dynamic obstacle layer contains two no-entry zones broadcast by the neighboring vehicle environmental messenger. The center coordinates of the no-entry zones are (50, 60) and (120, 140), respectively, with a radius of 5 grid cells. The vehicle's current location index number is (20, 30), and the center coordinates of the high-priority task nodes in the semantic priority sorting are (80, 90) and (150, 160), with corresponding priority coefficients of 1.5 and 2.0, respectively. The target point indices 8090 and 150160 are generated through a coordinate transformation algorithm, and the priority coefficients are marked on the grid map. The heuristic evaluation function of the improved A* variant algorithm is set as the product of the squared Euclidean distance and the priority coefficient. When the vehicle expands its nodes, the no-entry zone judgment module in the dynamic obstacle layer removes all nodes whose index distance from the center of the no-entry zone is less than or equal to 5. The cost function of the candidate node set is calculated as follows: ,in in and The coordinates of the current node. and The coordinates of the target point are used. Node expansion and priority queue sorting are performed, and the final conflict-free coarse driving trajectory effectively avoids restricted areas and effectively covers two high-priority task nodes. Verification results show that the total trajectory length is reduced, the coverage of key task nodes is significantly improved, and the probability of dynamic obstacle collisions is reduced to a low level acceptable to the business. S4.5: Perform smoothing optimization on the conflict-free coarse driving trajectory, forcibly embed the boundary constraints of the geometric model of the connecting buffer segment between semantic units, and verify the satisfaction of the trajectory with the semantic priority sorting sequence to generate a final conflict-avoiding driving path containing standardized connecting buffer segments for use by the subsequent consistency verification module.

[0016] Step S5: If the dynamic role type is identified as a task carrier role, then using the center coordinates of the task semantic topology unit as the sampling guide point, an energy consumption-aware path planning operation is performed to generate a task carrier type driving path with the goal of minimizing energy consumption. Specifically, this includes: S5.1: Obtain the set of task semantic topological units and corresponding center coordinate data generated in the previous steps, and use the spatial index mapping algorithm to perform rasterization discretization processing on the center coordinates to generate a sequence of sampling guide points containing unique spatial identifiers, which serves as the initial target constraint set for path search. S5.2: Based on the sampling guide point sequence and the current real-time vehicle pose data, a multi-dimensional cost function model is constructed to perform weighted fusion calculations on the vehicle kinematic constraint parameters, battery discharge efficiency curve and road friction coefficient to generate a dynamic energy consumption gradient field characterizing the energy loss rate per unit distance. Based on the sampling guide point sequence and the current real-time vehicle pose data, the pose parsing module is called to extract the vehicle's position vector and attitude quaternion information in the two-dimensional or three-dimensional global coordinate system, and the spatial identifiers in the sampling guide point sequence are mapped to the corresponding physical position coordinates. Vehicle kinematic constraint parameters, including maximum steering angle, minimum turning radius, acceleration limit, and load mass, are used as the first set of input variables. The multi-segment function parameters of the battery discharge efficiency curve (voltage-current-power relationship under charging and discharging conditions) are used as the second set of input variables. The dynamic monitoring value of the road friction coefficient is used as the third set of input variables. All of these are uniformly converted into a numerical matrix with standard dimensions. The multidimensional cost function construction module is called, and a weighted fusion mechanism is adopted with physical loss and dynamic constraints as the weight basis to map the three types of input variables to the energy consumption value under unit displacement. Then, interpolation correction is used to avoid data discontinuity between sampling points. When constructing the dynamic energy consumption gradient field, the path segment between the current position and the adjacent sampling guide point is used as the iteration unit, and the gradient calculation operator is called to numerically solve the rate of change of energy consumption per unit distance. in, Energy loss rate per unit distance , , These are the weighting coefficients corresponding to kinematic constraints, battery characteristics, and tribological characteristics, respectively. This is a power consumption function based on load and turning radius. The power consumption function is based on the battery efficiency curve. Let the frictional resistance consumption function be . For the kinematic parameter vector, This is the battery state vector. The coefficient of friction; The gradient field generation module is called to spatially interpolate the energy consumption calculation results on the global grid map to generate a continuous energy consumption distribution, and the gradient value is obtained through the numerical differentiation operator to characterize the rate of change of energy consumption with the path direction. in, The energy consumption gradient vector, , These are the two directional components of the local coordinate system along the path. The gradient field is segmented and stored according to the sampling guide point region, and then called in the subsequent bias sampling strategy to tilt the sampling density towards the low-energy gradient region; Through the above processing method, the sampling guide point sequence and pose data of the previous step are transformed into a dynamic energy consumption gradient field of energy loss rate per unit distance, so as to realize the real-time perception and constraint of energy consumption factors in the path planning stage. For example, in a sanitation vehicle with a rated load capacity of 1500 kg, the kinematic constraints are set as follows: maximum steering angle of 35 degrees, minimum turning radius of 6 meters, and acceleration limit of 1.5 m / s². The battery is a lithium-ion battery pack, and the discharge efficiency curve is fitted as a piecewise linear function: efficiency is 0.92 when power output is below 20 kW, and drops to 0.85 when power output is above 20 kW. The road friction coefficient is calculated using onboard acceleration and wheel speed sensors, with a real-time value of 0.65. The sampling guide point sequence contains 15 spatial identifiers, which are mapped to their corresponding coordinate system positions before a multidimensional cost function model is invoked, with weighting coefficients set to... , , The calculated energy consumption per unit distance for a certain path segment is... The gradient operator calculates the rate of change of energy consumption along the x-direction in kilowatt-hours per kilometer. The rate of change of kilowatt-hours per kilometer per meter along the y-direction is Kilowatt-hours per kilometer per meter. The dynamic gradient field generated after spatial interpolation shows that low-energy consumption areas are concentrated in the south and southwest path segments. Improving the node generation probability in these areas during bias sampling significantly improves the energy consumption optimization effect of path planning. S5.3: Based on the dynamic energy consumption gradient field, the bias sampling strategy is executed to improve the random node generation mechanism of the traditional fast expanding random tree algorithm, and the sampling probability density is concentrated in the low energy consumption gradient region to generate an initial expanding node cloud with optimal energy consumption. Based on the dynamic energy consumption gradient field, the unit distance energy loss rate matrix calculated in the previous step S5.2 is used as the input parameter to perform biased design on the random node generation module in the fast expanding random tree algorithm. A normalized sampling weight matrix is ​​constructed using the energy consumption gradient value of each spatial grid cell in the gradient field, and the energy consumption gradient value is converted into sampling probability weights through inverse scaling. The cumulative distribution function is used to generate the corresponding candidate sampling probability distribution vector to ensure that low-energy gradient regions have a higher probability of being selected during node generation. The probability density of the node position coordinates randomly generated by the original RRT algorithm is reset. The generated coordinates are mapped to the sampling weight matrix, and the node's landing position in the spatial grid is selected according to the probability distribution vector. An energy consumption threshold filtering mechanism is introduced to remove nodes whose landing point energy consumption gradient value is higher than a preset threshold, and to regenerate replacement nodes that meet the low energy consumption conditions to form a preliminary set of nodes with low energy consumption tendency. The spatial connectivity of a node set is determined by neighborhood detection, and the node groups that meet the connectivity conditions are aggregated into an initial extended node cloud. The mathematical calculation of the sampling probability weights can be performed using the following formula: in, For node sampling probability weights, This represents the energy consumption gradient value for the corresponding grid cell. To prevent extremely small positive constants with a denominator of zero; By using a bias sampling strategy and energy consumption threshold filtering, the dynamic energy consumption gradient field of the previous step is transformed into an initial extended node cloud with high probability density concentrated in the low energy consumption region, thereby achieving the expected technical effect of reducing the tendency of vehicle path energy consumption. For example, in a sanitation operation scenario in a certain urban area, the value range of the dynamic energy consumption gradient field matrix is ​​0.05 to 0.20 kWh / km, and the sampling weight matrix is ​​calculated according to the formula... Calculate, set The value is 0.001, and the low energy consumption threshold is set to 0.08 kWh / km. For a grid cell with an energy consumption gradient of 0.06 kWh / km, the sampling weight is calculated as follows: =15.97, this value occupies a significant proportion in the normalized weight distribution, thus significantly increasing the probability of this region being selected during the random node generation process. 300 initial nodes were generated through bias sampling. 78 nodes with energy consumption exceeding 0.08 kWh / km were filtered out using an energy consumption threshold and replaced with corresponding low-energy nodes, ultimately outputting 222 initial extended node clouds that meet connectivity requirements. In the path planning execution verification, the connected path tree derived from this cloud showed a significant reduction in average energy consumption per unit distance compared to the RRT path without bias sampling, resulting in a reduction in the vehicle's total cumulative energy consumption throughout the journey, achieving the technical goal of low-energy path planning. S5.4: Use the bidirectional connection optimization algorithm to reconstruct the topology of the parent and child nodes in the initial extended node cloud, and combine the dynamic energy consumption gradient field to calculate the cumulative energy consumption value of each connection edge, so as to select the minimum energy consumption connected path tree from the initial pose to the sampling guide point sequence. S5.5: Based on the minimum energy consumption connected path tree, apply the cubic spline interpolation algorithm to smooth the curvature continuity at the path turning point, and verify the coverage integrity of the corrected path to the task semantic topological unit boundary, so as to generate the final task carrier body driving path. Based on the node and edge data of the minimum energy consumption connected path tree, the geometric features at the path turning points are extracted to obtain the initial curvature value of each turning point and the tangential direction vector of the adjacent path segments. A cubic spline interpolation algorithm is used to establish an interpolation model for the transition region of adjacent path segments. By constructing a sequence of control points and solving the coefficients of the cubic polynomial, the interpolation curve satisfies the constraints of continuity of the first derivative and continuity of the second derivative at each inflection point. The curvature difference of the corrected path in the local transition region is calculated using the curvature continuity check formula, as follows: in, and These represent the curvature values ​​before and after correction at the same inflection point, ensuring... Below the preset curvature difference threshold; Spatial coverage verification is performed on the continuous coordinate sequence of the corrected path and the boundary data of the task semantic topological unit, using the following coverage calculation formula: in, The number of task semantic topological units to cover for path segments. The total number of path segments ensures coverage parameters. Meets the integrity threshold; The corrected path that meets the conditions of curvature continuity and coverage integrity is marked as the final mission carrier body travel path and output to the consistency verification module. By using cubic spline interpolation and coverage integrity verification, the minimum energy consumption connected path tree of the previous step is transformed into a final driving path with a smooth structure and complete semantic unit coverage, thereby achieving the technical effect of minimizing operating energy consumption and significantly improving operation coverage. For example, in a sanitation operation scenario, the minimum energy consumption connected path tree for vehicles contains 18 path nodes and 17 connecting edges, with the initial curvature at path turning points ranging from 0.15 to 0.42. The interpolation control point sequence adopts an equidistant distribution strategy, with a single interpolation interval length of 5 meters. The interpolation polynomial coefficients are solved using the least squares method. The preset threshold for curvature continuity verification is set to 0.05, and the corrected turning points are calculated using the above formula. All values ​​were within 0.03. The coverage integrity threshold was set at 0.95 during the testing process. =27, =28, the coverage rate is calculated. The value is 0.964, which meets the coverage requirements. In actual operation, the final output path significantly reduced the mean energy consumption gradient, improved the coverage of task semantic units, and ensured that the low-speed parking action of the vehicle in the critical operation area strictly matched the requirements of the operation strategy.

[0017] Step S6: If the dynamic role type is identified as an environmental messenger role, then extract the local road condition disturbance event features and perform lightweight encoding, update the dynamic obstacle confidence parameters in the neighboring vehicle path prediction model, without performing the local driving path regeneration operation. Specifically, this includes: S6.1: Acquire raw local road condition video stream data and lidar point cloud data collected by vehicle-mounted perception sensors, and use a multimodal spatiotemporal synchronization algorithm to perform timestamp alignment and spatial coordinate unification processing on the two types of data to generate a multi-source fusion environmental perception data frame with a unified spatiotemporal reference. S6.2: Based on the multi-source fusion environmental perception data frame, a dynamic target detection neural network is applied to extract features from non-static obstacles such as water trucks occupying the road and temporary construction barriers, so as to generate a local road condition disturbance event feature set containing obstacle category labels, three-dimensional bounding box coordinates and motion vectors. S6.3: Perform a semantic abstract mapping operation on the feature set of the local road condition disturbance events, and use predefined discretization coding rules to convert continuous three-dimensional bounding box coordinates into relative grid indices, motion vectors into direction level codes, and obstacle category labels into type identifiers to generate a structured disturbance event semantic description vector. Semantic abstract mapping is performed on obstacle category labels, 3D bounding box coordinates, and motion vectors in the feature set of local road disturbance events to generate structured disturbance event semantic description vectors. The input conditions required are the feature set data extracted by S6.2. The coordinates of the 3D bounding box are discretized according to the preset grid size parameters. The original continuous coordinate values ​​are divided by the grid cell size and rounded to obtain the relative grid index values, which are then mapped into an integer encoding matrix according to the spatial indexing rules. The direction quantization algorithm is performed on the motion vector data. By calculating the polar angle of the motion vector in the global orientation system and dividing it into a predefined set of direction levels, each direction level is mapped to a fixed-length binary code. Perform type identifier mapping on obstacle category labels according to the category coding table, match the original category text labels to the corresponding category index and convert them into discrete integer codes; The relative grid index value, orientation level code, and category integer code are combined into a structured vector in field order, and the field labels are retained for subsequent compression encoding processing; Through the above processing method, the result of the previous step is transformed into a unified encoded semantic description vector of the disturbance event, realizing the standardized expression of multi-source perception information and supporting fast broadcasting and efficient parsing of neighboring vehicles under low bandwidth channels. For example, when a sanitation vehicle detects a temporary construction site barrier, the center coordinates of the bounding box are (125.7, 88.3, ​​0.0) meters, the grid cell size is set to 1.0 meter, and a relative grid index of (125, 88, 0) is generated using discretization. The formula used here is: in For the grid index in the x-direction, This represents the floor operation. The motion vector is ( The speed is 0.5, 0.866 meters per second. The polar angle θ = 120° is calculated. In the predefined direction level set {0°, 45°, 90°, 135°, 180°, 225°, 270°, 315°}, the corresponding level is 135°, and the mapped direction level code is "100101". The obstacle category label "construction fence" corresponds to the integer code 5 in the category coding table. The index (125, 88, 0), the direction level code "100101", and the category code 5 are combined into a vector [125, 88, 0, 100101, 5] in the field order. After subsequent hash compression, the length of this vector is significantly shortened. After being broadcast to neighboring vehicles, it can significantly improve the dynamic obstacle resolution speed and maintain the accuracy of obstacle updates. S6.4: Based on the semantic description vector of the disturbance event, a lightweight hash compression algorithm is used to perform bit width pruning and serial splicing on each field in the vector to generate a lightweight encoded data packet of local road condition disturbance event with strictly limited length and containing key disturbance information. S6.5: Receive the lightweight encoded data packet of the local road condition disturbance event, parse it and input it into the uncertainty calculation module of the adjacent vehicle path prediction model, use the Bayesian update mechanism to correct the probability of dynamic obstacles in the corresponding spatial area in the model, so as to generate the updated dynamic obstacle confidence parameters and complete the local environment state synchronization.

[0018] Step S7: The generated conflict-avoidance driving path or task-carrying body driving path is sliced ​​according to task semantic topology units, the task intent satisfaction score covered by each path segment is calculated, and the average score from neighboring vehicles is aggregated to generate a distributed semantic consistency verification result. Specifically, this includes: S7.1: Obtain the continuous coordinate sequence of the conflict avoidance driving path or the driving path of the task carrier body and the spatial boundary definition of the task semantic topology unit. Use the spatial projection matching algorithm to map the continuous coordinate sequence to the discrete task semantic topology unit set and generate the path segment sequence with unit belonging identifier. S7.2: Based on the path segment sequence with unit affiliation identifier and the behavior intention label of the task semantic topology unit, call the action feature extractor to parse the vehicle motion state data in the path segment sequence and generate an action feature vector set containing the number of low-speed stops, the overlap of the coverage trajectory, and the proportion of operation time. S7.3: Based on the action feature vector set and the pre-set intent satisfaction rule base in the behavioral intent label, perform multi-dimensional weighted scoring operation to quantitatively evaluate the action feature vector set and generate a task intent satisfaction score that represents the degree of achievement of a specific behavioral intent for a single path segment. Based on the action feature vector set and the behavioral intent label of the task semantic topology unit extracted in the previous steps, the weight coefficients of each feature dimension are initialized and allocated by calling the preset intent satisfaction rule base. The weight coefficients are set according to the historical statistical results of the sensitivity of different behavioral intents to features, so as to ensure the relevance of the scoring model. The difference between the number of low-speed stops, the overlap of the coverage trajectory, and the proportion of operation time in the action feature vector set and the target interval value corresponding to the behavior intention label is calculated. The difference calculation result is used to characterize the degree of deviation between the actual action features and the target action features. The aforementioned deviation levels are multiplied item by item according to the weight coefficients in the rule base to form a multidimensional deviation weighted vector. This multidimensional deviation weighted vector is then normalized using the following scoring formula: in, Score the degree to which the task intent is satisfied. Let i be the weight coefficient of the i-th feature. The normalized feature matching value for the i-th feature. The total number of features; This formula is used to obtain the task intent satisfaction score for each path segment, thereby achieving a quantitative match between action features and behavioral intent. The calculated score is output to the subsequent distributed consensus aggregation module, so that it can be used as a local score input source to participate in the calculation of the regional collaborative mean. Through multidimensional weighted scoring calculation, the action feature vector set of the previous step is transformed into a quantitative task intent satisfaction score, so as to achieve accurate evaluation of the degree of achievement of specific behavioral intent by path segments; For example, in a certain sanitation operation area, the path segment sequence contains the following action characteristics: 8 low-speed stops, 0.92 overlap in coverage trajectory, and 0.35% of operation time. The corresponding behavioral intent label is "high spillover risk area," and the weight coefficients in the preset rule base are 0.5, 0.3, and 0.2, respectively. The difference between the number of low-speed stops and the target interval value of 10 is calculated to obtain... 2. The difference between the coverage trajectory overlap and the target value of 0.95 is calculated to obtain... 0.03 is obtained by calculating the difference between the percentage of work time and the target value of 0.4. 0.05. Normalize the above differences to obtain corresponding matching values ​​of 0.8, 0.97, and 0.875, and input them into the multidimensional weighted scoring formula: The calculated score is 0.882. This score is used as a local task intent satisfaction index and input into the distributed consensus aggregation algorithm. After aggregating the scores of neighboring vehicles, the scoring result is significantly improved, ensuring that this path segment can maintain a high operational quality in global collaboration. S7.4: Receive the local task intent satisfaction score from the lightweight semantic protocol stack data packet broadcast by the neighboring vehicle, and use the distributed consensus aggregation algorithm to perform weighted fusion processing on the local task intent satisfaction score and the local task intent satisfaction score of the neighboring vehicle to generate the regional collaborative task intent satisfaction average. The system receives the local task intent satisfaction score from the lightweight semantic protocol stack data packet broadcast from the neighboring vehicle as an external input signal, calls the protocol stack parsing module to parse the task unit type code, time-attenuation coefficient and cooperative constraint identifier field according to the preset bit order decoding rules, and extracts the local task intent satisfaction score value related to the task segment. The task intent satisfaction score calculated locally is paired with the parsed task intent satisfaction score of neighboring vehicles according to spatial adjacency to generate a one-to-one list of task segment score pairs, ensuring that the scores of the same segment in different vehicles are at the same index position during the fusion calculation process. Construct a weight allocation matrix for the distributed consensus aggregation algorithm. Combine vehicle role type, communication delay smoothing value and task urgency coefficient to apply different weight coefficients to the local score and neighbor vehicle score, so that the weight allocation matrix satisfies the normalization constraints of each row and column. Under the constraints of the weight allocation matrix, a weighted fusion calculation is performed, and the score of each pair of segments is processed according to the following formula: in, Let be the weight coefficient for the j-th segment corresponding to the i-th vehicle. The score represents the task intent satisfaction level for the i-th vehicle corresponding to the j-th segment. The average score of the merged segments; The average score of all merged segments is used to calculate the global mean, which is then used as a quantitative indicator of the multi-vehicle collaboration effect in the work area. By using distributed consensus aggregation and weight adjustment, the local and neighboring vehicle task intent satisfaction scores generated in the previous step are transformed into the global regional collaborative task intent satisfaction average, thereby achieving a comprehensive evaluation of the consistency and collaborative efficiency of multi-vehicle operations. For example, in a sanitation operation area, the task intent satisfaction scores for path segments calculated by local vehicles are 4.2, 3.8, and 5.0, respectively. The corresponding segment scores obtained from parsing the lightweight semantic protocol stack data packets broadcast by neighboring vehicles are 3.9, 4.1, and 4.8. Combining the vehicle role type as a scheduling agent, a communication latency smoothing value of 180ms, and a task urgency coefficient of 1.3, a weight allocation matrix is ​​constructed, where the local score has a weight of 0.6 and the neighboring vehicle score has a weight of 0.4. Weighted fusion is then performed on the first segment: The calculated result is 4.08. Processing all segments sequentially yields a fusion score list of 4.08, 3.92, and 4.92, with a global mean of 4.64. In practical applications, this mean is significantly higher than the set collaboration satisfaction value of 4.0, indicating that multi-vehicle collaboration in this area has achieved high consistency and operational quality, and collaboration efficiency has been significantly improved. S7.5: Based on the average satisfaction rate of regional collaborative task intent and the preset consistency judgment threshold logic, perform Boolean comparison operation to verify the compliance of the average satisfaction rate of regional collaborative task intent, and generate a distributed semantic consistency verification result for deciding whether to execute path locking or trigger replanning.

[0019] Step S8: Determine whether the distributed semantic consistency verification result is greater than a preset threshold. If the condition is met, lock and issue the corresponding driving path to the vehicle actuator. If the condition is not met, trigger a path replanning instruction and return to step S3 to re-execute the dynamic role type identification determination. Specifically, this includes: S8.1: Obtain the continuous coordinate sequence of the conflict avoidance driving path or the task carrier type driving path and the boundary data of the task semantic topology unit. Use the spatiotemporal mapping algorithm to cut the continuous coordinate sequence into a set of discrete path segments that correspond one-to-one with the task semantic topology unit, and generate path semantic slice data with unique unit identifiers. S8.2: Based on the set of discrete path segments and behavioral intention labels in the path semantic slice data, call the intention matching degree evaluation model to compare and analyze the vehicle motion trajectory features in each path segment with the preset behavioral intention constraints, and calculate the task intention satisfaction score that represents the local operation quality of a single vehicle. S8.3: Receive the task intent satisfaction score value broadcast from neighboring vehicles and the task intent satisfaction score value calculated locally, and use the distributed weighted average fusion algorithm to aggregate the multi-source task intent satisfaction score values ​​to generate a distributed semantic consistency verification result scalar. S8.4: Based on the distributed semantic consistency verification result scalar and the preset threshold parameter, perform logical comparison operation to determine the cooperative effectiveness of the current path scheme. If the distributed semantic consistency verification result scalar is greater than the preset threshold parameter, generate a path locking instruction; if it is less than or equal to the preset threshold parameter, generate a path replanning trigger signal. S8.5: Based on the path locking instruction, the corresponding conflict avoidance driving path or task carrier type driving path is sent to the vehicle execution mechanism, or the dynamic role type identification determination process is reset based on the path replanning trigger signal and the previous steps are returned to re-execute the role evaluation operation to complete the closed-loop scheduling control.

[0020] The present invention also provides a sanitation vehicle dispatching system, which uses the above-mentioned sanitation vehicle dispatching method to dispatch sanitation vehicles.

[0021] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0022] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and rules of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for dispatching sanitation vehicles, characterized in that, Includes the following steps: S1: Obtain the original task point set coordinate data within the sanitation operation area, extract multi-dimensional state parameters based on the original task point set coordinate data and perform weighted fusion to generate task semantic topology units; S2: Based on the behavioral intent tag of the task semantic topology unit, perform serialization mapping on the task unit type code, time-effect decay coefficient and collaborative constraint identifier to generate semantic protocol stack data packets; S3: Real-time collection of the current vehicle's operating indicators, and determination of the operating indicators based on preset threshold logic to generate the vehicle's current dynamic role type identifier. S4: If the dynamic role type is identified as a scheduling agent role, then the semantic priority order in the semantic protocol stack data packet is used as the guiding factor to perform a conflict-avoidance path planning operation and generate a conflict-avoidance driving path. S5: If the dynamic role type is identified as a task carrier role, then the center coordinates of the task semantic topology unit are used as the sampling guide point to perform energy consumption-aware path planning operation and generate a task carrier type driving path. S6: If the dynamic role type is identified as an environmental messenger role, then extract and encode the local road condition disturbance event features, update the dynamic obstacle confidence parameters in the neighbor vehicle path prediction model, and do not perform the local driving path regeneration operation.

2. The sanitation vehicle dispatching method according to claim 1, characterized in that, Following S6, the following is also included: S7: Slice the conflict avoidance driving path or the task carrier driving path according to the task semantic topology unit, calculate the task intent satisfaction score covered by each path segment, and aggregate the average score of neighboring vehicle feedback to generate distributed semantic consistency verification results. S8: Determine whether the distributed semantic consistency verification result is greater than the preset threshold. If the condition is met, lock and send the corresponding driving path to the vehicle actuator. If the condition is not met, trigger the path replanning instruction and return to S3 to re-execute the dynamic role type identifier determination.

3. The sanitation vehicle dispatching method according to claim 1, characterized in that, The multidimensional state parameters include historical overflow rate, waste composition heat map, and surrounding pedestrian density.

4. The sanitation vehicle dispatching method according to claim 1, characterized in that, S3 specifically includes: The system collects raw data on the vehicle's remaining load, battery balance, deviation rate of the last three tasks, and communication quality with neighboring vehicles in real time. It then performs timestamp alignment and outlier removal on these raw data to generate a standardized four-dimensional operational index vector. Based on the four-dimensional operation index vector, noise suppression and trend smoothing are performed on each index, and the smoothed values ​​of remaining load, battery balance, task completion deviation rate and neighbor vehicle communication quality are extracted to construct a dynamic state feature set. Based on the preset role determination rule library, the remaining load smooth value in the dynamic state feature set is compared with the first load threshold, the battery balance smooth value is compared with the first power threshold, the task completion deviation rate smooth value is compared with the deviation tolerance threshold, and the neighbor vehicle communication quality smooth value is compared with the communication delay threshold to generate preliminary role determination intermediate variables. Based on the intermediate variables for the initial role determination, a multi-condition priority arbitration logic is executed. If the conditions of high load and low latency are met, it is mapped to the scheduling agent role code. If the conditions of low load and high power are met, it is mapped to the task carrier role code. If the conditions of path intersection risk and communication fluctuation are met, it is mapped to the environmental messenger role code, and a unique dynamic role type identifier is output.

5. A sanitation vehicle dispatching method according to claim 4, characterized in that, S3 further includes: The dynamic role type identifier is validated and its status is maintained. If the validation passes, the validated dynamic role type identifier is written into the role configuration bit of the vehicle's local control register, and the role response interrupt signal of the downstream path planning module is triggered, thus completing the final establishment and issuance of the vehicle's current dynamic role type identifier.

6. The sanitation vehicle dispatching method according to claim 1, characterized in that, S4 specifically includes: Based on the semantic protocol stack data packets generated by S2, the task unit type code and timing decay coefficient are extracted, and dynamic weight assignment is performed on the task semantic topology units in the current sanitation operation area to generate a semantic priority sorting sequence. Based on the spatial distribution characteristics of adjacent high-priority task semantic topology units in the semantic priority sorting sequence, the theoretical minimum safe distance between unit boundaries is calculated, and a buffer segment expansion operation is performed in combination with vehicle kinematic constraint parameters to generate a geometric model of the connecting buffer segment between semantic units. The geometric model of the inter-semantic unit connection buffer segment is mapped to a dynamic obstacle layer, and the dynamic obstacle confidence parameters updated by the environmental messenger role broadcast by neighboring vehicles are fused to construct a multi-layer conflict avoidance grid map. Based on the multi-layer conflict-avoidance grid map, and using the center coordinates of the high-priority task semantic topology unit in the semantic priority sorting sequence as the heuristic target point, a global path search is performed to generate a conflict-free coarse driving trajectory. The conflict-free coarse driving trajectory is smoothed and optimized, the boundary constraints of the geometric model of the connection buffer segment between the semantic units are forcibly embedded, and the satisfaction of the trajectory with the semantic priority sorting sequence is verified to generate a conflict-avoiding driving path.

7. A sanitation vehicle dispatching method according to claim 6, characterized in that, The multi-layered conflict-avoidance grid map includes a static road network topology and a dynamic game-theoretic restricted area.

8. A sanitation vehicle dispatching method according to claim 1, characterized in that, S5 specifically includes: Obtain the set of task semantic topology units generated by S1 and the corresponding center coordinate data, perform rasterization discretization on the center coordinates, and generate a sampling guide point sequence; Based on the sampling guide point sequence and the current real-time vehicle pose data, a multi-dimensional cost function model is constructed to perform weighted fusion calculations on the vehicle kinematic constraint parameters, battery discharge efficiency curve and road friction coefficient to generate a dynamic energy consumption gradient field. Based on the dynamic energy consumption gradient field, an offset sampling strategy is executed to improve the random node generation mechanism of the traditional fast expanding random tree algorithm, concentrating the sampling probability density in the low energy consumption gradient region to generate an initial expanding node cloud. The parent and child nodes in the initial extended node cloud are reconstructed topologically. The cumulative energy consumption value of each connecting edge is calculated by combining the dynamic energy consumption gradient field. The minimum energy consumption connected path tree from the initial pose to the sampling guide point sequence is selected. Based on the minimum energy consumption connected path tree, the curvature continuity at the path turning points is smoothly corrected, and the coverage integrity of the corrected path to the task semantic topological unit boundary is verified to generate the task carrier body driving path.

9. A sanitation vehicle dispatching method according to claim 8, characterized in that, The sampling guide point sequence contains a unique spatial identifier.

10. A sanitation vehicle dispatching system, characterized in that: The sanitation vehicle dispatching method according to any one of claims 1-9 is used for sanitation vehicle dispatching.

Images