Double-vehicle cooperative non-contact medical material intelligent distribution method and system

Abstract

The invention provides a double-vehicle cooperative non-contact medical material intelligent distribution method and system, and relates to the technical field of logistics robots. The method comprises the following steps: constructing a global topology network based on a distribution environment and obtaining a state vector of a carrying unit; comprehensive execution cost is calculated by using a multi-dimensional cost function, and an optimal task allocation sequence and an initial planning path are generated; in the driving process, gray level distribution data predicted by Kalman filtering and a node vision correction vector are fused, and real-time motion control parameters are generated to correct the course; based on a dynamic road right arbitration mechanism of a space-time dimension, predicting a future running track of the carrying unit, and calculating a dynamic priority score when space-time overlapping is detected; and generating a conflict resolution strategy according to the score, controlling the low-priority unit to execute deceleration avoidance, and locking the right of passage of the high-priority unit. The problems of deadlock and collision of multi-vehicle dispatching in a medical scene are solved, and the material distribution efficiency and the positioning robustness are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of logistics robot technology, and in particular to a dual-vehicle collaborative contactless intelligent delivery method for medical supplies. Background Technology

[0002] As the global population ages and the prevalence of chronic diseases continues to rise, hospitals and elderly care institutions are facing increasing pressure on medical resources. Traditional manual delivery of medical supplies is not only inefficient, but also prone to problems such as dispensing the wrong medicine or sending it to the wrong ward. Especially when dealing with public health emergencies, traditional delivery methods pose a risk of cross-infection and cannot meet the urgent need for contactless delivery.

[0003] Existing technologies include solutions that utilize automation to improve drug delivery efficiency, such as microcontroller-based intelligent delivery systems. These systems typically use a microcontroller as the core control unit, integrating a machine vision module to complete tasks like ward number recognition and path tracking. Such systems generally employ wireless communication modules to enable communication between the two delivery vehicles, utilize motor drive chips to control vehicle movement, and employ grayscale sensor modules for path recognition, thereby achieving independent operation of a single vehicle or basic collaborative functions between two vehicles.

[0004] However, existing technologies have shortcomings in multi-vehicle collaboration. The main issue is that the collaboration logic relies on preset time-sequence triggers and cannot dynamically allocate tasks based on real-time status, resulting in low overall delivery efficiency. Furthermore, existing technologies lack effective conflict resolution and right-of-way allocation mechanisms in scenarios with multiple vehicles coexisting, such as intersections. Additionally, the single visual or grayscale tracking navigation method has weak anti-interference capabilities and is prone to positioning deviations or getting lost under varying lighting conditions or complex paths, failing to meet the efficient and stable delivery needs of complex medical scenarios. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, the purpose of this invention is to provide a dual-vehicle collaborative contactless intelligent delivery method and system for medical supplies. This invention solves the problems in the prior art, such as the rigid collaborative scheduling mechanism leading to low system operating efficiency, the lack of a dynamic right-of-way arbitration mechanism easily causing path conflicts, and the single navigation mode and poor environmental adaptability.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A method for intelligent, contactless delivery of medical supplies using a dual-vehicle collaborative system includes:

[0008] A global topology network is constructed based on a preset delivery environment, and the state vectors of multiple transport units are acquired in real time. The state vectors include: the dynamic positioning coordinates of the transport unit, battery power data, and task load status. The global topology network includes: multiple static topology coordinates and path weight data connecting the static topology coordinates.

[0009] The comprehensive execution cost of the task to be executed relative to each of the carrier units is calculated based on the global topology network and the state vector.

[0010] The optimal task allocation sequence is obtained based on the comprehensive execution cost, and the corresponding initial planning path is planned based on the optimal task allocation sequence.

[0011] During the execution along the initial planned path, grayscale distribution data of the ground path is collected in real time, and the grayscale distribution data is predicted using the Kalman filter algorithm to obtain the basic lateral deviation correction amount.

[0012] The Euclidean distance between the dynamic positioning coordinates and the static topological coordinates is calculated in real time. When the Euclidean distance is less than a preset trigger threshold, the visual correction process for the current node is activated to obtain the visual correction vector.

[0013] Real-time motion control parameters are obtained based on the basic lateral deviation correction amount and the visual correction vector to drive the vehicle unit to correct the heading angle and driving speed.

[0014] The dynamic right-of-way arbitration mechanism based on spatiotemporal dimensions predicts the future running trajectory of the carrier unit according to the real-time motion control parameters. When the future running trajectories of different carrier units are detected to overlap spatiotemporally in the same static topological coordinate region, the dynamic priority score of each carrier unit is calculated.

[0015] A conflict resolution strategy is generated based on the dynamic priority score, and after the spatiotemporal overlap is removed, the vehicle unit is controlled to resume tracking the initial planned path. The conflict resolution strategy includes: controlling the vehicle unit with the lower dynamic priority score to perform deceleration and avoidance actions, and locking the right-of-way of the vehicle unit with the higher dynamic priority score.

[0016] The path tracking, correction, and arbitration steps are continuously executed until the dynamic positioning coordinates are detected to coincide with the static topological coordinates of the target node. Based on visual positioning, the unloading of materials is confirmed, and the state vector is updated to trigger the next round of path reconstruction loop.

[0017] A dual-vehicle collaborative contactless intelligent delivery system for medical supplies includes:

[0018] The network construction and status acquisition module is used to construct a global topology network based on a preset delivery environment and acquire the status vectors of multiple transport units in real time. The status vectors include: the dynamic positioning coordinates of the transport unit, battery power data and task load status. The global topology network includes: multiple static topology coordinates and path weight data connecting the static topology coordinates.

[0019] The comprehensive cost calculation module is used to calculate the comprehensive execution cost of the task to be executed relative to each of the carrier units based on the global topology network and the state vector;

[0020] The task allocation and path planning module is used to obtain the optimal task allocation sequence based on the comprehensive execution cost, and to plan the corresponding initial planning path based on the optimal task allocation sequence.

[0021] The deviation prediction and correction module is used to collect grayscale distribution data of the ground path in real time during the execution along the initial planned path, and use the Kalman filter algorithm to predict the grayscale distribution data to obtain the basic lateral deviation correction amount.

[0022] The node visual correction module is used to calculate the Euclidean distance between the dynamic positioning coordinates and the static topological coordinates in real time. When the Euclidean distance is less than a preset trigger threshold, the visual correction process for the current node is activated to obtain the visual correction vector.

[0023] The multi-source fusion control module is used to obtain real-time motion control parameters based on the basic lateral deviation correction amount and the visual correction vector to drive the carrier unit to correct the heading angle and driving speed.

[0024] The dynamic right-of-way arbitration module is used for a dynamic right-of-way arbitration mechanism based on the spatiotemporal dimension. It predicts the future running trajectory of the carrier unit according to the real-time motion control parameters. When it is detected that the future running trajectories of different carrier units overlap spatiotemporally in the same static topological coordinate region, it calculates the dynamic priority score of each carrier unit.

[0025] The conflict resolution execution module is used to generate a conflict resolution strategy based on the dynamic priority score, and after the spatiotemporal overlap is removed, control the vehicle unit to resume tracking the initial planned path. The conflict resolution strategy includes: controlling the vehicle unit with the lower dynamic priority score to perform deceleration and avoidance actions, and locking the right-of-way of the vehicle unit with the higher dynamic priority score.

[0026] The closed-loop monitoring and interaction module is used to continuously perform path tracking, correction and arbitration steps until the dynamic positioning coordinates coincide with the static topological coordinates of the target node. Based on visual positioning, the unloading of materials is confirmed, and the state vector is updated to trigger the next round of path reconstruction loop.

[0027] The present invention discloses the following technical effects:

[0028] This invention provides a dual-vehicle collaborative contactless intelligent delivery method and system for medical supplies. Addressing the problems of rigid collaborative scheduling mechanisms, lack of dynamic right-of-way arbitration, and poor adaptability to navigation environments in existing technologies, this invention offers the following advantages: First, by constructing a task allocation model based on a multi-dimensional cost function, dynamic optimal matching between transport units and delivery tasks is achieved, breaking through the efficiency bottleneck of traditional serial scheduling and significantly improving the overall system throughput. Second, a dynamic right-of-way arbitration mechanism based on spatiotemporal dimensions is introduced, utilizing conflict resolution strategies to intelligently lock and allocate right-of-way at intersections, effectively eliminating deadlock and collision risks in multi-vehicle operation. Finally, a closed-loop navigation architecture fusing visual correction and Kalman filter prediction overcomes the shortcomings of single sensors being susceptible to interference from lighting and path contamination, significantly enhancing the positioning accuracy and operational robustness of contactless delivery in complex medical environments. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart of a dual-vehicle collaborative contactless intelligent delivery method for medical supplies, provided as an embodiment of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] like Figure 1As shown, this invention provides a dual-vehicle collaborative contactless intelligent delivery method for medical supplies, comprising:

[0034] Step 100: Construct a global topology network based on a preset delivery environment and acquire the state vectors of multiple transport units in real time. The state vectors include: the dynamic positioning coordinates of the transport unit, battery power data, and task load status. The global topology network includes: multiple static topology coordinates and path weight data connecting the static topology coordinates.

[0035] Step 200: Calculate the comprehensive execution cost of the task to be executed relative to each of the carrier units based on the global topology network and the state vector;

[0036] Step 300: Obtain the optimal task allocation sequence based on the comprehensive execution cost, and plan the corresponding initial planning path based on the optimal task allocation sequence;

[0037] Step 400: During the execution along the initial planned path, grayscale distribution data of the ground path is collected in real time, and the grayscale distribution data is predicted using the Kalman filter algorithm to obtain the basic lateral deviation correction amount.

[0038] Step 500: Calculate the Euclidean distance between the dynamic positioning coordinates and the static topological coordinates in real time. When the Euclidean distance is less than a preset trigger threshold, activate the visual correction process for the current node to obtain the visual correction vector.

[0039] Step 600: Obtain real-time motion control parameters based on the basic lateral deviation correction amount and the visual correction vector to drive the vehicle unit to correct the heading angle and travel speed;

[0040] Step 700: A dynamic right-of-way arbitration mechanism based on spatiotemporal dimensions is used to predict the future running trajectory of the carrier unit according to the real-time motion control parameters. When it is detected that the future running trajectories of different carrier units overlap spatiotemporally in the same static topological coordinate region, the dynamic priority score of each carrier unit is calculated.

[0041] Step 800: Generate a conflict resolution strategy based on the dynamic priority score, and after the spatiotemporal overlap is resolved, control the vehicle unit to resume tracking the initial planned path. The conflict resolution strategy includes: controlling the vehicle unit with the lower dynamic priority score to perform deceleration and avoidance actions, and locking the right-of-way of the vehicle unit with the higher dynamic priority score.

[0042] Step 900: Continuously execute path tracking, correction and arbitration steps until the dynamic positioning coordinates coincide with the static topological coordinates of the target node. Based on visual positioning, complete the material unloading confirmation and update the state vector to trigger the next round of path reconstruction loop.

[0043] Furthermore, the specific implementation process of step 100 is as follows:

[0044] This embodiment first discretizes the physical space of the hospital corridor or delivery area, establishing a two-dimensional Cartesian coordinate system with the pharmacy dispensing window as the origin. Using a laser rangefinder or high-precision measuring tape, the geometric center coordinates of intersections, ward entrances, and charging station locations are measured in the field. These key locations are marked as graph theory nodes, and the absolute coordinate values ​​of each node are recorded. Further, this embodiment constructs an adjacency matrix based on the connectivity of physical paths. When calculating the path weight connecting two nodes, not only the Euclidean distance between them is considered, but a turning cost factor is also introduced. For path segments requiring a 90-degree turn, this embodiment adds a preset time penalty value to the physical length, normalizes the weighted values, and stores them in the corresponding elements of the adjacency matrix, thereby completing the construction of a global topology network containing spatial location information and traffic cost information.

[0045] This embodiment establishes point-to-point data transmission links with each carrier unit through a wireless communication module, and performs periodic communication using a specific data frame format. This data frame format sequentially includes a frame header, device ID, data payload length, payload, and checksum. Upon receiving the raw byte stream uploaded by the underlying hardware, this embodiment first checks the frame header identifier to synchronize data packets. Then, it uses a cyclic redundancy check (CRC) algorithm to verify the integrity of the data frame. If the CRC calculation result is inconsistent with the received checksum, the frame is determined to be noise interference and discarded. Only valid data frames that pass the CRC check are retained for subsequent parsing processes, thereby ensuring the reliability of data interaction in medical equipment environments with high electromagnetic interference.

[0046] This embodiment performs bit-domain parsing on the unpacked payload data and extracts key status information according to a predefined communication protocol. For dynamic positioning coordinates, this embodiment reads the encoder's cumulative pulse count and converts it into a relative mileage value. Simultaneously, it reads the latest road sign ID identified by the vision module, combining the two to update the real-time position index of the vehicle unit in the global topology network. For battery power data, this embodiment directly reads the ADC voltage sampling value uploaded by the power management chip and maps it to the remaining battery percentage. For task load status, this embodiment reads the flag bits in the status register to determine whether the vehicle unit is currently in an idle waiting, fully loaded delivery, or returning to charge state. Finally, all the parsed data is encapsulated into a state vector for subsequent task allocation algorithms to use.

[0047] Furthermore, the specific implementation process of step 200 is as follows:

[0048] This embodiment first parses the received delivery task instruction to be executed, extracting the target ward number or pharmacy window ID. By consulting a pre-built static address mapping table, these logical addresses are converted into the corresponding target static topological coordinates in the global topology network. This embodiment then constructs a graph-based search space in the control unit's memory, using the dynamic positioning coordinates of the current transport unit as the starting node and the target static topological coordinates as the ending node, and calls the A* (A-Star) heuristic search algorithm or Dijkstra's algorithm for path planning. During the calculation process, this embodiment not only calculates the geometric connection length between nodes but also accumulates the actual Euclidean distance of each physical path segment, thereby outputting a shortest path length value accurate to the centimeter level. This value is defined as the quantized value of the path's Euclidean distance, serving as the spatial dimension benchmark for subsequent cost calculations.

[0049] To prevent the vehicle unit from shutting down due to battery depletion during mission execution, this embodiment constructs a piecewise inverse proportional energy penalty function model. This embodiment reads the battery percentage data uploaded by the battery management system in real time and sets a safety threshold (e.g., 20%). When the read battery percentage is below this safety threshold, this embodiment triggers an exponential penalty mechanism, using an exponential function to generate a high penalty value approaching infinity, thereby forcibly prohibiting the low-battery vehicle unit from accepting long-distance missions at the algorithm level. Conversely, when the battery percentage is above the safety threshold, this embodiment uses an inverse proportional linear function, where the generated penalty value decreases linearly as the remaining battery power increases. That is, the more abundant the battery power, the lower its corresponding energy state score, thus favoring vehicles with higher battery power to undertake more missions.

[0050] This embodiment further introduces an assessment of the impact of dynamic road conditions on delivery efficiency. It extracts the inherent weight data of all topological segments covered by the aforementioned shortest path and retrieves historical traffic frequency statistics recorded in the system database for that time period (e.g., morning peak check-in period). This embodiment multiplies or weights the inherent weights with the historical frequencies to calculate the cumulative resistance value of the planned path, thereby quantifying the road segment congestion coefficient. Finally, this embodiment presets three sets of weight adjustment factors, corresponding to the importance of spatial distance, energy consumption, and road congestion, respectively. The previously calculated path Euclidean distance quantification value, energy status score, and road segment congestion coefficient are substituted into a linear weighted summation formula. After normalization, a unique dimensionless numerical result is output, which is the comprehensive execution cost of the transportation unit for the current task, used for subsequent task bidding and allocation decisions.

[0051] Furthermore, the specific implementation process of step 300 is as follows:

[0052] This embodiment first establishes a two-dimensional mapping relationship between the set of tasks to be assigned and the set of available transport units. Using the comprehensive execution cost data calculated in the previous steps, a cost mapping matrix is ​​constructed. In this matrix, each row vector corresponds to a specific transport unit, and each column vector corresponds to a delivery task to be executed. The specific comprehensive execution cost values ​​are filled into the matrix elements at the intersection of rows and columns. This embodiment then sets minimizing the overall system operating cost as the global constraint objective and uses a combinatorial optimization algorithm to iteratively solve the cost mapping matrix. Under the premise of satisfying the load capacity limitations of the transport units, an allocation scheme that minimizes the sum of the selected elements in the matrix is ​​calculated. This determines which transport unit each task belongs to and its execution order in a multi-task scenario, ultimately generating an optimal task allocation sequence that includes clear task ownership and time priority.

[0053] This embodiment parses the generated optimal task allocation sequence to extract the target location information of the current high-priority task. By consulting the topology map database, it directly locks the target's static topological coordinates. Given that the transport unit may be located at any position in the road network rather than on a standard graph node when receiving a task, this embodiment simultaneously reads the current dynamic positioning coordinates of the transport unit to which the task belongs. It traverses all node data in the global topology network, calculates the Euclidean distance between the dynamic positioning coordinates and each graph node, and retrieves the nearest graph node by comparison. This node is then anchored as the starting static topological coordinates for this navigation task, thus completing the spatial mapping from a continuous physical coordinate system to a discrete topology network, ensuring that the path planning algorithm has valid input endpoints.

[0054] This embodiment performs path optimization between the determined initial and target static topological coordinates, invoking the A-Star heuristic graph search algorithm or Dijkstra's algorithm to perform a breadth-first or depth-first search in the global topological network. During the search process, this embodiment reads pre-stored path weight data as the traversal cost of the topological graph edges. This cost covers physical distance, congestion coefficient, and turning cost. The algorithm accumulates the total weights of different path branches to select a path with the minimum cumulative cost from the starting point to the destination. This embodiment finally discretizes this path and outputs it as a list of nodes composed of a series of consecutive static topological coordinates arranged in sequence. This list of nodes is defined as the initial planned path and sent to the underlying motion control module for execution.

[0055] Furthermore, the specific implementation process of step 400 is as follows:

[0056] This embodiment utilizes a linear grayscale sensor array mounted on the bottom of the carrier unit to read the analog voltage signal of the ground path in a high-frequency sampling mode, and quantizes it into a one-dimensional grayscale distribution data array through an analog-to-digital converter. This embodiment employs the Otsu method or a preset dynamic threshold algorithm to binarize this grayscale distribution data array, dividing the pixels into a set of black pixels representing the background and a set of white pixels representing the path. This embodiment traverses the binarized array, retrieving and locking the start and end indices of the white pixel regions, marking them as the left and right boundary indices of the path, respectively. This embodiment calculates the numerical center of the left and right boundary indices using the arithmetic mean method, performs a difference operation with the inherent physical geometric center index of the sensor array, thereby calculating the original geometric offset of the carrier unit relative to the center of the guidance path at the current moment, and defines it as the path observation deviation value.

[0057] To suppress measurement noise caused by ground stains, sudden changes in lighting, or sensor jitter, this embodiment constructs a discrete linear dynamic system model based on the Kalman filter algorithm. This embodiment defines a two-dimensional system state vector containing the lateral deviation and its first derivative (i.e., the rate of change of deviation). Based on the principles of uniform linear kinematics and combined with the sensor sampling period, a state transition matrix is ​​constructed, and an observation matrix describing the uncertainty of the measurement system is established. At the beginning of each control cycle, this embodiment uses the optimal posterior estimate calculated in the previous time step to perform time update calculations using the state transition matrix, deduce the prior state estimate for the current time step, and synchronously update the prediction covariance matrix. This allows for a mathematical prediction of the theoretically correct lateral position and motion trend of the carrier unit at the current time step.

[0058] This embodiment uses the path observation deviation value calculated above as the actual observation input, calculates the residual between it and the prior state estimate, and combines the prediction covariance matrix with the preset measurement noise covariance matrix to solve for the Kalman gain matrix. This embodiment uses the Kalman gain matrix to perform weighted correction on the prior state estimate. If the observation noise is large, the algorithm automatically reduces the confidence weight of the current observation value; conversely, it increases the weight, thereby obtaining the optimal posterior state estimate containing the minimum mean square error. Finally, this embodiment extracts the position component from the optimal posterior state estimate and uses it as the basic lateral deviation correction amount after filtering and smoothing, which is then fed into the underlying motion control algorithm to achieve smooth and interference-resistant path tracking control.

[0059] Furthermore, the specific implementation process of step 500 is as follows:

[0060] In this embodiment, as the transport unit travels along the initially planned path, the background control program continuously traverses the node data in the path list, calculating in real time the spatial Euclidean distance between the transport unit's current dynamic positioning coordinates and the next static topological coordinates to be traversed. This embodiment sets a specific visual capture radius as a judgment benchmark, which is typically calibrated based on the camera's field of view and optimal recognition distance. When the real-time calculated spatial Euclidean distance is detected to be less than the preset visual capture radius, this embodiment immediately determines that the transport unit has entered the node region and generates a visual activation trigger signal, thereby activating the visual recognition subsystem. This on-demand triggering mechanism effectively avoids the visual module idling throughout the entire process, reducing the system's computational power consumption.

[0061] In response to the vision activation trigger signal, this embodiment controls the machine vision module to acquire images of the node scene in front of the current field of view. To eliminate background clutter interference, this embodiment performs color segmentation on the original image based on a pre-set color threshold space, quickly locates the region of interest containing the digital identifier, and performs image cropping. Subsequently, this embodiment uses an adaptive thresholding algorithm to binarize the cropped image. This algorithm can automatically calculate the threshold based on the local brightness distribution of the image, thereby effectively filtering out ambient lighting noise caused by uneven corridor lighting or ground reflection, generating a clear black and white binary feature image, providing a high-quality data foundation for subsequent feature matching.

[0062] This embodiment calls a pre-stored standard digital feature template and uses a normalized cross-correlation algorithm to perform high-precision search and matching in the binarized feature image. Considering that the vehicle unit may experience angular deflection during operation, this embodiment introduces a rotation correction mechanism in the matching process. The binarized feature image is rotated at multiple levels within a preset angle range with a fixed step size, generating a series of rotated candidate sub-images. The matching confidence of each candidate sub-image with the template is calculated one by one, and the matching result with the highest confidence is selected as the optimal matching target. This embodiment extracts the pixel center coordinates and corresponding rotation angle of the target in the image coordinate system. Combined with the camera's intrinsic parameter matrix and installation height parameters, the pixel deviation of the image plane is inversely solved into geometric position offset and heading angle deviation values ​​in the world coordinate system using the perspective transformation principle. Finally, a visual correction vector is generated for subsequent motion control correction.

[0063] Furthermore, the specific implementation process of step 600 is as follows:

[0064] This embodiment performs orthogonal decomposition on the received visual correction vector, breaking it down into a heading angle compensation component representing the direction adjustment requirement and a longitudinal distance compensation component representing the stopping or deceleration requirement. To address the issue of inconsistent dimensions among multi-source data, this embodiment introduces a normalization mapping mechanism. Based on the correspondence between the camera's field of view and the vehicle unit's steering sensitivity, the basic lateral deviation correction amount in units of distance is converted into a tracking deviation component in units of angle. This ensures that the tracking data from the grayscale sensor and the correction data from the vision module are additive, thus achieving data alignment at the control input layer.

[0065] This embodiment constructs a composite control strategy based on feedforward compensation and main feedback adjustment. The normalized tracking deviation component is used as the main feedback quantity to maintain straight-line travel, and the visually calculated heading angle compensation component is used as the predictive feedforward correction quantity. The total heading error of the system is obtained through linear superposition calculation, thereby eliminating the lag phenomenon that may occur at nodes in single feedback control. Simultaneously, this embodiment calculates the difference between the longitudinal distance compensation component and the target cruise speed set for the current mission, generating the system speed error. This embodiment further designs a dual-loop controller. The total heading error of the system is input into the position loop PD controller, which uses proportional and derivative terms for fast response and oscillation suppression to calculate the target angular velocity adjustment. The system speed error is input into the speed loop PI controller, which uses proportional and integral terms to eliminate steady-state error and calculate the target linear velocity reference quantity.

[0066] This embodiment establishes a differential kinematic model based on the physical and mechanical structure of the carrier unit. The lateral distance between the grounding centers of the left and right drive wheels, i.e., the wheelbase parameter, is pre-measured and stored. This embodiment substitutes the previously calculated target angular velocity adjustment and target linear velocity reference values ​​into the kinematic model, and calculates the required target speeds for the left and right drive wheels according to the differential steering principle. Specifically, this embodiment obtains independent control commands for the left and right wheels by superimposing or subtracting the differential component converted from the product of angular velocity and half the wheelbase from the reference linear velocity. These two target speeds are then output as real-time motion control parameters to the underlying motor driver, thereby driving the carrier unit to precisely execute heading correction and speed adjustment actions.

[0067] The expression for the differential kinematics model is:

[0068] ;

[0069] in, and These represent the target linear velocities of the left and right drive wheels, respectively. The target linear velocity reference quantity is calculated and output by the speed loop PI controller; The target angular velocity adjustment is calculated and output by the position loop PD controller; The wheelbase parameter represents the wheelbase of the transport unit.

[0070] Specifically, after obtaining the target linear velocity reference value and the target angular velocity adjustment value, this embodiment uses a pre-constructed differential kinematic model to transform these two abstract control commands into speed commands executable by the underlying motor. Specifically, this model, based on the rigid body motion characteristics of the vehicle unit, decomposes the overall motion of the vehicle into translation and rotation around the center point. To calculate the target execution linear velocity of the left drive wheel, this embodiment uses the target linear velocity reference value as a base value and subtracts a differential adjustment component from it; wherein, the differential adjustment component is equal to half the product of the target angular velocity adjustment value and the wheelbase parameter. Correspondingly, to calculate the target execution linear velocity of the right drive wheel, this embodiment uses the target linear velocity reference value as a base value and adds the same differential adjustment component to it. The wheelbase parameter mentioned here specifically refers to the lateral physical distance between the grounding center points of the left and right drive wheels of the vehicle unit. Through this logic of increasing and decreasing, when the target angular velocity adjustment is not zero, the two drive wheels will generate a speed difference, thereby generating the torque required for steering, driving the carrier unit to perform precise heading correction while maintaining the overall forward speed.

[0071] To further illustrate the above calculation process, a numerical simulation is performed using a specific control cycle in this embodiment as an example: First, by measuring the chassis structure of the vehicle unit in the field, the wheelbase parameter is determined to be 0.2 meters. Second, during system operation, the speed loop proportional-integral controller calculates the target linear velocity reference value as 0.5 meters per second based on the current task requirements. This value represents the straight-line cruising speed that the vehicle unit expects to maintain. Simultaneously, the position loop proportional-derivative controller calculates the target angular velocity adjustment value as 0.2 radians per second based on the visually recognized heading deviation. This value represents the directional magnitude by which the vehicle unit needs to make a slight leftward adjustment. Based on the above parameters, the system first calculates the differential adjustment component, i.e., 0.2 radians per second multiplied by 0.2 meters and then divided by 2, resulting in 0.02 meters per second. Subsequently, the system performs the final composite calculation: the target execution linear velocity of the left drive wheel is equal to 0.5 minus 0.02, i.e., 0.48 meters per second; the target execution linear velocity of the right drive wheel is equal to 0.5 plus 0.02, i.e., 0.52 meters per second. Ultimately, the underlying driver controls the right wheel to rotate at a slightly higher speed than the left wheel, thereby driving the carrier unit to achieve a smooth left-turn correction.

[0072] Furthermore, the specific implementation process of step 700 is as follows:

[0073] This embodiment establishes a kinematic recursive model based on discrete time steps, setting the current dynamic positioning coordinates of the carrier unit as the initial state at time zero, and using the real-time acquired linear and angular velocities as input control variables. This embodiment sets a time window covering several seconds into the future, performing frame-by-frame iterative calculations according to a preset fixed time step. In each frame's calculation, the linear velocity is decomposed into lateral and longitudinal components based on the heading angle determined in the previous frame. The planar coordinates of the next frame are updated through numerical integration, and the heading angle of the next frame is updated based on the angular velocity. Finally, this embodiment associates and stores the spatial coordinates corresponding to each discrete moment with an absolute timestamp, generating a future trajectory composed of a series of time-stamped spatial point sets.

[0074] This embodiment maps the future trajectories generated by all active vehicle units to a grid map or node graph of the global topology network. A spatial indexing algorithm quickly searches for shared static topological coordinate regions among the trajectory sets of different vehicle units, such as the same intersection node or one-way corridor segment. If spatial overlap is detected, this embodiment further extracts the estimated timestamps of each vehicle unit's arrival at the center point of the shared region and calculates the time difference between the relevant vehicle units. This embodiment compares the absolute value of this time difference with a preset safety time interval threshold. If the absolute value of the time difference is less than the threshold, it means that the two vehicles will occupy the same physical space in extremely close proximity, posing a collision risk. This embodiment then determines that the current state is a spatiotemporal overlap and triggers subsequent arbitration logic.

[0075] To quantify the passage priority of different transport units, this embodiment extracts the task attributes and motion status data of the transport units involved in real time, specifically including the task urgency level, current instantaneous speed value, and remaining path length from the target destination. This embodiment uses the minimax method to perform dimensionless normalization processing on the above heterogeneous data, mapping it to a numerical range of zero to one. This embodiment constructs a multi-dimensional weighted scoring formula, setting the task urgency level and current instantaneous speed value as positive gain factors, i.e., the more urgent the task or the faster the speed, the higher the score, to encourage the rapid passage of emergency supplies and reduce energy consumption caused by high-speed emergency braking; simultaneously, the remaining path length from the target is set as a negative attenuation factor, i.e., the farther from the destination, the lower the score. This embodiment multiplies each normalized value by its corresponding preset weight and then performs algebraic summation to calculate and output a unique scalar value as a dynamic priority score, which serves as the basis for determining right-of-way allocation.

[0076] Furthermore, the specific implementation process of step 800 is as follows:

[0077] This embodiment first compares the dynamic priority scores of each transport unit involved in the spatiotemporal overlap. Based on the scores, the vehicles involved are divided into different logical roles. The transport unit with the highest score is marked as the passing party, and the other transport units with lower scores are marked as the yielding party. For the transport unit marked as the yielding party, this embodiment reads its current odometer value in real time. By comparing it with map data, it calculates the physical distance between the current front position of the transport unit and the static topological coordinate boundary of the conflict area, and defines it as the remaining available braking distance. This embodiment combines the maximum braking deceleration parameters preset by the transport unit and uses kinematic formulas to generate a smooth trapezoidal or S-shaped deceleration curve. By gradually reducing the motor speed command, the yielding party is driven to smoothly reduce its speed to zero before reaching the boundary line of the conflict area, and then the traction power is cut off. The transport unit is controlled to enter a position holding waiting state, and the parking brake is activated to prevent the vehicle from rolling away.

[0078] This embodiment targets the vehicle unit marked as the passing party. In the global topology network data structure managed in the background, a software-level exclusive occupancy lock is set for the static topology coordinate region with spatiotemporal overlap that the vehicle is about to enter. This status bit indicates that the physical region is exclusively accessible to the passing party within the current time window. To prevent the passing party from mistakenly triggering its own vehicle collision avoidance braking mechanism due to the detection of a stationary vehicle waiting to yield on the roadside during the crossing process, this embodiment temporarily disables the ultrasonic or lidar collision avoidance deceleration logic of the passing party for the current specific conflict area through low-level instructions, establishing a temporary safety trust whitelist. This allows the passing party to maintain its currently set cruising speed or perform acceleration actions, quickly and without interruption crossing the conflict area to maximize intersection traffic efficiency.

[0079] This embodiment continuously monitors the real-time dynamic positioning coordinates of the passing vehicle using high-frequency positioning data, calculates its relative positional relationship with the boundary of the conflict area, and confirms that the right-of-way conflict has been resolved when it is determined that the rear coordinates of the passing vehicle have completely crossed the exit boundary of the conflict area. This embodiment then releases the exclusive occupancy lock for that area in the global topology network, resetting the area status to idle. Simultaneously, this embodiment sends a reset trigger signal to the yielding vehicle in a waiting state via the communication link, driving the yielding vehicle to release its parking brake and exit its waiting position, reloading its initial planned path data before the interruption, and activating the tracking control algorithm to resume normal material delivery tasks.

[0080] Furthermore, the specific implementation process of step 900 is as follows:

[0081] In this embodiment, throughout the entire process of the transport unit moving towards the target node, the aforementioned path tracking, visual correction, and dynamic right-of-way arbitration subroutines are called cyclically at millisecond intervals to ensure that the transport unit always follows the planned path and avoids dynamic conflicts. Simultaneously, this embodiment calculates in real-time the straight-line distance between the transport unit's current dynamic positioning coordinates and the static topological coordinates of the target node specified in the task order, and continuously compares this distance with a preset arrival determination threshold. When the straight-line distance is detected to be less than the arrival determination threshold, and the transport unit's real-time speed has decreased to zero according to the deceleration curve, this embodiment determines that the transport unit has successfully arrived at the predetermined unloading area, and then sends a command to terminate the chassis motor's motion control logic, keeping the vehicle in a parking brake state.

[0082] After confirming a stable stop, this embodiment reactivates the visual perception module and controls the camera to perform a secondary, detailed scan of the side-mounted bed label, ward door sign, or pharmacy window sign. This embodiment utilizes image recognition algorithms to extract QR codes, barcodes, or specific character features from the scanned image, and compares the decoded content with the target ID in the current task instruction to prevent misdelivery due to positioning drift. If the comparison results match, this embodiment generates an unloading permission signal, driving the electromagnetic lock or mechanical latch of the vehicle's storage compartment to unlock. A voice prompt is issued through the vehicle's human-machine interface, or an LED indicator flashes at a specific frequency, guiding medical personnel to retrieve medical supplies without contact.

[0083] This embodiment utilizes an infrared beam sensor array or gravity sensor installed inside the locker to monitor the status of items inside in real time. Once it detects that items have been removed and the locker door has closed again, a status update procedure is immediately executed. This embodiment modifies the task load status field in the status vector, changing it from a fully loaded delivery state to a task completed state or an idle waiting state, and simultaneously updates the battery level data. This embodiment uploads the updated full-dimensional status vector to the central dispatch terminal via a wireless network and actively queries the task queue. If a new task exists in the queue or a return-to-base charging instruction is received, the current static topology coordinates are used as the new starting point to trigger the next round of path planning and reconstruction loop, thereby achieving a continuous and automated closed-loop delivery service.

[0084] This embodiment also provides a dual-vehicle collaborative contactless medical supplies intelligent delivery system, including:

[0085] The network construction and status acquisition module is used to construct a global topology network based on a preset delivery environment and acquire the status vectors of multiple transport units in real time. The status vectors include: the dynamic positioning coordinates of the transport unit, battery power data and task load status. The global topology network includes: multiple static topology coordinates and path weight data connecting the static topology coordinates.

[0086] The comprehensive cost calculation module is used to calculate the comprehensive execution cost of the task to be executed relative to each of the carrier units based on the global topology network and the state vector;

[0087] The task allocation and path planning module is used to obtain the optimal task allocation sequence based on the comprehensive execution cost, and to plan the corresponding initial planning path based on the optimal task allocation sequence.

[0088] The deviation prediction and correction module is used to collect grayscale distribution data of the ground path in real time during the execution along the initial planned path, and use the Kalman filter algorithm to predict the grayscale distribution data to obtain the basic lateral deviation correction amount.

[0089] The node visual correction module is used to calculate the Euclidean distance between the dynamic positioning coordinates and the static topological coordinates in real time. When the Euclidean distance is less than a preset trigger threshold, the visual correction process for the current node is activated to obtain the visual correction vector.

[0090] The multi-source fusion control module is used to obtain real-time motion control parameters based on the basic lateral deviation correction amount and the visual correction vector to drive the carrier unit to correct the heading angle and driving speed.

[0091] The dynamic right-of-way arbitration module is used for a dynamic right-of-way arbitration mechanism based on the spatiotemporal dimension. It predicts the future running trajectory of the carrier unit according to the real-time motion control parameters. When it is detected that the future running trajectories of different carrier units overlap spatiotemporally in the same static topological coordinate region, it calculates the dynamic priority score of each carrier unit.

[0092] The conflict resolution execution module is used to generate a conflict resolution strategy based on the dynamic priority score, and after the spatiotemporal overlap is removed, control the vehicle unit to resume tracking the initial planned path. The conflict resolution strategy includes: controlling the vehicle unit with the lower dynamic priority score to perform deceleration and avoidance actions, and locking the right-of-way of the vehicle unit with the higher dynamic priority score.

[0093] The closed-loop monitoring and interaction module is used to continuously perform path tracking, correction and arbitration steps until the dynamic positioning coordinates coincide with the static topological coordinates of the target node. Based on visual positioning, the unloading of materials is confirmed, and the state vector is updated to trigger the next round of path reconstruction loop.

[0094] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0095] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A dual-vehicle collaborative contactless intelligent delivery method for medical supplies, characterized in that, include: A global topology network is constructed based on a preset delivery environment, and the state vectors of multiple transport units are acquired in real time. The state vectors include: the dynamic positioning coordinates of the transport unit, battery power data, and task load status. The global topology network includes: multiple static topology coordinates and path weight data connecting the static topology coordinates. The comprehensive execution cost of the task to be executed relative to each of the carrier units is calculated based on the global topology network and the state vector. The optimal task allocation sequence is obtained based on the comprehensive execution cost, and the corresponding initial planning path is planned based on the optimal task allocation sequence. During the execution along the initial planned path, grayscale distribution data of the ground path is collected in real time, and the grayscale distribution data is predicted using the Kalman filter algorithm to obtain the basic lateral deviation correction amount. The Euclidean distance between the dynamic positioning coordinates and the static topological coordinates is calculated in real time. When the Euclidean distance is less than a preset trigger threshold, the visual correction process for the current node is activated to obtain the visual correction vector. Real-time motion control parameters are obtained based on the basic lateral deviation correction amount and the visual correction vector to drive the carrier unit to correct the heading angle and travel speed. The dynamic right-of-way arbitration mechanism based on spatiotemporal dimensions predicts the future running trajectory of the carrier unit according to the real-time motion control parameters. When the future running trajectories of different carrier units are detected to overlap spatiotemporally in the same static topological coordinate region, the dynamic priority score of each carrier unit is calculated. A conflict resolution strategy is generated based on the dynamic priority score, and after the spatiotemporal overlap is removed, the vehicle unit is controlled to resume tracking the initial planned path. The conflict resolution strategy includes: controlling the vehicle unit with the lower dynamic priority score to perform deceleration and avoidance actions, and locking the right-of-way of the vehicle unit with the higher dynamic priority score. The path tracking, correction, and arbitration steps are continuously executed until the dynamic positioning coordinates are detected to coincide with the static topological coordinates of the target node. Based on visual positioning, the unloading of materials is confirmed, and the state vector is updated to trigger the next round of path reconstruction loop.

2. The intelligent delivery method for contactless medical supplies via dual-vehicle collaboration according to claim 1, characterized in that, A global topology network is constructed based on a pre-defined delivery environment, and the state vectors of multiple transport units are acquired in real time, including: The intersections, ward entrances, and pharmacy loading points in the delivery environment are discretized into graph theory nodes, and the absolute position of each graph theory node is calibrated in the global coordinate system to generate the static topological coordinates. Based on the physical road network structure of the delivery environment, an adjacency matrix describing the connection relationship between nodes is established, and the physical path length between adjacent static topological coordinates is normalized with the preset turning cost to generate the initial path weight data, so as to determine the construction of the global topological network. The system periodically receives data frames sent by each of the aforementioned carrier units, unpacks and verifies the validity of the data frames, extracts real-time mileage count values ​​and visual positioning identifiers, and obtains the dynamic positioning coordinates. Based on the data frame, the voltage reading and task queue flag are extracted to obtain battery power data and task load status.

3. The intelligent delivery method for contactless medical supplies via dual-vehicle collaboration according to claim 1, characterized in that, The comprehensive execution cost of the task to be executed relative to each of the carrier units is calculated based on the global topology network and the state vector, including: The target location of the task to be executed is analyzed, and the target location is mapped to target static topological coordinates; In the global topology network, a path search algorithm is used to calculate the shortest path length from the current dynamic positioning coordinates of the carrier unit to the target static topology coordinates, which is used as the quantized value of the Euclidean distance of the path. Read the battery power data in the state vector, construct an inverse proportional energy penalty function to obtain an energy state score, wherein the inverse proportional energy penalty function is configured to output an exponentially increasing cost value when the battery power data is lower than a preset safety threshold, and output a linear cost value that is inversely proportional to the remaining power when the battery power data is higher than the safety threshold. Extract the path weight data of all paths located on the shortest path in the global topology network, and combine it with the historical traffic frequency statistics at the current moment to calculate the cumulative sum of the path weight data, so as to obtain the road segment congestion coefficient that reflects the traffic resistance of the current road segment; Preset weight adjustment factors corresponding to distance, energy and congestion dimensions respectively. Use the weight adjustment factors to linearly weight and sum the quantified value of the Euclidean distance of the path, the energy status score and the road segment congestion coefficient, and output a unique numerical result as the comprehensive execution cost.

4. The intelligent delivery method for contactless medical supplies via dual-vehicle collaboration according to claim 1, characterized in that, The optimal task allocation sequence is obtained based on the comprehensive execution cost, and the corresponding initial planning path is planned based on the optimal task allocation sequence, including: A cost mapping matrix is ​​constructed using the comprehensive execution cost, wherein the row vectors of the cost mapping matrix correspond to different carrier units, the column vectors correspond to different tasks to be executed, and the matrix elements are the corresponding comprehensive execution cost values; With the goal of minimizing the total system cost, the cost mapping matrix is ​​solved by combination optimization to determine the carrier unit and execution order of each task to be executed, and to generate the optimal task allocation sequence. The optimal task allocation sequence is analyzed, the target position of the current task is extracted and anchored as the target static topological coordinates, the dynamic positioning coordinates of the carrier unit to which the current task belongs are read, and the graph theory node with the closest Euclidean distance to the dynamic positioning coordinates is searched in the global topological network to obtain the initial static topological coordinates. A heuristic graph search algorithm is used to optimize the path between the initial static topological coordinates and the target static topological coordinates. During the optimization process, the path weight data is used as the traversal cost of the graph edges. The minimum cost path is calculated by accumulating the data, and a list of nodes consisting of a series of consecutive static topological coordinates is output as the initial planned path.

5. The intelligent delivery method for contactless medical supplies via dual-vehicle collaboration according to claim 1, characterized in that, During the execution along the initially planned path, grayscale distribution data of the ground path is collected in real time, and the grayscale distribution data is predicted using a Kalman filter algorithm to obtain the basic lateral deviation correction amount, including: The grayscale distribution data collected in real time is subjected to binarization threshold segmentation to extract the left and right boundary indices of the path; Calculate the geometric center of the left boundary index and the right boundary index, and define the difference between the geometric center and the physical center of the image sensor of the carrier unit as the path observation deviation value at the current moment; Construct a system state vector that includes the lateral deviation state and the deviation change rate state, and establish a state transition matrix that describes the linear motion characteristics of the carrier unit, as well as an observation matrix that describes the noise characteristics of the gray-scale distribution data; Based on the optimal posterior estimate of the previous moment, the prior state estimate of the current moment is derived using the state transition matrix, and the prediction covariance matrix is ​​updated synchronously to predict the theoretical lateral position of the carrier unit at the current moment. Calculate the residual between the path observation bias and the prior state estimate, and calculate the Kalman gain by combining the prediction covariance matrix; The prior state estimate is weighted and corrected using the Kalman gain to obtain the optimal posterior state estimate. The lateral deviation component in the optimal posterior state estimate is then output as the basic lateral deviation correction.

6. The intelligent delivery method for contactless medical supplies via dual-vehicle collaboration according to claim 1, characterized in that, The real-time calculation of the Euclidean distance between the dynamic positioning coordinates and the static topological coordinates, and the activation of the visual correction process for the current node when the Euclidean distance is less than a preset trigger threshold, to obtain the visual correction vector, including: Traverse the list of nodes in the initial planned path and calculate in real time the spatial Euclidean distance between the current dynamic positioning coordinates of the carrier unit and the next static topological coordinate to be traversed. When the spatial Euclidean distance is detected to be less than the preset visual capture radius, a visual activation trigger signal is generated; In response to the vision activation trigger signal, acquire the node scene image of the current field of view; The node scene image is cropped based on a preset color threshold space, and the cropped image is converted into a binary feature image using an adaptive threshold algorithm to filter out ambient lighting noise. The pre-stored digital feature template is invoked, and a normalized cross-correlation algorithm is used to search and match in the binarized feature image. During the matching process, the binarized feature image is subjected to multi-level rotation transformation within a preset angle range to generate a series of rotation candidate sub-images. The matching confidence of each rotation candidate sub-image with the digital feature template is calculated, and the matching result with the highest confidence is selected as the optimal matching target. Extract the pixel center coordinates and corresponding rotation angle of the optimal matching target in the image coordinate system, and calculate the pixel deviation from the physical center of the image; Based on the camera's intrinsic parameter matrix and installation height, the pixel deviation and rotation angle are converted into geometric position offset and heading angle deviation values ​​in the world coordinate system, and combined to generate the visual correction vector.

7. The intelligent delivery method for contactless medical supplies via dual-vehicle collaboration according to claim 1, characterized in that, The step of obtaining real-time motion control parameters based on the basic lateral deviation correction amount and the visual correction vector to drive the vehicle unit to correct its heading angle and speed includes: The visual correction vector is decomposed into a heading angle compensation component and a longitudinal distance compensation component; The basic lateral deviation correction is normalized to a tracking deviation component with the same dimensions as the heading angle compensation component; Using the tracking deviation component as the main feedback quantity and the heading angle compensation component as the feedforward correction quantity, the total heading error of the system is obtained by linear superposition calculation. Based on the longitudinal distance compensation component and the target velocity of the current task, the system velocity error is calculated; Construct a position loop PD controller and a velocity loop PI controller, input the total heading error of the system into the position loop PD controller to calculate the target angular velocity adjustment, and input the system velocity error into the velocity loop PI controller to calculate the target linear velocity reference. A differential kinematic model is established based on the wheel track parameters of the transport unit. The target rotational speeds of the left and right drive wheels are calculated using the target angular velocity adjustment and the target linear velocity reference. The target rotational speeds are output as the real-time motion control parameters to drive the transport unit to perform differential steering and speed adjustment. The expression for the differential kinematics model is: in, and These represent the target linear velocities of the left and right drive wheels, respectively. The target linear velocity reference quantity is calculated and output by the speed loop PI controller; The target angular velocity adjustment is calculated and output by the position loop PD controller; The wheelbase parameter represents the wheelbase of the transport unit.

8. The intelligent delivery method for contactless medical supplies via dual-vehicle collaboration according to claim 1, characterized in that, The spatiotemporal dynamic right-of-way arbitration mechanism predicts the future trajectory of the transport unit based on the real-time motion control parameters. When it detects that the future trajectories of different transport units spatiotemporally overlap in the same static topological coordinate region, it calculates the dynamic priority score of each transport unit, including: Using the dynamic positioning coordinates as the initial state and the real-time motion control parameters as input variables, frame-by-frame position integration and attitude recursion are performed within a preset future time window to generate the future running trajectory containing a timestamp sequence and a spatial coordinate sequence. The future operating trajectories of different carrier units are mapped to the global topology network, and the existence of a shared static topology coordinate region is searched. If a shared region exists, the time difference between the arrival of different carrier units in the shared region is calculated. When the absolute value of the time difference is less than a preset safety time interval threshold, it is determined that a spatiotemporal overlap has occurred. Extract the mission attributes and motion states of each of the aforementioned carrier units that have spatiotemporal overlap, and normalize the mission urgency level, current instantaneous velocity value, and remaining path length to the target. A weighted summation algorithm is used, with the task urgency level and the current instantaneous speed value as positive gain factors and the remaining path length to the target as negative attenuation factors, to calculate and output a unique scalar value as the dynamic priority score.

9. The intelligent delivery method for contactless medical supplies via dual-vehicle collaboration according to claim 1, characterized in that, The step of generating a conflict resolution strategy based on the dynamic priority score, and controlling the transport unit to resume tracking the initial planned path after the spatiotemporal overlap is resolved, includes: Compare the dynamic priority scores of the carrier units that have spatiotemporal overlap, mark the carrier unit with the highest score as the passing party, and mark the carrier unit with the lower score as the yielding party; Calculate the safe braking distance from the current position of the avoiding party to the boundary of the static topological coordinate region; Based on the safe braking distance, a deceleration speed curve is generated, driving the avoidance vehicle to perform a stopping action before entering the static topological coordinate region and enter a position holding waiting state; Based on the aforementioned access method, an exclusive occupancy lock is set for the static topological coordinate region where spatiotemporal overlap occurs in the global topological network; The system blocks the collision avoidance and deceleration commands issued by the passing party to the current area, allowing it to maintain its current speed or accelerate through the area. The system monitors the dynamic positioning coordinates of the passing vehicle in real time. When it is determined that the passing vehicle has completely left the static topology coordinate area, the exclusive occupancy lock is released, and a reset trigger signal is sent to the avoiding vehicle to drive the avoiding vehicle to exit the position and remain in a waiting state.

10. A dual-vehicle collaborative contactless intelligent delivery system for medical supplies, characterized in that, include: The network construction and status acquisition module is used to construct a global topology network based on a preset delivery environment and acquire the status vectors of multiple transport units in real time. The status vectors include: the dynamic positioning coordinates of the transport unit, battery power data and task load status. The global topology network includes: multiple static topology coordinates and path weight data connecting the static topology coordinates. The comprehensive cost calculation module is used to calculate the comprehensive execution cost of the task to be executed relative to each of the carrier units based on the global topology network and the state vector; The task allocation and path planning module is used to obtain the optimal task allocation sequence based on the comprehensive execution cost, and to plan the corresponding initial planning path based on the optimal task allocation sequence. The deviation prediction and correction module is used to collect grayscale distribution data of the ground path in real time during the execution along the initial planned path, and use the Kalman filter algorithm to predict the grayscale distribution data to obtain the basic lateral deviation correction amount. The node visual correction module is used to calculate the Euclidean distance between the dynamic positioning coordinates and the static topological coordinates in real time. When the Euclidean distance is less than a preset trigger threshold, the visual correction process for the current node is activated to obtain the visual correction vector. The multi-source fusion control module is used to obtain real-time motion control parameters based on the basic lateral deviation correction amount and the visual correction vector to drive the carrier unit to correct the heading angle and driving speed. The dynamic right-of-way arbitration module is used for a dynamic right-of-way arbitration mechanism based on the spatiotemporal dimension. It predicts the future running trajectory of the carrier unit according to the real-time motion control parameters. When it is detected that the future running trajectories of different carrier units overlap spatiotemporally in the same static topological coordinate region, it calculates the dynamic priority score of each carrier unit. The conflict resolution execution module is used to generate a conflict resolution strategy based on the dynamic priority score, and after the spatiotemporal overlap is removed, control the vehicle unit to resume tracking the initial planned path. The conflict resolution strategy includes: controlling the vehicle unit with the lower dynamic priority score to perform deceleration and avoidance actions, and locking the right-of-way of the vehicle unit with the higher dynamic priority score. The closed-loop monitoring and interaction module is used to continuously perform path tracking, correction and arbitration steps until the dynamic positioning coordinates coincide with the static topological coordinates of the target node. Based on visual positioning, the unloading of materials is confirmed, and the state vector is updated to trigger the next round of path reconstruction loop.

Images