Method and system for vehicle-to-pedestrian coordination in avp pick-up phase

Abstract

The application discloses a kind of AVP pick-up stage under human-vehicle coordination arrival prediction and scheduling method and system, it is related to electric digital data processing technical field, method includes: extracting the user and vehicle coordinate in historical period to construct space-time feature cluster;Extract trajectory divergence value as motion divergence, combine the noise tolerance value based on map semantics to calculate confidence mapping kernel, to generate vehicle displacement constraint boundary;Combine vehicle steering geometry value with the constraint boundary to construct two-dimensional feasible region, calculate the density value of vehicle history point falling into the region, when density value meets the standard, fusion current double-end coordinate obtains timing phase difference;Finally, according to the phase difference matching virtual anchor point in map semantics, issue parking or speed regulation instruction.The application has the advantages of strong anti-interference, high scheduling accuracy and stable and reliable.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing technology, specifically to a method and system for predicting and scheduling human-vehicle cooperative arrival during the AVP vehicle retrieval phase. Background Technology

[0002] In current Automated Valet Parking (AVP) retrieval scenarios, the goal is to achieve accurate meeting between people and vehicles at the retrieval point. However, existing technologies face physical and kinematic limitations in complex indoor building topologies.

[0003] First, the underground parking environment, with its concrete load-bearing walls and overhead metal cable trays, causes the positioning signals of user mobile terminals to be affected by multipath effects and electromagnetic reflection interference, resulting in high-frequency trajectory jumps. Existing systems often use a globally fixed single threshold to filter noise, making it difficult for users in weak signal areas such as elevator lobbies to pass stability checks, thus affecting the normal response of the collaborative scheduling system. Second, when assessing the consistency of vehicle trajectories, existing methods often rely on scalar comparisons within a one-dimensional radial distance interval, without considering the Ackerman steering geometry and kinematic constraints of the vehicle chassis. This easily leads to misjudging lateral displacements generated during normal cornering or obstacle avoidance as positioning drift, resulting in the passive cancellation of scheduling commands and system control oscillations. Finally, when generating scheduling compensation instructions, existing technologies mostly adopt continuous linear scaling logic, that is, when the system predicts that the vehicle will arrive earlier than the user, it directly reduces the real-time cruising speed of the vehicle proportionally. However, in the network topology of underground parking garages with limited space and mostly one-way streets where overtaking is not allowed, this cooperative strategy of making vehicles travel long distances at low speeds will block the normal traffic flow behind, thereby causing congestion in the internal road network of the parking lot. Summary of the Invention

[0004] In response to the technical problems described in the background art, the present invention provides a method and system for predicting and scheduling human-vehicle cooperative arrival during the AVP vehicle retrieval stage.

[0005] A method for predicting and scheduling human-vehicle cooperative arrival during the AVP (Automated Guided Vehicle) vehicle retrieval phase includes: obtaining the prediction period at the current moment, extracting user coordinates and vehicle coordinates within the historical period, and constructing a spatiotemporal feature cluster; extracting trajectory divergence values ​​as motion divergence based on the spatiotemporal feature cluster, calculating a noise tolerance value based on map semantics, and calculating a confidence mapping kernel to generate vehicle displacement constraints when the motion divergence is less than or equal to the noise tolerance value; constructing a two-dimensional feasible region by combining the vehicle's steering geometry and displacement constraints, calculating the density value of vehicle historical points falling into the two-dimensional feasible region, and if the density value is greater than or equal to a preset lower convergence limit, fusing the user coordinates and vehicle coordinates at the current moment to obtain the temporal phase difference; matching virtual anchor points in map semantics based on the temporal phase difference and a preset standard time difference, and issuing parking or speed adjustment commands.

[0006] Optionally, a spatiotemporal feature cluster is constructed, including: obtaining a time reference for absolute time and clock-aligning the time axis for extracting coordinates; dividing the time axis into continuous and non-overlapping prediction periods; and within the historical period, arranging user coordinates and vehicle coordinates in ascending order of timestamps to form a spatiotemporal feature cluster containing coordinates and timestamps.

[0007] Optionally, the trajectory divergence value is extracted as the motion divergence, including: extracting the difference vector between the displacement vectors between two adjacent sampling periods; accumulating the magnitude of the difference vector, using the accumulated magnitude value to characterize the acceleration divergence, and using it as the motion divergence.

[0008] Optionally, calculating a noise tolerance value based on map semantics includes: extracting the grid attribute where the user coordinates are located in the map semantics; obtaining a basic fluctuation value; calculating an exponentially decaying distance compensation term based on the vertical distance between the user coordinates and the load-bearing wall; calculating a linear interference term in combination with the metal density value of the upper region; and summing the basic fluctuation value, the distance compensation term, and the linear interference term to obtain the noise tolerance value.

[0009] Optionally, calculating the confidence mapping kernel to generate the vehicle's displacement constraint boundary includes: obtaining the confidence mapping kernel by using the ratio of the difference between the noise tolerance value and the motion divergence to the noise tolerance value, and combining it with a preset upper limit of the mapping kernel; reading the half-width of the preset standard displacement boundary, multiplying it by the confidence mapping kernel to obtain the shrinkage compensation value, and using the upper and lower limits of the standard displacement boundary to reduce the shrinkage compensation value inward, respectively, to generate a displacement constraint boundary with physical constraints.

[0010] Optionally, a two-dimensional feasible region is constructed, and the density value of vehicle history points in the spatiotemporal feature cluster falling into the two-dimensional feasible region is calculated. This includes: reading the steering angle and wheelbase values ​​fed back by the chassis, and calculating the theoretical heading angle of the vehicle within the prediction period based on the steering geometry principle; constructing a two-dimensional feasible region with a fan-shaped envelope by combining the displacement constraint boundary and the theoretical heading angle; for each vehicle history point in the spatiotemporal feature cluster, when the actual displacement between adjacent points is within the displacement constraint boundary, and the deviation between the actual heading angle and the theoretical heading angle does not exceed the yaw tolerance scaled by the confidence mapping kernel, it is determined that the vehicle history point falls into the two-dimensional feasible region; the density value is obtained by dividing the number of vehicle history points falling into the two-dimensional feasible region by the total number of verification points.

[0011] Optionally, obtaining the temporal phase difference includes: inputting the user coordinates into a spatial mapping matrix based on map pathpoints, and obtaining the user's arrival time through interpolation; calculating the vehicle's arrival time by combining the vehicle coordinates, driving speed, and remaining mileage; and subtracting the vehicle's arrival time from the user's arrival time to obtain the temporal phase difference with algebraic sign properties.

[0012] Optionally, matching virtual anchor points in the map semantics and issuing parking or speed adjustment commands includes: when the timing phase difference is positive and greater than a preset parking threshold, and the map semantics indicate that the vehicle is currently in a one-way lane, blocking the proportional speed reduction command; searching for area coordinates that do not interfere with the main lane traffic flow along the vehicle's planned path towards the pick-up point as virtual anchor points, issuing parking commands to control the vehicle to drive to the virtual anchor point and brake; continuously calculating the user's remaining arrival time during the vehicle's parking period, and issuing a release parking command to allow the vehicle to start moving when the remaining arrival time is equal to the physical time required for the vehicle to start from the virtual anchor point and accelerate to the pick-up point; if the timing phase difference is less than or equal to the parking threshold, or the vehicle is in a wide environment, generating a proportional coefficient based on the timing phase difference and issuing a speed adjustment command.

[0013] A human-vehicle cooperative arrival prediction and scheduling system under the AVP vehicle retrieval stage is also provided. The system includes: an acquisition module, used to acquire the prediction period at the current moment, extract user coordinates and vehicle coordinates within the historical period, and construct a spatiotemporal feature cluster; a first processing module, used to extract trajectory divergence values ​​as motion divergence based on the spatiotemporal feature cluster, calculate a noise tolerance value based on map semantics, and calculate a confidence mapping kernel to generate the vehicle's displacement constraint boundary when the motion divergence is less than or equal to the noise tolerance value; a second processing module, used to combine the vehicle's steering geometry and displacement constraint boundary to construct a two-dimensional feasible region, calculate the density value of vehicle historical points in the spatiotemporal feature cluster falling into the two-dimensional feasible region, and if the density value is greater than or equal to a preset convergence lower limit value, fuse the user coordinates and vehicle coordinates at the current moment to obtain the temporal phase difference; and a control module, used to match virtual anchor points in map semantics based on the temporal phase difference and a preset standard time difference, and issue parking commands or speed adjustment commands.

[0014] Optionally, the acquisition module is also used to: acquire the time reference of absolute time, clock-align the time axis of the extracted coordinates; divide the time axis into continuous and non-overlapping prediction periods, and within the historical period, arrange the user coordinates and vehicle coordinates in ascending order of timestamps to form a spatiotemporal feature cluster containing coordinates and timestamps.

[0015] The beneficial effects of this invention are reflected in: In the entire AVP vehicle retrieval phase, the human-vehicle cooperative arrival prediction and scheduling method firstly achieves time axis alignment of heterogeneous data from both ends through the spatiotemporal feature cluster construction mechanism in S1, transforming discrete coordinate points into continuous kinematic sequences and filtering out noise interference caused by single sampling delays. Furthermore, S2 introduces a dynamic noise baseline tolerance calculation model based on map semantics, converting building topology attributes such as load-bearing wall distance and metal coverage into physical attenuation parameters, giving adaptive tolerance to weak signal areas such as elevator lobbies, and improving the scheduling lag problem caused by existing fixed thresholds. Further, S3 changes the single one-dimensional distance comparison method, combining chassis front wheel angle and wheelbase parameters to construct a two-dimensional physically feasible domain corridor following Ackerman steering geometry, distinguishing between normal cornering and positioning drift, and calculating the temporal phase difference based on this, improving the reliability of the prediction. Finally, S4 proposes a discontinuous spatiotemporal parking mechanism for narrow one-way road environments. When a vehicle arrives early, it changes its low-speed driving strategy and instead uses map semantic matching to execute parking and waiting at roadside virtual anchor points. A wake-up command is issued at the node where the user's remaining arrival time equals the physical time. This design reduces the occupation of public lane resources in indoor parking lots while enabling people and vehicles to meet, and optimizes the global traffic efficiency in multi-vehicle scheduling scenarios. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0017] Figure 1 This is a schematic diagram illustrating the steps of the human-vehicle cooperative arrival prediction and scheduling method in the AVP vehicle retrieval stage of the present invention; Figure 2 This is a schematic diagram of a portion of steps in S1 of the AVP vehicle retrieval stage method of the present invention; Figure 3 This is a schematic diagram of a portion of steps in step S2 of the AVP vehicle retrieval stage of the present invention. Figure 4This is a schematic diagram of part of step S3 in the AVP vehicle retrieval stage of the present invention; Figure 5 This is a schematic diagram of a portion of step S4 in the AVP vehicle retrieval stage of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0019] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0020] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] This invention provides a method for predicting and scheduling human-vehicle cooperative arrival during the AVP vehicle retrieval phase, such as... Figure 1 As shown, in one embodiment, the method includes: S1. Obtain the prediction period at the current moment, extract the user coordinates and vehicle coordinates within the historical period, and construct a spatiotemporal feature cluster; S2. Extract the trajectory divergence value as the motion divergence based on the spatiotemporal feature cluster, calculate the noise tolerance value based on map semantics, and when the motion divergence is less than or equal to the noise tolerance value, calculate the confidence mapping kernel to generate the vehicle's displacement constraint boundary. S3. Combining the vehicle's steering geometry and displacement constraints, construct a two-dimensional feasible region. Calculate the density value of the vehicle's historical points falling into the two-dimensional feasible region in the spatiotemporal feature cluster. If the density value is greater than or equal to the preset lower limit of convergence, then fuse the user coordinates and vehicle coordinates at the current moment to obtain the temporal phase difference. S4. Based on the time phase difference and the preset standard time difference, match the virtual anchor points in the map semantics and issue a stop command or speed adjustment command.

[0022] In this embodiment, it should be noted that in S1, during the automatic valet parking and vehicle retrieval stage, in order to solve the technical problems of asynchronous sampling delay between the mobile terminal and the vehicle positioning signal and the susceptibility of single-point data to environmental interference causing scheduling inaccuracies, a time-aligned spatiotemporal feature cluster is constructed through S1.

[0023] Specifically, a unified time benchmark is established using a network time protocol, and hardware-level clock synchronization is performed between the mobile terminal positioning sensor and the vehicle-mounted autonomous driving positioning system, setting the basic discrete prediction period to 200 milliseconds. At the current moment, not only are instantaneous spatial coordinates collected, but also two adjacent historical prediction periods are traced back along the historical timeline to extract high-frequency location data recorded by both ends within this timeframe. These timestamped discrete physical coordinate points are integrated and arranged chronologically to construct a multi-period spatiotemporal feature cluster. This data processing logic changes the existing collaborative methods that rely solely on instantaneous single sampling points for decision-making, transforming scattered and disordered location information into a continuous trajectory sequence with kinematically coherent properties. It solves the measurement error problem caused by time asynchrony, eliminates initial interference caused by single communication delays or accidental occlusion, and achieves a unified benchmark for the physical states of both the human and vehicle ends in both time and space dimensions, providing structured data support for subsequent extraction of features such as velocity and acceleration.

[0024] It should be noted that the basic discrete prediction cycle was determined based on the comprehensive historical test results of the vehicle sensor sampling frequency and data communication bus bandwidth. Specifically, during the R&D phase, data packets from multiple sources, including ultrasonic, visual, and mobile terminal sensors, were collected in various underground parking garage environments, and the time distribution curves of their stable arrival at the vehicle's central computing unit were statistically analyzed. For example, experiments showed that when the sampling window was less than 100 milliseconds, due to fluctuations in the underground network, the packet loss rate and asynchrony rate would surge to over 30%, resulting in invalid data splicing. When the window was greater than 300 milliseconds, the real-time performance of kinematic features lagged significantly, making it impossible to capture high-frequency trajectory jumps in a timely manner. By fitting the normal distribution of 100,000 communication delay data points, taking the upper limit of its 99% confidence interval coverage, and combining it with redundancy processing, the basic discrete prediction cycle was finally calibrated offline to 200 milliseconds, which satisfies both the need to ensure high synchronization rate of multi-source heterogeneous data and reflect the real-time response capability of dynamic scheduling.

[0025] In S2, to address the severe multipath effect of positioning signals caused by load-bearing walls and metal cable trays in underground parking garages, and the problem that existing fixed thresholds can easily misjudge users in weak signal areas such as elevator lobbies as untrustworthy, thus causing scheduling deadlock, S2 introduces a dynamic noise tolerance mechanism based on map semantics.

[0026] Specifically, by calculating and accumulating the magnitude of the coordinate difference vector, the divergence value of the current user trajectory is extracted to be 0.8 meters. The high-precision map is then used to obtain the vertical distance of the user from the load-bearing wall as 2.0 meters and the metal coverage density above it as 0.8. Taking a tolerance of 0.5 meters for the open foundation, and combining this with the load-bearing wall weight of 1.0 meter, which is exponentially reduced using a 2.0-meter distance and a 4.0-meter reference constant, plus the product of the metal weight of 0.5 meters and the density of 0.8, the dynamic noise tolerance value is calculated to be 1.5065 meters. Since the 0.8-meter divergence is within this tolerance, and combined with the preset upper limit of 0.8, the confidence mapping kernel is calculated to be 0.4689. Using this value, the standard displacement boundary from 0.2 to 1.0 meters is contracted and compensated by 0.187 meters, resulting in a new constraint boundary from 0.387 to 0.813 meters. This calculation logic establishes a numerical mapping between the judgment tolerance and the building topology attributes, which solves the technical problem of misjudging positioning fluctuations in complex occlusion environments, and gives the harsh environment a moderate adaptive physical tolerance, avoiding the state of cooperative interruption in weak signal areas.

[0027] In S3, after the deadlock in user-side signal determination is resolved, in order to overcome the problem that the existing one-dimensional radial distance comparison is prone to misjudging the normal lateral displacement of the vehicle when cornering as positioning drift, S3 constructs a two-dimensional physical feasible domain in combination with the vehicle chassis attributes.

[0028] Specifically, by reading the vehicle's wheelbase and front wheel steering angle data, the theoretical heading angle was calculated to be 0.523 radians. Combined with the displacement constraints of 0.387 to 0.813 meters generated in the previous step, and the standard yaw tolerance of 0.174 radians after scaling with a 0.4689 mapping kernel, a fan-shaped inspection corridor with directional restrictions was constructed. Historical points were extracted from the feature cluster, and their actual radial displacement was measured to be 0.6 meters, falling within the dynamic constraints. Furthermore, the deviation of the actual heading angle of 0.550 radians from the theoretical value was 0.027 radians, which is less than the calculated allowable tolerance of 0.0815 radians. Therefore, this point was determined to conform to the kinematic laws. After statistical analysis of all historical trajectory points, the density of points meeting the above constraints reached 0.95, higher than the preset convergence lower limit of 0.9. This judgment logic links the vehicle's Ackerman steering geometry constraints with spatial stability assessment, solving the problem of misjudgment when the vehicle is rejected due to excessive lateral displacement during detours or steering maneuvers. It also filters out lateral jump data caused solely by sensor errors, improving the accuracy of trajectory consistency verification.

[0029] After confirming that the spatial trajectories of both the vehicle and the person have effective convergence characteristics, it is necessary to transform the spatial coordinate data into a time-dimensional basis for collaborative scheduling in order to solve the technical problem of mismatch between the arrival time of the vehicle and the user at the pick-up point.

[0030] Specifically, by extracting the spatial mapping matrix of the high-precision map pathpoints, the current user's two-dimensional coordinates are substituted into linear interpolation to obtain the user's arrival time as 65 seconds. Simultaneously, combining the vehicle's current coordinates, actual speed, and remaining physical mileage of the planned route, the vehicle's arrival time is calculated to be 25 seconds. Subtracting the user's arrival time from the vehicle's arrival time yields a temporal phase difference with algebraic properties of +40 seconds, which physically indicates that the vehicle will arrive at the pick-up point 40 seconds earlier than the user. A preset 0-second standard time difference is obtained, and this 40-second positive phase difference is assessed to have exceeded the set 15-second parking trigger threshold. At the same time, the high-precision map semantics are retrieved, identifying that the vehicle is currently in a one-way street environment with a width of only 3.2 meters, where overtaking is not permitted. This computational logic accurately maps rigorously validated spatial displacement data to the temporal correspondence between people and vehicles, and, combined with lane topology characteristics, provides quantitative conditions for generating instructions that do not interfere with the overall traffic flow of the parking lot, preventing abnormal situations caused by directly issuing dispatch instructions.

[0031] In S4, to address the issue that using the existing proportional speed reduction in narrow one-way streets with a width of only 3.2 meters would force vehicles to travel long distances at low speeds, thereby obstructing traffic flow behind them, S4 implemented discontinuous spatiotemporal parking scheduling.

[0032] Specifically, the system blocks deceleration commands that could cause congestion. Following the planned path towards the vehicle pickup point, it locates a virtual anchor point 20 meters away that does not interfere with the main lane traffic flow. A command is issued to control the vehicle to travel to this anchor point at a normal cruising speed and apply brakes. During the vehicle's parking period, the system continuously tracks and calculates the user's remaining arrival time. When the user walks, reducing their remaining time to 15 seconds—a value exactly equal to the theoretical time required for the vehicle to restart from the virtual anchor point and accelerate to the final pickup point—a command to release the parking order is issued to the drive-by-wire chassis, allowing the vehicle to start moving again. This non-continuous parking mechanism solves the space occupation problem caused by simply reducing speed to compensate for time differences. It abandons the slow-driving strategy that reduces road utilization and instead uses map semantic intelligent matching to execute waiting actions on idle roadside resources. While ensuring the basic goal of timely encounters between people and vehicles, it releases public lane resources and improves the traffic efficiency of parking lot road networks in multi-vehicle concurrent scheduling scenarios.

[0033] In summary, the human-vehicle cooperative arrival prediction and scheduling method in the entire AVP vehicle retrieval stage firstly achieves time axis alignment of heterogeneous data from both ends through the spatiotemporal feature cluster construction mechanism in S1, transforming discrete coordinate points into continuous kinematic sequences and filtering out noise interference caused by single sampling delays. Furthermore, S2 introduces a dynamic noise baseline tolerance calculation model based on map semantics, converting building topology attributes such as load-bearing wall distance and metal coverage into physical attenuation parameters, giving adaptive tolerance to weak signal areas such as elevator lobbies, and improving the scheduling lag problem caused by existing fixed thresholds. Further, S3 changes the single one-dimensional distance comparison method, combining chassis front wheel angle and wheelbase parameters to construct a two-dimensional physically feasible domain corridor following Ackerman steering geometry, distinguishing between normal cornering and positioning drift, and calculating the temporal phase difference based on this, improving the reliability of the prediction. Finally, S4 proposes a discontinuous spatiotemporal parking mechanism for narrow one-way road environments. When a vehicle arrives early, it changes its low-speed driving strategy and instead uses map semantic matching to execute parking and waiting at roadside virtual anchor points. A wake-up command is issued at the node where the user's remaining arrival time equals the physical time. This design reduces the occupation of public lane resources in indoor parking lots while enabling people and vehicles to meet, and optimizes the global traffic efficiency in multi-vehicle scheduling scenarios.

[0034] like Figure 2 As shown, in one specific embodiment, S1 includes: S11, obtaining the time base of absolute time and clock aligning the time axis of the extracted coordinates. A network time protocol is used to ensure that the user terminal and the vehicle terminal sample within a unified time frame.

[0035] S12. Divide the time axis into continuous and non-overlapping prediction periods. At the current moment, determine the current prediction period and trace back to the historical time axis to extract the two immediately adjacent historical periods.

[0036] S13. Within the historical period, arrange user coordinates and vehicle coordinates in ascending order according to timestamps. Combine the coordinate points with the corresponding acquisition time to form a spatiotemporal feature cluster containing multi-dimensional motion information, providing a data foundation for subsequent extraction of acceleration features.

[0037] In this embodiment, it should be noted that in S11, during the automatic valet parking and vehicle retrieval phase, a spatiotemporal reference alignment mechanism is established to address the asynchronous sampling delay issue between the mobile terminal and the vehicle's positioning signals. A unified absolute time reference is obtained through a network time protocol, and hardware-level clock synchronization is performed between the mobile terminal's positioning sensor and the vehicle's autonomous driving positioning system. Based on this, a basic discrete time granularity is set, i.e., a preset sampling time interval of 200 milliseconds, and the entire time axis is divided into continuous and non-overlapping prediction periods. This computational logic incorporates the originally independent and frequency-discrete data streams from both ends into a unified time framework for management, resolving the initial data misalignment interference caused by single communication delays or clock deviations. This ensures the time-dimensional consistency of the physical states of both the driver and vehicle, providing a reliable time basis for subsequent kinematic feature calculations.

[0038] In S12, after the unified division of the basic time axis is completed, to avoid the susceptibility of single-point data to environmental interference leading to inaccurate evaluations, S12 performs the division of the prediction period and the tracing of historical data. At the current running moment, the current 200-millisecond sampling window is defined as the current prediction period. Tracing back along the historical time axis, the preceding 200-millisecond time window immediately adjacent to the current period is extracted as the first historical prediction period, and the 200-millisecond time window further back is extracted as the second historical prediction period. Through this calculation logic of extracting a total of 400-millisecond continuous time windows from the past, a data range with a certain time span can be obtained. This step solves the limitations of relying solely on instantaneous snapshots for decision-making, expanding the evaluation perspective from isolated points in time to continuous time periods, and providing a time range basis for subsequent analysis and positioning of the dynamic changing trends of the data.

[0039] In S13, after obtaining a clear historical time window, S13 is responsible for integrating the data within these time periods into a structured spatiotemporal feature cluster. Within the two historical prediction periods totaling 400 milliseconds, high-frequency location data from both ends is extracted, and user location data and vehicle location data are arranged and combined strictly according to the ascending order of timestamps. This data not only contains two-dimensional or three-dimensional spatial physical coordinate points, but also is tightly bound to the corresponding absolute timestamps, thus forming a multi-dimensional data set containing location and time correlations. This arrangement logic transforms the originally scattered and disordered coordinate information into a continuous trajectory sequence with kinematic coherence properties, solving the problem that messy data cannot be used for dynamic feature analysis. It enables discrete positioning points to possess the prerequisites for kinematic analysis such as velocity and acceleration, providing an ordered data structure to support the subsequent extraction of high-frequency jitter features of the trajectory.

[0040] like Figure 3As shown, in one specific embodiment, S2 includes: S21, extracting trajectory divergence values ​​as motion divergence. The user coordinate sequence within the historical period is traversed, and the difference vector between displacement vectors between two adjacent sampling periods is extracted. The magnitude of this difference vector is then accumulated. This accumulated magnitude value characterizes the degree of divergence of trajectory acceleration; in this embodiment, it is defined as motion divergence. The divergence of uniform motion tends to zero, while non-physical jumps caused by multipath effects will lead to an increase in this divergence.

[0041] S22. Calculate the noise tolerance value based on map semantics. Extract the grid attribute where the user's coordinates are located in the map semantics, obtain the basic fluctuation value, and combine it with the distance attenuation value of the vertical distance from the coordinates to the load-bearing wall, and the metal density value of the area above, to calculate the noise tolerance value. The calculation formula is as follows:

[0042] in, This represents the dynamically calculated noise tolerance value, in meters. This represents the basic fluctuation value, in meters. This represents the environmental compensation coefficient, with the unit being meters. Represents the natural constant, which is dimensionless; This indicates the distance from the user's coordinates to the nearest concrete load-bearing wall, in meters. This represents the reference attenuation constant, expressed in meters. This represents the environmental impact weighting coefficient, with the unit being meters; This represents the metal cable tray coverage density coefficient, ranging from 0 to 1, and is dimensionless. This calculation step allows for a reasonable tolerance in areas prone to signal interference.

[0043] Among them, the calculation formula involves , , as well as The model parameters were obtained through large-scale gridded radio frequency fingerprinting and regression analysis within specific types of underground buildings. In practice, testers used high-precision reference equipment to perform tens of thousands of high-frequency samples at grid points covering different metal cable tray densities and distances from load-bearing walls, recording the actual multipath positioning divergence at each point. Subsequently, using the terminal distance from the wall and metal cable tray density as independent variables and the measured divergence limit as the dependent variable, a nonlinear least squares method was used to jointly estimate the parameters of the attenuation exponential model and the linear interference model. For example, during the fitting process, it was found that the mean divergence in the unobstructed area was stable at 0.5 meters, thus determining the base fluctuation value to be 0.5 meters; while at a distance of 4.0 meters from the wall, the divergence increment attenuated to approximately the reciprocal of the natural logarithm base at the wall base, thus calculating the reference attenuation constant as 4.0 meters.

[0044] S23. Calculate the confidence mapping kernel to generate the vehicle's displacement constraints. When the motion divergence is determined to be less than or equal to the noise tolerance value, calculate the confidence mapping kernel using the following formula:

[0045] in, This represents the confidence mapping kernel, which is dimensionless. This represents a function that takes the minimum value. Indicates the noise tolerance value, in meters; This represents the trajectory divergence value, in meters. This represents the preset upper limit of the mapping, which is dimensionless. Then, the half-width of the preset standard displacement boundary is read and multiplied by the confidence mapping kernel to obtain the contraction compensation value. This contraction compensation value is then reduced inwards using the upper and lower limits of the standard displacement boundary to generate the displacement constraint boundary.

[0046] Among them, the preset upper limit value of mapping in the formula The calibration was determined through a combination of mechanical response limits and historical error tolerance tests on the drive-by-wire chassis. The specific calibration process involved: in a closed test track, different gradients of positioning noise were artificially injected into the vehicle, and the vehicle was instructed to perform correction maneuvers. The maximum instantaneous displacement reduction ratio that the chassis steering actuator could withstand, while avoiding mechanical resonance and dangerous side-stepping, was recorded. Analysis of thousands of extreme condition data revealed that when the reduction ratio exceeded 80%, the chassis response exhibited high-frequency oscillations; therefore, a physical critical value of 0.8 was taken as the upper limit of the mapping. For example, if the normalization ratio calculated under extreme disturbance reached 0.95, the system would be forcibly truncated and the value would be... This ensures that no matter how good the signal quality is at the user end, the generated shrinkage compensation value will not exceed the safety execution boundary of the vehicle's underlying drive-by-wire chassis.

[0047] In this embodiment, it should be noted that after constructing the spatiotemporal feature cluster in S21, S21 quantifies the reliability of user positioning data by extracting trajectory divergence values. The user coordinate sequence within the historical period is traversed, the difference vector between displacement vectors of two adjacent sampling periods is calculated, and the magnitude of this difference vector is accumulated to calculate the current trajectory divergence value as 0.8 meters. Specifically, the formula for calculating the extracted trajectory divergence value is as follows: ;in, This represents the trajectory divergence value, in meters. This represents the total number of coordinate sampling points extracted within the historical period; Indicates the first The displacement vector corresponding to each sampling period; This represents the magnitude of the difference vector between two adjacent sampling periods. The calculated current trajectory divergence value is 0.8 meters.

[0048] Since the difference in displacement vectors kinematically represents the change in acceleration (the second derivative of the trajectory), the calculation logic of this cumulative magnitude value can intuitively reflect the high-frequency divergence of the trajectory. This step defines it as motion divergence, solving the problem of difficulty in quantifying irregular jumps in mobile phone positioning caused by multipath effects in underground parking garages. Through this calculation, it is possible to distinguish between the normal trajectory of a user's smooth walking and non-physical jumps caused by signal interference, transforming positioning stability into a numerical indicator that can be used for subsequent threshold comparison.

[0049] In S22, during the automated valet parking vehicle retrieval phase, to transform the complex architectural physical topology of the underground parking lot into a quantifiable indicator that can be processed by a computer, S22 introduces a dynamic noise baseline tolerance calculation expression based on high-precision map semantics. .

[0050] The computational logic of this expression aims to mathematically model and reconstruct the multipath fading behavior of electromagnetic waves in a confined space; all its parameters have a clear physical context. In the formula, This represents the final allowable location noise tolerance (in meters), which is composed of the following three parts.

[0051] The first part is This represents the basic fluctuation tolerance constant under an ideal, open, and unobstructed environment, which establishes the baseline for the inherent thermal noise and measurement error of the sensor itself.

[0052] The second part is the nonlinear exponential term characterizing the influence of concrete obstacles. ,in It is the vertical physical distance (in meters) between the user and the nearest load-bearing wall. It is a reference constant (meter) used to eliminate the dimensions of exponential physical quantities and adjust the decay rate. Environmental compensation weighting coefficient for load-bearing walls (meters); using natural constants. The negative exponential form is because when radio signals approach large, high-density building entities, the multipath interference intensity caused by their reflection and scattering exhibits a nonlinear, non-proportional growth characteristic, with the effect becoming more severe the closer the distance. The exponential decay model can fit this physical law.

[0053] The third part is the linear term characterizing the effect of the top metal mesh. ,in is the dimensionless metal cable tray coverage density coefficient (value from 0 to 1). The corresponding weighting coefficient (meters) is given. Since the metal mainly exhibits large-area parallel reflection of the signal, its interference intensity has an approximately linear relationship with the coverage density.

[0054] Based on data from specific application scenarios, a baseline fluctuation value for open environments is set. Meters, the user is in the elevator lobby, a distance from the load-bearing wall meters, reference attenuation constant Meters, environmental compensation coefficient for load-bearing walls Meter, metal density Weighting coefficient of environmental impact of metals Meters; substituting into the expression, we can calculate... This calculation process addresses the technical problem that existing automated valet parking systems, which use a single global fixed threshold to filter noise, make it difficult for users in weak signal areas such as elevator lobbies to pass stability checks. By upgrading the static threshold to a dynamic tolerance calculated in real-time based on the semantics of a high-precision map grid, reasonable spatial positioning tolerance is provided at the physical level in harsh environments. This avoids deadlocks in the collaborative scheduling logic caused by objective environmental factors and improves the adaptability of the two-end mutual trust mechanism in different building topology environments.

[0055] In S23, after obtaining the tolerance, the system is responsible for generating dynamically scaled vehicle displacement constraint bounds. Specifically, after determining the physical tolerance of the current environment, it is necessary to map the data quality from the user end to the control constraint requirements for the vehicle end. The expression for calculating the dynamic confidence mapping kernel in S23 is as follows: .

[0056] The calculation of this expression is based on the margin assessment theory in feedback control, used to quantify the reliability of the current user positioning signal and generate subsequent scaling parameters. In the formula, The actual motion divergence of the user (i.e., trajectory divergence value, in meters) is extracted by the second derivative of the trajectory. This represents the difference between the maximum noise margin allowed by the current environment and the actual trajectory divergence generated by the user. This difference physically represents the "residual safety margin" of the current signal. Divide this difference by... Normalization eliminates the differences in specific physical dimensions, generating a dimensionless relative proportion between 0 and 1, reflecting the degree to which the user signal quality is better than the average level of the environment.

[0057] To ensure that the vehicle's underlying drive-by-wire chassis does not exceed the physical limits of mechanical response due to excessively narrow interval contraction when applying displacement constraints, the expression is nested with... The function (minimum function) is compared with a preset upper limit threshold. Compare them.

[0058] Based on the scene data, it is known Meters, User Trajectory Divergence Meters, the set upper limit Substituting into the formula, the normalized ratio is calculated as follows: Since 0.4689 is less than the upper limit of 0.8, after... The function ultimately outputs a confidence mapping kernel after processing. After obtaining the dimensionless mapping kernel, it is applied to the generation of vehicle physical displacement constraint boundaries. It is assumed that the pre-calibrated standard displacement boundary of the vehicle within a single prediction cycle is... Meters, its interval half-width is Meters. The shrinkage compensation value is calculated by multiplying the mapping kernel by the half-width. The standard displacement boundary is calculated by adding 0.187 meters to the lower limit of 0.2 meters to obtain 0.387 meters, and subtracting 0.187 meters from the upper limit of 1.0 meters to obtain 0.813 meters. This results in the adaptively reduced radial displacement constraint boundary. rice.

[0059] This computational logic solves the technical problem of fragmented data evaluation at both ends in human-vehicle collaboration. It establishes a bridge, enabling the signal quality of high-frequency sampling at the user end to be directly converted into tolerance requirements for vehicle displacement fluctuations. The more stable the user end signal (i.e., the smaller the divergence), the larger the mapping kernel, and the tighter the allowable displacement range at the vehicle end is compressed, thereby forcing the filtering out of minute positioning drifts at the vehicle end. This closed-loop numerical mapping mechanism improves the synchronization requirements and control accuracy of overall collaborative scheduling.

[0060] It should also be noted that the pre-calibrated standard displacement boundary is not subjectively set, but rather obtained based on the statistical extreme values ​​of the target vehicle's dynamic bench tests and historical continuous sampling data. Specifically, within a baseline prediction period (e.g., 200 milliseconds), the true radial displacement values ​​of adjacent points are extracted under all operating conditions, including idle start-up, full-throttle acceleration, and emergency braking. Since the system speed limit is 5 m / s, the theoretical maximum physical displacement within 200 milliseconds is 1.0 meter, while the normal minimum drift background of the sensor is approximately 0.2 meters. By performing Kalman filtering smoothing and extreme value removal on the collected 100,000 normal displacement data points, a statistical boundary containing 99% of normal physical motion behavior was determined, with a lower limit of 0.2 meters and an upper limit of 1.0 meter. Using this data as a benchmark, the standard displacement boundary is set to 0.2 meters to 1.0 meters, ensuring that the initial, uncontracted displacement limit fully conforms to the actual dynamic reachable envelope of the vehicle model within the prediction period.

[0061] like Figure 4 As shown, in one specific implementation, S3 includes: S31, constructing a two-dimensional feasible region. The steering angle and wheelbase values ​​fed back from the chassis are read, and the theoretical heading angle of the vehicle within the prediction period is calculated based on steering geometry principles. Combining the displacement constraint boundary and the theoretical heading angle, a two-dimensional feasible region with a fan-shaped envelope is constructed. For each vehicle historical point in the spatiotemporal feature cluster, a determination is made, and the determination value... The logic is as follows:

[0062] in, Indicates the first The logical determination value of each point satisfying the constraint; This represents the minimum value of the displacement constraint boundary, in meters. This represents the actual displacement between adjacent points, in meters. This represents the maximum value of the displacement constraint boundary, in meters. This indicates the actual heading angle, expressed in radians. This represents the theoretical heading angle, expressed in radians. This represents the confidence mapping kernel, which is dimensionless. Indicates yaw tolerance, in radians.

[0063] The standard yaw tolerance value is calculated based on the control dead zone of the target vehicle's chassis steering actuator and the Ackermann geometric steady-state deviation, obtained through large-scale real-vehicle field calibration experiments. In a flat, dry test track, the test vehicle performed dozens of standard radius turns and straight-line driving maneuvers at a typical cruising speed of 2.5 m / s during the automated valet parking phase. A high-precision integrated navigation system simultaneously recorded the theoretical chassis heading angle and the vehicle's actual trajectory heading angle. Statistical analysis of approximately 20,000 sets of normal driving heading deviation data revealed that, due to the combined effects of tire lateral stiffness, subtle changes in road friction coefficient, and mechanical assembly clearances, the inherent steady-state deviation averaged approximately 0.08 radians, with the maximum transient deviation peaking at around 0.15 radians. To cover 99% of normal mechanical and physical lateral deviation fluctuations, this peak value was compensated and mapped with a 15% safety tolerance, ultimately fixing the standard yaw tolerance offline to 0.174 radians.

[0064] S32. Calculate the density value of convergence for two-dimensional spatial positioning. (This applies to all points.) The results are summed and divided by the total number of verification points. If the density value is greater than or equal to the preset lower convergence limit, the vehicle-side positioning signal is determined to match the steering trend.

[0065] S33. Calculate the user's arrival time and the vehicle's arrival time. Input the user's coordinates into the spatial mapping matrix of the map waypoints, and obtain the user's arrival time through interpolation. At the same time, combine the vehicle's coordinates, driving speed, and remaining mileage to calculate the vehicle's arrival time.

[0066] S34. Obtain the timing phase difference. Subtract the vehicle's arrival time from the user's arrival time to obtain the timing phase difference, which has an algebraic sign. A positive value indicates that the user arrives later than the vehicle.

[0067] In this embodiment, it should be noted that in S31, in order to overcome the problem that one-dimensional distance comparison can easily misjudge the normal lateral displacement of the vehicle when cornering as positioning drift, S31 constructs a two-dimensional feasible domain in combination with the vehicle chassis attributes.

[0068] Specifically, to verify whether the vehicle's historical trajectory truly reflects its motion state, the expression for determining whether a vehicle's historical point falls within a two-dimensional physically feasible region corridor is set as follows: .

[0069] The operation logic of this piecewise function abandons the existing one-dimensional scalar distance comparison and introduces the Ackermann steering geometry model in vehicle dynamics to construct a dual physical envelope in two-dimensional space that includes longitudinal displacement and lateral yaw.

[0070] The first condition of the expression This is a radial length constraint, used to verify the first... The actual physical displacement between adjacent trajectory points Does the (meter) fall within the dynamic contraction range calculated in the previous step? This eliminates the possibility of a long-distance spatial jump caused by the loss of positioning signals.

[0071] The second condition of the expression It belongs to the lateral angle constraint, among which The heading angle (in radians) of the actual displacement vector. To read the theoretical heading angle (in radians) calculated in real time from the chassis front wheel steering angle and vehicle wheelbase, the absolute difference between the actual and theoretical heading angles is calculated and compared to see if it is less than that obtained through a confidence mapping kernel. Dynamically scaled standard yaw tolerance This can verify whether the vehicle's displacement direction is consistent with the actual steering trend of its mechanical chassis. The judgment result is only valid when both radial and lateral conditions are simultaneously satisfied (logical "AND" relationship). Only output 1 if it is true, otherwise output 0.

[0072] Based on the scene data, the generated radial constraint bounds are known to be... Meters, theoretical heading angle radians, standard yaw tolerance radians, mapping kernel For an extracted actual radial displacement meters, actual heading angle The trajectory point of the arc, first verify the radial condition, 0.6 meters is in Within the specified range, it holds true; secondly, verify the angle conditions, and the deviation between reality and theory. Radius, the upper limit of the allowable angle tolerance is calculated as follows. Since 0.027 radians is less than 0.0815 radians, the condition still holds; therefore, the binary logic decision value at this point is... The value is assigned to 1. This complex logical calculation process solves the technical problem that existing technologies easily misjudge normal obstacle avoidance or lateral displacement during turning as sensor positioning drift when evaluating vehicle trajectories. By cross-validating the mechanical feedback data (front wheel steering angle) from the chassis base layer with the visual or radar positioning data (heading angle) from the top-level software, a fan-shaped dynamic tolerance corridor is constructed, effectively filtering out data jumps caused solely by sensor errors and improving the rationality and physical authenticity of multi-source heterogeneous data fusion verification.

[0073] In S32, after completing the logical determination of the two-dimensional spatial state of each independent trajectory point, it is responsible for calculating the convergence density value of the entire historical sequence in two-dimensional spatial positioning. The logical determination values ​​of all vehicle historical points in the feature cluster are accumulated and divided by the total number of points participating in the verification. For example, if 19 out of 20 historical points in the feature cluster satisfy the sector corridor constraint condition, the calculated density value is 0.95. This value is compared with a preset convergence lower limit of 0.9; if it is greater than or equal to this lower limit, the positioning signal is determined to match the physical steering trend. This calculation logic, which uses the overall density proportion instead of the absolute judgment of a single point, solves the vulnerability problem of the overall sequence being rejected due to individual instantaneous positioning jumps, filters out the very few lateral abnormal jump data caused by occasional sensor errors, and improves the overall accuracy and fault tolerance of the trajectory consistency comprehensive verification.

[0074] It should be noted that the aforementioned preset lower convergence limit is obtained offline through statistical distribution model calibration based on a limited number of historical parking data. Specifically, before system deployment, no fewer than 5,000 sets of historical vehicle trajectory data, including normal driving and positioning drift, are collected. These data are then substituted into the two-dimensional feasible region logical judgment expression to calculate the true convergence density value for each set. The set of density values ​​under normal driving conditions is fitted using a normal distribution function, and the lower bound corresponding to its one-sided 95% confidence interval is taken as the preset lower convergence limit. For example, in the measured dataset of an underground parking garage, the mean convergence density of normal cornering and straight-line samples is concentrated at 0.96, with a standard deviation of 0.03. Through confidence interval calculation and combined with engineering safety redundancy, the preset lower convergence limit is finally calibrated offline to 0.90, thereby ensuring that normal physical displacement points are not mistakenly filtered out while occasional jumps are filtered.

[0075] In S33, after confirming that the dual-end spatial trajectories have effective convergence characteristics, it is responsible for transforming spatial coordinate data into a time-dimensional basis for collaborative scheduling. The current user's two-dimensional coordinates are substituted into a spatial mapping matrix based on high-precision map pathpoint calibration. This spatial mapping matrix records the baseline walking time from each road network node to the vehicle pickup point. By projecting the user's coordinates onto the nearest road network node and performing linear interpolation based on the distance ratio between the user's current coordinates and adjacent nodes, the user's arrival time is calculated to be 65 seconds. Simultaneously, combining the vehicle's current coordinates, actual driving speed, and remaining physical mileage of the planned route, the vehicle's arrival time is calculated to be 25 seconds. This linear interpolation and mileage calculation logic smoothly converts non-linear spatial distances into the time cost required for travel. This step solves the problem of the lack of a unified measurement scale for the spatial location of users and vehicles, accurately converting discrete three-dimensional spatial physical coordinates into a unified one-dimensional time expectation metric, providing a comparable quantitative time parameter basis for subsequent direct evaluation of the encounter timing between people and vehicles at the pickup point.

[0076] In S34, after obtaining the expected arrival times of both ends, a subtraction operation is performed to obtain the timing phase difference used to guide scheduling. Subtracting the vehicle arrival time of 25 seconds from the calculated user arrival time of 65 seconds yields a timing phase difference with an algebraic sign of +40 seconds. This directional subtraction calculation logic not only quantifies the time gap between the arrival times of both parties, but its positive or negative sign also directly represents the order of the person and vehicle on the timeline in a physical sense. A positive value indicates that the vehicle will arrive at the pick-up point earlier than the user, while a negative value means that the user needs to wait for the vehicle. This step solves the problem of directly determining the direction of deviation from the current collaborative progress, providing a key quantitative indicator that includes both the magnitude and direction of the deviation, serving as a direct input variable to trigger subsequent acceleration, deceleration, or parking strategies.

[0077] like Figure 5 As shown, in one specific embodiment, S4 includes: S41, evaluating the time-series phase difference. A standard time difference is obtained, and the deviation between the time-series phase difference and this value is calculated. Simultaneously, lane width information is extracted from the map semantics.

[0078] The standard time difference is set to 0. This parameter is a baseline zero directly defined through mathematical modeling of an ideal human-vehicle collaborative intersection scenario, representing the optimal expectation of seamless connection. In the logical architecture of the collaborative scheduling system, the complete consistency between the user's arrival and the vehicle's arrival is defined as the absolute steady state pursued by the system. By analyzing user experience feedback data from tens of thousands of historical vehicle retrieval events, the developers found that when the person and the vehicle arrive at the retrieval point exactly simultaneously (with a deviation within plus or minus 2 seconds, uniformly smoothed to 0 seconds in the mathematical model), the user's waiting anxiety is lowest and the time spent ineffectively occupying the lane is zero. For example, when it is calculated that the user will arrive in 30 seconds and the vehicle will also arrive in 30 seconds, the physical expected deviation between the two is 0. The system determines that no additional speed adjustment or parking energy consumption commands are required, and the vehicle can proceed smoothly as originally planned. Therefore, through historical utility evaluation, the system establishes 0 seconds as the preset standard time difference, serving as the absolute origin for all collaborative command correction calculations.

[0079] S42. Execute parking dispatch. When the timing phase difference is positive and greater than the preset parking threshold, and the map semantics indicate that the vehicle is currently in a one-way lane, block the proportional speed reduction command. Retrieve the coordinates of an area along the planned path towards the vehicle pick-up point that does not interfere with the main lane traffic flow as a virtual anchor point. Issue a parking command to control the vehicle to drive to the virtual anchor point and brake.

[0080] The parking threshold was determined by analyzing historical pedestrian-vehicle intersection queuing congestion data and combining it with micro-simulation of parking lot traffic flow. During the development phase, traffic flow data from multiple commercial underground parking garages during peak hours were collected to establish a one-way vehicle following queuing model, simulating the impact of different early arrival time differences on the traffic flow efficiency behind. Simulation results show that when an automated parking vehicle arrives at the retrieval point early and waits on the main road for more than 15 seconds, the probability of chain congestion behind it increases exponentially; while short stops of less than 15 seconds can usually be absorbed by the natural wave gaps in the traffic flow. Based on this simulation extreme value and engineering experience, the preset parking threshold was strictly calibrated to 15 seconds. For example, when the phase difference is positive 12 seconds, the system will issue a proportional deceleration command to make it coast slowly; while when the phase difference reaches positive 40 seconds, the 15-second threshold is triggered, and the system immediately switches to a strategy of finding virtual anchor points to execute discontinuous parking, effectively preventing road network deadlock.

[0081] S43. Perform restart or continuous compensation. During vehicle parking, continuously calculate the user's remaining arrival time. When the change in timing phase difference meets the condition that the remaining arrival time equals the physical time required for the vehicle to start from the virtual anchor point and travel to the pick-up point, issue a release parking command to start the vehicle. If the timing phase difference is less than or equal to the parking threshold, or the vehicle is in a wide-open environment, generate a proportional coefficient based on the timing phase difference and issue a speed adjustment command.

[0082] In this embodiment, it should be noted that in S41, to ensure that subsequent dispatch commands conform to the current physical environment conditions, S41 comprehensively evaluates the magnitude of the timing phase difference and the lane topology attributes. The acquired positive 40-second timing phase difference is compared with the preset 0-second standard time difference, and it is determined that the deviation has exceeded the set 15-second parking trigger threshold. Simultaneously, high-precision map semantics are retrieved to extract that the vehicle is currently in a one-way street environment with a width of only 3.2 meters, where overtaking is not permitted. This evaluation logic, which superimposes the magnitude of the time-dimensional deviation with the lane capacity in the spatial dimension, solves the problem of environmental conflicts caused by issuing commands solely based on time differences. It prevents potential congestion risks caused by unreasonable deceleration actions in confined spaces and provides sufficient boundary conditions to support correct decisions between regular proportional speed adjustment and discontinuous parking.

[0083] In S42, once it's determined that a vehicle arrived excessively early and its environment was constrained, S42 implemented discontinuous parking scheduling. This blocked the conventional proportional speed reduction command, retrieved a virtual anchor point 20 meters away along the planned path towards the pick-up point that wouldn't interfere with the main lane's traffic flow, and issued a command to control the vehicle to travel to that anchor point at a normal cruising speed and brake to a stop. This computational logic of abandoning speed reduction and instead finding available space to park solves the problem of vehicles traveling long distances at low speeds in a 3.2-meter-wide one-way street, obstructing normal traffic flow behind them. Through discontinuous, spatially skipping stops, it changes the existing practice of simply relying on speed reduction to consume waiting time. Without reducing the overall lane flow efficiency, it effectively absorbs the 40-second time difference caused by vehicles arriving early, avoiding prolonged occupation of public lane resources.

[0084] In S43, to ensure that vehicles can resume driving in a timely manner after parking to complete the rendezvous, S43 performs precise wake-up and continuous compensation operations. While the vehicle is parked at the anchor point, the remaining arrival time of the user is continuously tracked and calculated. When the user walks, reducing their remaining time to 15 seconds—a value exactly equal to the theoretical time required for the vehicle to restart from the virtual anchor point and accelerate to the final pick-up point—a command to release the parking stop is sent to the drive-by-wire chassis, causing the vehicle to start moving. This comparative calculation logic based on equal time points solves the problem of difficulty in accurately determining the vehicle restart timing. For situations with small deviations or in wide-open environments, speed adjustment is performed according to a proportional coefficient calculated based on the deviation. This step, while ensuring the basic goal of timely encounters between people and vehicles, completes a closed-loop connection from parking and waiting to resumption of coordinated scheduling, improving the overall traffic network utilization efficiency in multi-vehicle concurrent scheduling scenarios in underground parking lots.

[0085] This invention also provides a human-vehicle cooperative arrival prediction and scheduling system during the AVP vehicle retrieval phase, the system comprising: The acquisition module is used to obtain the prediction period at the current moment, extract user coordinates and vehicle coordinates within the historical period, and construct a spatiotemporal feature cluster. The first processing module is used to extract trajectory divergence values ​​as motion divergence based on spatiotemporal feature clusters, calculate noise tolerance values ​​based on map semantics, and calculate confidence mapping kernels to generate vehicle displacement constraint boundaries when the motion divergence is less than or equal to the noise tolerance values. The second processing module is used to combine the vehicle's steering geometry and displacement constraints to construct a two-dimensional feasible region, calculate the density value of the vehicle's historical points in the spatiotemporal feature cluster falling into the two-dimensional feasible region, and if the density value is greater than or equal to the preset lower limit of convergence, then the user coordinates and vehicle coordinates at the current moment are fused to obtain the temporal phase difference. The control module is used to match virtual anchor points in the map semantics based on the timing phase difference and the preset standard time difference, and to issue parking or speed adjustment commands.

[0086] In one specific implementation, the acquisition module is further configured to: acquire the time reference of absolute time, clock-align the time axis of the extracted coordinates; divide the time axis into continuous and non-overlapping prediction periods, and within the historical period, arrange the user coordinates and vehicle coordinates in ascending order of timestamps to form a spatiotemporal feature cluster containing coordinates and timestamps.

[0087] To enable those skilled in the art to fully understand and implement the technical solutions described in this specification, the following section, using a specific application scenario, provides a detailed deduction and data analysis of the entire process of the human-vehicle collaborative arrival prediction and scheduling method and system under the AVP vehicle retrieval stage.

[0088] Imagine a typical underground shopping mall parking lot environment. During Phase S1, the system aligns both clocks via Network Time Protocol (NTP) and sets the basic discrete prediction period to 200ms. Assume the user has just exited the elevator lobby on the second basement level, finding themselves in a complex area of ​​load-bearing walls and metal pipes. Meanwhile, the vehicle, having received a retrieval command, is automatically cruising at 2.5m / s on a single lane 150m from the retrieval point. The system traces back two prediction periods (400ms) to the historical timeline, extracting high-frequency coordinate data from both the user and vehicle terminals. These discrete physical coordinates, each with an absolute timestamp, are packaged to construct a spatiotemporal feature cluster for the current logical decision.

[0089] Upon entering phase S2, the system first executes sub-step S21, which calculates the current trajectory divergence value (i.e., motion divergence) by accumulating the magnitude of the second-order difference vector of the user's historical coordinates. The distance is 0.8m. Due to the severe signal multipath effect in the elevator lobby, the system calls the high-precision map semantic grid in sub-step S22 to extract the current user's vertical distance from the load-bearing wall. m, density coefficient of the metal cable tray above Set the basic fluctuation value for open environment. m, environmental compensation coefficient for load-bearing wall m, reference attenuation constant m, weighting coefficient for the environmental impact of metals m. Substituting these physical parameters into the noise tolerance formula based on map semantics: ; calculation yielded m. This indicates that the system, by sensing the building topology, provides a legitimate signal fluctuation tolerance of 1.5065m at the physical level for this harsh environment.

[0090] Subsequently, in sub-step S23, the system comparison revealed that the motion divergence of 0.8m was less than the calculated noise tolerance value of 1.5065m, determining that the current user trajectory had not undergone a non-physical jump. Next, a preset mapping upper limit value was set. Substituting into the confidence mapping kernel formula: Calculations show that the difference accounts for approximately 0.4689%, therefore, we take... Assume the preset standard displacement boundary is... m, with a half-width of 0.4m. Using this mapping kernel... Shrink it, and the shrinkage compensation value is m. The final generated lower bound of the vehicle displacement constraint. m, upper limit m. This indicates that the system adaptively tightens its tolerance for vehicle displacement fluctuations based on the signal quality at the user end.

[0091] In phase S3, sub-step S31 is executed to construct the two-dimensional feasible region. The chassis provides feedback on the theoretical heading angle calculated from the current front wheel steering angle and wheelbase. rad (approximately 30 degrees), the set yaw tolerance rad (approximately 10 degrees). At this point, the first [unit] in the spatiotemporal feature cluster is extracted. The actual radial displacement at each vehicle historical point m, actual heading angle rad. Substituting into the two-dimensional feasible region logical decision expression: Obviously, m strictly falls into Within the m interval, and the angle deviation rad is much smaller than the dynamic tolerance allowed. rad. Therefore, the judgment value at this point is... This means that the historical point fully complies with the dual physical constraints of Ackermann steering geometry and radial contraction.

[0092] Execute sub-steps S32 to S34, assuming that 19 out of 20 historical points in the feature cluster are... The density value is 0.95, which is greater than the preset convergence lower limit of 0.9, indicating that double-blind mutual trust is established. The system then calculates the user's arrival time as 65s using high-precision map interpolation and the vehicle's arrival time as 25s based on the remaining mileage. The difference between the two (65s minus 25s) is calculated to yield the timing phase difference. This positive value physically indicates that if a vehicle arrives 40 seconds earlier than the user, it will cause congestion at the pick-up point if no intervention is taken.

[0093] Finally, in stage S4, spatial parking scheduling is executed. Stage S41 evaluation revealed a 40-second temporal phase difference significantly greater than the preset parking threshold (e.g., 15 seconds), and map semantic extraction indicated the vehicle was currently in a narrow, one-way street (only 3.2 meters wide) where overtaking was not permitted. In stage S42, the system directly blocked the linear deceleration command (because coasting at extremely low speeds would severely obstruct traffic flow behind), instead retrieving an abandoned widened avoidance zone 20 meters ahead as a virtual anchor point. After the command was issued, the vehicle entered the virtual anchor point at normal speed and braked to a stop. In stage S43, the system silently timed the event until the remaining time for the user to reach the destination dropped to 15 seconds. This time, precisely equal to the theoretical time (15 seconds) required for the vehicle to restart from the virtual anchor point and travel to the final pick-up point, the system accurately issued a release command. The vehicle restarted, ultimately achieving a high-precision coordinated intersection in both physical space and time, validating the effectiveness and reliability of the scheduling logic in extremely confined spaces.

[0094] The preferred embodiments of the present disclosure have been described in detail above with reference to the accompanying drawings. However, the present disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present disclosure, various simple modifications can be made to the technical solutions of the present disclosure, and these simple modifications all fall within the protection scope of the present disclosure.

[0095] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0096] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for predicting and scheduling human-vehicle cooperative arrival during the AVP vehicle retrieval phase, characterized in that, The methods include: Obtain the prediction period at the current moment, extract user coordinates and vehicle coordinates from the historical periods, and construct a spatiotemporal feature cluster; The trajectory divergence value is extracted as the motion divergence based on the spatiotemporal feature cluster. A noise tolerance value based on map semantics is calculated. When the motion divergence is less than or equal to the noise tolerance value, a confidence mapping kernel is calculated to generate the vehicle's displacement constraint bound. By combining the vehicle's steering geometry with the displacement constraint, a two-dimensional feasible region is constructed. The density value of the vehicle's historical points in the spatiotemporal feature cluster falling into the two-dimensional feasible region is calculated. If the density value is greater than or equal to a preset lower convergence limit, the user coordinates and vehicle coordinates at the current moment are fused to obtain the temporal phase difference. Based on the time phase difference and the preset standard time difference, the virtual anchor points in the map semantics are matched, and a stop command or speed adjustment command is issued.

2. The method for predicting and scheduling human-vehicle cooperative arrival during the AVP vehicle retrieval phase according to claim 1, characterized in that, The construction of the spatiotemporal feature cluster includes: Obtain the absolute time reference and clock-align the time axis of the extracted coordinates; The time axis is divided into continuous and non-overlapping prediction periods. Within the historical period, user coordinates and vehicle coordinates are arranged in ascending order of timestamps to form the spatiotemporal feature cluster containing coordinates and timestamps.

3. The method for predicting and scheduling human-vehicle cooperative arrival during the AVP vehicle retrieval phase according to claim 1, characterized in that, The extraction of trajectory divergence values ​​as motion divergence includes: Extract the difference vector of displacement vectors between two adjacent sampling periods; The magnitude of the difference vector is accumulated, and the accumulated magnitude value is used to characterize the acceleration divergence, and is used as the motion divergence.

4. The method for predicting and scheduling human-vehicle cooperative arrival during the AVP vehicle retrieval phase according to claim 1, characterized in that, The calculation of the noise tolerance value based on map semantics includes: Extract the grid attribute where the user's coordinates are located in the map semantics; Obtain the basic fluctuation value, calculate the distance compensation term with exponential decay based on the vertical distance of the user coordinates from the load-bearing wall, and calculate the linear interference term in combination with the metal density value of the upper region. Summate the basic fluctuation value, the distance compensation term and the linear interference term to obtain the noise tolerance value.

5. The method for predicting and scheduling human-vehicle cooperative arrival during the AVP vehicle retrieval phase according to claim 1, characterized in that, The calculation of the confidence mapping kernel to generate the vehicle's displacement constraint bounds includes: The confidence mapping kernel is obtained by using the ratio of the difference between the noise tolerance value and the motion divergence to the noise tolerance value, and in combination with a preset mapping upper limit value. Read the half-width of the preset standard displacement boundary, multiply it by the confidence mapping kernel to obtain the shrinkage compensation value, and use the upper and lower limits of the standard displacement boundary to reduce the shrinkage compensation value inward to generate the displacement constraint boundary with physical constraints.

6. The method for predicting and scheduling human-vehicle cooperative arrival during the AVP vehicle retrieval phase according to claim 1, characterized in that, The construction of the two-dimensional feasible region and the calculation of the density value of vehicle historical points in the spatiotemporal feature cluster falling into the two-dimensional feasible region include: Read the steering angle and wheelbase values ​​fed back by the chassis, and calculate the theoretical heading angle of the vehicle within the prediction period based on the steering geometry principle; By combining the displacement constraint boundary and the theoretical heading angle, the two-dimensional feasible region with a fan-shaped envelope is constructed; For each vehicle history point in the spatiotemporal feature cluster, if the actual displacement between adjacent points is within the displacement constraint and the deviation between the actual heading angle and the theoretical heading angle does not exceed the yaw tolerance scaled by the confidence mapping kernel, the vehicle history point is determined to fall into the two-dimensional feasible region. The density value is obtained by dividing the number of vehicle history points falling into the two-dimensional feasible region by the total number of verification points.

7. The method for predicting and scheduling human-vehicle cooperative arrival during the AVP vehicle retrieval phase according to claim 1, characterized in that, The acquisition of the timing phase difference includes: The user coordinates are input into a spatial mapping matrix based on map path point calibration, and the user's arrival time is obtained through interpolation. Based on the vehicle coordinates, driving speed, and remaining mileage, calculate the vehicle's arrival time; Subtracting the arrival time of the vehicle from the arrival time of the user yields the timing phase difference with algebraic sign properties.

8. The method for predicting and scheduling human-vehicle cooperative arrival during the AVP vehicle retrieval phase according to claim 1, characterized in that, The process of matching virtual anchor points in the map semantics and issuing pause or speed adjustment commands includes: When the timing phase difference is positive and greater than the preset parking threshold, and the map semantics indicate that the vehicle is currently in a one-way lane, the proportional deceleration command is blocked. The coordinates of an area that does not interfere with the main lane traffic flow are retrieved along the vehicle's planned path toward the vehicle pick-up point and used as the virtual anchor point. The parking command is then issued to control the vehicle to drive to the virtual anchor point and brake. During the vehicle's parking period, the remaining arrival time of the user is continuously calculated. When the remaining arrival time is equal to the physical time required for the vehicle to start from the virtual anchor point and accelerate to the pick-up point, a parking release command is issued to start the vehicle. If the timing phase difference is less than or equal to the parking threshold, or if the vehicle is in a wide environment, a proportional coefficient is generated based on the timing phase difference, and the speed control command is issued.

9. A human-vehicle cooperative arrival prediction and scheduling system during the AVP vehicle retrieval phase, characterized in that, The system includes: The acquisition module is used to obtain the prediction period at the current moment, extract user coordinates and vehicle coordinates within the historical period, and construct a spatiotemporal feature cluster. The first processing module is used to extract trajectory divergence values ​​as motion divergence based on the spatiotemporal feature clusters, calculate noise tolerance values ​​based on map semantics, and calculate confidence mapping kernels to generate vehicle displacement constraint boundaries when the motion divergence is less than or equal to the noise tolerance values. The second processing module is used to combine the vehicle's steering geometry with the displacement constraint to construct a two-dimensional feasible region, calculate the density value of the vehicle's historical points in the spatiotemporal feature cluster falling into the two-dimensional feasible region, and if the density value is greater than or equal to a preset lower convergence limit, then the user coordinates and vehicle coordinates at the current moment are fused to obtain the temporal phase difference. The control module is used to match the virtual anchor points in the map semantics according to the time phase difference and the preset standard time difference, and issue a stop command or speed adjustment command.

10. The AVP vehicle retrieval stage human-vehicle cooperative arrival prediction and scheduling system according to claim 9, characterized in that, The acquisition module is also used for: Obtain the absolute time reference and clock-align the time axis of the extracted coordinates; The time axis is divided into continuous and non-overlapping prediction periods. Within the historical period, user coordinates and vehicle coordinates are arranged in ascending order of timestamps to form the spatiotemporal feature cluster containing coordinates and timestamps.

Images