A multi-vehicle cooperative tracking system and a communication and obstacle avoidance method using the system.

Abstract

This invention discloses a multi-vehicle cooperative tracking system and a communication and obstacle avoidance method using this system. The method, executed by the multi-vehicle cooperative tracking system, includes: assigning a unique ID to each vehicle and planning an initial global path for each vehicle; each vehicle periodically broadcasting its own status data packets during operation; each vehicle receiving and processing status data packets from other vehicles, comparing its own and other vehicles' pre-defined grid sets, determining whether there are intersecting grids and calculating the estimated conflict time, and determining the conflict level; each vehicle autonomously negotiating to determine its passage priority; vehicles with lower priority autonomously performing corresponding avoidance actions based on the negotiation results, updating their own status and broadcasting it, while the central control unit performs system-level path load balancing adjustments. This invention significantly outperforms existing technologies in terms of safety, efficiency, and robustness, and can fully meet the needs of multi-vehicle cooperative operations in high-density, dynamic environments in warehousing and logistics workshops.

Description

Technical Field

[0001] This invention relates to the field of automatic control and intelligent transportation technology, specifically to a multi-vehicle cooperative tracking system and a communication and obstacle avoidance method using the system. It is particularly suitable for communication, tracking and active obstacle avoidance of multi-vehicle cooperative operations in high-density, dynamic environments such as warehousing and logistics and industrial manufacturing. It is compatible with the synchronous cooperative operation of 5-50 intelligent vehicles and supports typical application scenarios such as complex intersections and multi-path intersections. Background Technology

[0002] With the development of automation technology, multi-agent collaborative operations are increasingly being used in warehousing and logistics, intelligent inspection, and industrial manufacturing. Among them, multi-vehicle collaborative tracking systems are one of the core technologies for achieving efficient and flexible operations, and their performance directly affects operational efficiency, equipment safety, and system throughput.

[0003] Currently, existing multi-vehicle cooperative tracking systems mainly have the following problems:

[0004] 1. Lack of effective collaboration easily leads to "information silos": Most systems involve multiple vehicles independently performing the same line-following task, with no communication or coordination mechanism between the vehicles. When paths intersect or vehicles need to meet, deadlocks (i.e., two or more vehicles waiting for each other and unable to move) or collisions can easily occur, resulting in extremely low system efficiency and reliability.

[0005] 2. Passive and inefficient obstacle avoidance strategies: Common obstacle avoidance methods rely on onboard sensors (such as ultrasonic and infrared sensors) to brake or detour only when a nearby obstacle (i.e., another vehicle) is detected. This approach has a lag (response time is typically 0.8-1.5 seconds), leading to frequent starts and stops in the convoy, disrupted operation, and more than 300% more vehicle starts and stops compared to a collaborative system. Furthermore, it is prone to triggering chain reactions in complex paths, potentially causing the entire system to malfunction.

[0006] 3. Inefficient resource allocation: When there are multiple paths or multiple task points, traditional methods cannot dynamically allocate the optimal path to the vehicle, which may lead to some paths being congested (occupancy rate exceeding 80%), while other paths are idle (occupancy rate below 30%). As a result, the system throughput cannot reach the optimal level, and the resource utilization rate is only 50%-60%.

[0007] While existing multi-vehicle cooperative path planning and obstacle avoidance technologies have achieved conflict coordination to some extent, they still suffer from the following key bottlenecks:

[0008] 1. Passive and Lagging Conflict Prediction: As described in reference document D1 (CN115079702A), obstacle avoidance relies on real-time detection of whether the distance between the vehicle and the obstacle is less than a safety threshold, which is a "perception-response" type of passive obstacle avoidance. This method can only trigger emergency braking or replanning when a conflict is imminent, resulting in frequent starts and stops of the convoy, disrupted operation, and a risk of collision due to response delays. Its fundamental flaw lies in the lack of accurate prediction and utilization of the vehicle's future short-term trajectory; the conflict prediction lead time is usually less than 0.5 seconds, failing to allow sufficient time for collaborative decision-making.

[0009] 2. The centralized nature of collaborative decision-making severely hinders system robustness: As shown in comparison documents D2 (CN119437246A) and D3 (CN119124205A), their conflict detection and resolution strategies heavily rely on the central control unit for calculation and arbitration. Once the central node fails or the communication link is interrupted, the entire system's collaborative capability is lost, and the vehicle cannot autonomously handle sudden conflicts, potentially leading to partial or even global system paralysis. Tests have shown that within 30 seconds of a central unit failure, the collision risk of this type of centralized system increases by 80%, demonstrating the inherent vulnerability of the existing "centralized decision-making" architecture.

[0010] 3. Insufficient accuracy in environmental modeling and conflict detection: For example, in comparison file D2 (CN119437246A), the first grid size used for global planning is on the order of "10-20km," which is only suitable for macro-path planning and completely incapable of capturing and predicting vehicle-scale (centimeter-level) motion trajectories and micro-conflicts. Existing technologies generally do not combine high-precision positioning data (such as the 10cm-level accuracy of UWB) with refined environmental modeling, resulting in omissions or misjudgments in conflict detection, with a misjudgment rate as high as 10%-15%, seriously affecting obstacle avoidance accuracy and safety.

[0011] 4. Rigid and Simplistic Priority Decision-Making Mechanism: As shown in comparison document D3 (CN119124205A), priorities are mainly based on static or semi-static factors such as task energy consumption and completion rate, lacking comprehensive consideration of dynamic scenarios (such as road attributes and real-time distance). In complex intersections and multi-vehicle convergence scenarios, a single rule is prone to decision-making conflicts or deadlocks, with a deadlock incidence rate of approximately 8%-10%, reducing collaborative efficiency.

[0012] 5. Lack of a global resource dynamic optimization perspective: Existing methods mostly focus on resolving individual conflicts, failing to consider the overall system operational efficiency and monitor and adjust the global traffic flow distribution in real time. This can easily lead to local path congestion while other paths remain idle, limiting the system's throughput under high load. In high-density operation scenarios (number of vehicles ≥ 20), the throughput of existing systems is only 40%-50% of the theoretical optimal value. Summary of the Invention

[0013] This invention aims to overcome the shortcomings of the prior art and provide a communication and obstacle avoidance method and system for multi-vehicle cooperative tracking. The method aims to achieve proactive and forward-looking conflict warning (prediction lead time of 1-3 seconds) through "prediction of future trajectory occupancy under a high-precision virtual grid map", achieve decentralized and highly robust collaborative decision-making through "distributed negotiation based on multi-dimensional dynamic priority rules", and combine centralized global optimization to ultimately achieve efficient, safe and smooth operation of the multi-vehicle system, reduce the collision rate to below 1%, and increase the system throughput to more than 85% of the theoretical optimal value.

[0014] To achieve the above objectives, the present invention provides a communication and obstacle avoidance method for multi-vehicle cooperative tracking, the method being executed by a multi-vehicle cooperative tracking system, and comprising the following steps:

[0015] S1. System Initialization and Global Path Planning: The central control unit assigns a unique ID to each vehicle and plans an initial global path for each vehicle based on a preset global map and global map information. The global map is divided into several virtual grids to form a high-precision virtual grid map, which is then distributed to all vehicles. The grid encoding rule of the virtual grid map is "column number (uppercase English letter) + row number (Arabic numeral)". The mapping relationship between the grid coordinates and the global coordinate system is: X = (column number - 1) × 0.2m, Y = (row number - 1) × 0.2m, with the column number increasing sequentially from left to right and the row number increasing sequentially from bottom to top.

[0016] S2. Distributed State Awareness and Broadcasting: During operation, each vehicle periodically broadcasts its own state data packets through a wireless communication module. The state data packets include at least the vehicle ID, current position (mapped to grid coordinates), velocity vector (linear velocity v∈[0.1,1.0] m / s, angular velocity ω∈[-π / 2,π / 2] rad / s), and the set of virtual grids to be occupied in the next 1-3 seconds based on the current motion state, i.e., the pre-occupied grid set.

[0017] S3. Proactive Conflict Detection: Each vehicle receives and processes status data packets from other vehicles locally. It compares its own pre-defined grid set with that of other vehicles using a grid intersection detection algorithm (time complexity O(n+m), where n and m are the number of grids pre-defined for each vehicle) to determine if there are any intersecting grids. Simultaneously, it calculates the estimated conflict time based on the current speed and position of both vehicles and determines the conflict level according to predefined rules. If there is no conflict, the vehicle continues to run along its planned path and speed, and updates its status data packet in the next broadcast cycle. If a risky conflict exists, a local cooperative obstacle avoidance decision is immediately triggered, and when the conflict level is high-risk, an emergency broadcast is also triggered.

[0018] S4. Local Cooperative Obstacle Avoidance Decision: When a risk conflict is detected, the relevant vehicles autonomously negotiate based on preset multi-dimensional dynamic priority rules to determine the passage priority.

[0019] S5. Autonomous Avoidance and System Coordination: Based on the negotiation results, the low-priority vehicle can autonomously perform any of the avoidance actions, such as pre-deceleration, temporary stopping, or switching to a local backup path.

[0020] ① Pre-deceleration (deceleration a∈[-0.5,-0.2] m / s², uniform deceleration begins 3-5 grids before the collision point, and the target velocity is reduced to 30%-50% of the original velocity);

[0021] ② Temporary parking (parking in the safe area 2 grids before the conflict point, with parking attitude angle error ≤ ±1° and parking brake response time ≤ 100ms).

[0022] ③ Switch to a local backup path (executed when pre-deceleration or stopping cannot avoid conflict);

[0023] After executing the action, it updates its own status and broadcasts it. At the same time, the central control unit monitors the overall traffic status and performs system-level path load balancing adjustments.

[0024] Furthermore, in step S1, the global map information includes obstacle distribution, road attributes, and task point locations; the global path planning adopts an improved A* algorithm that introduces a path smoothing coefficient, the path smoothing coefficient being in the range of 0.1-0.3, and the driving trajectory is optimized by Gaussian blurring of the path inflection points to reduce the steering energy consumption of the car.

[0025] Furthermore, in step S2, the prediction method for the predetermined occupied grid is as follows: First, each vehicle obtains its current motion state and velocity vector through the positioning module, and fuses the obtained data using a Kalman filter; then, the trajectory points for the next 1-3 seconds are calculated based on the differential driving kinematics model, and the sampling interval of the trajectory points is 35-60ms; then, the trajectory points are mapped onto the high-precision virtual grid map to obtain the corresponding grid set; the broadcast period of the status data packet is 200-500ms.

[0026] Furthermore, in step S3, the conflict level determination rules specifically include:

[0027] No conflict: No intersecting grids, or the number of intersecting grids = 0;

[0028] Low-risk collisions: Number of intersecting grids = 1 and expected collision time > 1 second;

[0029] Medium-risk conflict: 1-2 intersecting grids with an expected conflict time of 0.5-1s, or 3-5 intersecting grids with an expected conflict time >1s;

[0030] High-risk collisions: ≥3 intersecting grids and expected collision time ≤1s;

[0031] Furthermore, in step S4, the multi-dimensional dynamic priority rules are executed in descending order of hierarchy; the hierarchy from highest to lowest is as follows:

[0032] ① Road attribute priority: Travel on main roads takes priority over travel on secondary roads. Main roads are defined as passages with a width of no less than 3 virtual grids and marked as main roads.

[0033] ② Task urgency and priority: Task urgency = Task deadline / Current progress × Weighting factor, the higher the value, the higher the priority; Completion progress = Distance traveled / Total path distance;

[0034] ③ Distance priority to conflict points: Distance is calculated using Manhattan distance. The smaller the distance, the higher the priority. The unit of distance is the number of grid cells.

[0035] ④ Car ID Priority: Cars with smaller ID numbers have priority.

[0036] Furthermore, when the task type is an urgent task, the correction factor is 1.5; when the task type is a normal task, the correction factor is 1.0; the task type is marked by the central control unit during task allocation. Specifically, the criteria for determining the task type are determined by the system's preset rules: urgent tasks include, but are not limited to, the following situations—(1) the task deadline is less than or equal to a preset threshold (e.g., 15 minutes); (2) the task is manually marked as urgent; (3) the delayed execution of the task will cause the associated production line to stop. Normal tasks are regular handling tasks other than urgent tasks. When allocating tasks, the central control unit automatically marks the task type according to the above criteria and determines the value of the weight correction factor accordingly.

[0037] Furthermore, in step S5, based on the decision result, the car with lower priority performs an active avoidance action, and the action type is selected according to the conflict level:

[0038] Low-risk conflict: Pre-deceleration (deceleration a∈[-0.5,-0.2] m / s², target velocity reduced to 50% of original velocity);

[0039] Medium-risk conflict: Pre-deceleration (target speed reduced to 30% of original speed) or temporary stop;

[0040] High-risk conflict: Temporary stop (emergency deceleration, deceleration a≤-0.8m / s²) or switch to an alternative local route.

[0041] Temporary parking must meet the following conditions: parking must be within the safe zone of two grids before the conflict point, the parking attitude angle error must be ≤ ±1°, and the parking brake response time must be ≤ 100ms. The safe zone for temporary parking is determined as follows: the grid and the adjacent grid are not covered by any grid to be occupied by any vehicle, and there are no static obstacles (such as shelves or equipment) in the grid.

[0042] The triggering condition for switching to an alternative local path is: if pre-deceleration or stopping cannot avoid conflict (i.e., there are still intersecting grids after execution and the expected conflict time is ≤0.3s), the vehicle sends an alternative path request to the central control unit. The request data packet includes its own ID, current position, conflicting grid information, and original path information. Within 300ms, the central control unit calculates and generates an alternative path using an improved A* algorithm. The alternative path must meet the following requirements: it does not conflict with grids pre-occupied by other vehicles, its total length increase is ≤20%, and its deviation from the original path is ≤5 grids. The central control unit issues the alternative path to the requesting vehicle. The vehicle uses a B-spline curve interpolation algorithm (number of control points ≥4, order 3) to achieve seamless switching between the original path and the alternative path. During the switching process, the speed fluctuation is ≤±0.1m / s, and the attitude angle change rate is ≤0.5rad / s.

[0043] The system-level path load balancing adjustment is specifically as follows: The central control unit calculates the virtual grid occupancy rate of each path in real time (occupancy rate = number of pre-defined grids / total number of grids in the path × 100%). When the occupancy rate of a certain path exceeds 70% (the preset threshold can be configured within the range of 50%-80%), it actively adjusts the global path for newly planned tasks or some vehicles in transit. When adjusting the path, it ensures that the deviation between the new path and the original path is ≤5 grids, and the increase in the total path length is ≤20%.

[0044] The present invention also provides a multi-vehicle cooperative tracking system, comprising:

[0045] The central control unit is used to perform system initialization, global path planning, high-precision virtual grid map management, and system-level load balancing scheduling.

[0046] Multiple intelligent vehicles are provided, each including at least a main control unit, a wireless communication module, a positioning module, and a motion control module. The main control unit is used for data fusion, collision detection, and decision calculation. The wireless communication module is used to receive the status of other vehicles. The positioning module is used to obtain the pose of its own vehicle, calculate, and broadcast a status data packet containing a predetermined occupied grid. The motion control module includes a motor drive unit and an attitude adjustment unit. The attitude adjustment unit uses a PID control algorithm to stabilize the vehicle's driving attitude during obstacle avoidance maneuvers, with an attitude stability error ≤ ±0.3°.

[0047] Furthermore, the intelligent vehicle also includes an environmental perception module for assisting in positioning and static obstacle detection, the environmental perception module including an ultrasonic sensor and a color sensor for identifying path marking lines.

[0048] Furthermore, the central control unit is equipped with a visual monitoring interface that displays the vehicle's location, path occupancy rate, conflict warning information, and system operating status in real time, and supports manual intervention for path adjustment and emergency stop control.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. Achieving a leap from passive obstacle avoidance to active, forward-looking obstacle avoidance: By introducing a "20cm×20cm high-precision virtual grid" and "predicted grid occupancy 1-3s in the future," combined with Kalman filter data fusion and differential-driven kinematics model, this invention can predict potential risks 1-3 seconds before physical conflict occurs, significantly improving the prediction lead time compared to the existing 0.5s, thus allowing ample time for collaborative decision-making. Testing shows that the conflict prediction accuracy of this invention is ≥98%, the frequency of emergency start-stop for the vehicle is reduced by more than 80% compared to existing technologies, and operational smoothness is significantly improved.

[0051] 2. A hybrid robust architecture of "centralized planning + distributed decision-making" was constructed: the global path is optimized and generated by the central control unit, while real-time obstacle avoidance decisions are made autonomously by each vehicle locally based on consistent rules, without relying on the central control unit. Even if the central control unit fails, the vehicles can still safely avoid obstacles through distributed negotiation, significantly improving system robustness. Tests showed that after a central control unit failure, the collision rate of this invention's system increased by only 5%-8%, while the collision rate of traditional centralized systems increased by more than 80%.

[0052] 3. A multi-dimensional dynamic priority rule adapted to complex scenarios is proposed: By integrating hierarchical decision-making rules based on multiple factors such as road attributes, task urgency, and real-time distance, this invention can flexibly and unambiguously handle complex scenarios such as intersections and multi-vehicle convergences. Testing shows that the deadlock rate of this invention is ≤1%, significantly lower than the 8%-10% of existing technologies, and the collaborative efficiency is improved by more than 30%.

[0053] 4. Improved accuracy and reliability of collision detection: By directly mapping high-precision positioning data such as UWB / IMU to a centimeter-level virtual grid, combined with an efficient grid intersection detection algorithm, the collision detection false positive rate is ≤2%, solving the problem of inaccurate detection caused by excessively large grids in existing technologies. Simultaneously, the reliability of status data transmission is ensured through CRC-32 checksum and TDMA communication mechanism, achieving a data transmission success rate of ≥99.9%.

[0054] 5. Balancing local efficiency with global optimization: While efficiently resolving real-time conflicts in a distributed manner, the central control unit implements global load balancing by monitoring grid occupancy and dynamically optimizes path allocation. Testing shows that in high-density operation scenarios (30 vehicles operating simultaneously), the throughput of this invention's system reaches over 85% of the theoretical optimal value, a significant improvement over the 40%-50% of existing technologies, and path resource utilization is increased.

[0055] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0056] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0057] Figure 1 This is a flowchart of a communication and obstacle avoidance method for multi-vehicle cooperative tracking in a selected embodiment of the present invention;

[0058] Figure 2 This is a schematic diagram of a virtual grid map in a selected embodiment of the present invention; it shows the current position, global path, and predetermined grid space occupied by vehicles A and B.

[0059] Figure 3 This is a schematic diagram of a conflict scenario in a preferred embodiment of the present invention where the predetermined occupied grids of vehicles A and B intersect;

[0060] Figure 4 This is a schematic diagram of obstacle avoidance decision-making based on dynamic priority rules (such as main road priority) in a preferred embodiment of the present invention. Detailed Implementation

[0061] The present invention will now be described in detail with reference to the embodiments shown in the accompanying drawings. However, it should be noted that these embodiments are not intended to limit the present invention. Equivalent transformations or substitutions in function, method, or structure made by those skilled in the art based on these embodiments are all within the scope of protection of the present invention.

[0062] This embodiment provides a multi-vehicle cooperative tracking system, which includes a central control unit and multiple intelligent vehicles, with the following specific structure:

[0063] The central control unit uses an industrial PC (CPU model Intel Core i7-12700H, memory ≥16GB, storage ≥512GB SSD), running the Ubuntu 20.04 operating system and the ROS Noetic framework, to perform system initialization, global path planning, high-precision virtual grid map management, and system-level load balancing scheduling.

[0064] Multiple intelligent vehicles, each including at least a main control unit, a wireless communication module, a positioning module, and a motion control module; wherein:

[0065] The main control unit uses an STM32F407ZGT6 microcontroller (168MHz main frequency, Flash ≥ 1MB, RAM ≥ 192KB) for data fusion, collision detection, and decision calculation; the wireless communication module uses an ESP32-WROOM-32 module, supports the 802.11n protocol, has a communication frequency band of 2.4GHz, a transmission rate ≥ 15Mbps, a communication distance ≥ 50m (unobstructed), and uses a TDMA time division multiple access mechanism to avoid communication collisions.

[0066] The positioning module adopts one of the following two schemes: ① Encoder (1024 lines resolution) and MPU6050 IMU fusion positioning (fusion algorithm is extended Kalman filter EKF, sampling frequency 100Hz), positioning accuracy ≤ ±2cm; ② DW1000 chip UWB positioning module (positioning refresh rate 20Hz, static positioning accuracy ≤ ±10cm, dynamic positioning accuracy ≤ ±15cm).

[0067] The motion control module includes a motor drive unit (model L298N, output current ≥2A) and an attitude adjustment unit. The attitude adjustment unit adopts a PID control algorithm (proportional coefficient Kp=5.0, integral coefficient Ki=0.1, derivative coefficient Kd=0.5, control cycle 10ms) to stabilize the vehicle's driving attitude during obstacle avoidance maneuvers, with an attitude stability error ≤±0.3°.

[0068] The intelligent vehicle is configured to: obtain its own pose through the positioning module, calculate and broadcast a status data packet containing a pre-defined occupied grid; receive the status of other vehicles through the wireless communication module, independently complete conflict detection based on the pre-defined occupied grid and collaborative obstacle avoidance decision based on multi-dimensional dynamic priority rules, and autonomously execute avoidance actions.

[0069] In this embodiment of the invention, the intelligent vehicle also includes an environmental perception module, which uses an HC-SR04 ultrasonic sensor (detection distance 2-400cm, measurement accuracy ±3mm) and a TCS34725 color sensor (used to identify path marking lines, identification distance 0-10cm) for auxiliary positioning and static obstacle detection.

[0070] In this embodiment of the invention, the central control unit is equipped with a visual monitoring interface (developed based on Qt 5.15) that displays the vehicle's location, path occupancy rate, conflict warning information, and system operating status in real time, and supports manual intervention for path adjustment and emergency stop control.

[0071] In this embodiment of the invention, the vehicle receives status information from other vehicles and performs conflict prediction locally. Once a path conflict is predicted, a priority-based distributed decision-making process is immediately triggered, without waiting for central instructions, ensuring real-time response. Based on the decision result, the vehicle performs avoidance actions such as deceleration and stopping, and then continues the cycle to achieve continuous safe driving.

[0072] The multi-vehicle cooperative tracking system of this invention has the architectural advantage of combining "centralized management" and "distributed control". The central control unit is responsible for macro-level planning and monitoring, while the vehicles are responsible for local real-time coordination. The two work together to ensure the efficiency and reliability of the system through information exchange.

[0073] like Figure 1 As shown, this embodiment of the invention also provides a communication and obstacle avoidance method for multi-vehicle cooperative tracking, executed by the aforementioned multi-vehicle cooperative tracking system; the method specifically includes the following steps:

[0074] S1: System Initialization and Global Path Planning

[0075] The central control unit (industrial PC) assigns a unique identifier (ID) to each vehicle in the system. The ID number is a consecutive integer from 1 to 50, ensuring that the vehicles can be distinguished.

[0076] The central control unit acquires global map information (including obstacle distribution, road attributes (main roads / minor roads), and task point locations) and the target tasks of all vehicles (such as start point, end point, task deadline, and task type). It then uses an improved A* algorithm to plan an initial globally optimal path from the start point to the end point for each vehicle. The improved A* algorithm introduces a path smoothing coefficient (range 0.1-0.3, default 0.2) and performs Gaussian blurring on path inflection points (3×3 fuzzy kernel size), reducing the number of vehicle turns (more than 40% less than the traditional A* algorithm) and lowering driving energy consumption (15%-20% reduction). Finally, it sends path information (including path point coordinates, desired speed, and road attribute annotations) to the corresponding vehicle.

[0077] The central control unit divides the global map into a high-precision virtual grid map of 20cm × 20cm, establishing a one-to-one mapping relationship between the grid coordinates and the global coordinate system (X = (column number - 1) × 0.2m, Y = (row number - 1) × 0.2m). This map and grid coding rules (column numbers are represented by uppercase English letters A, B, C... in sequence, and row numbers are represented by Arabic numerals 1, 2, 3... in sequence) are then distributed to all vehicles to ensure that all vehicles perform state calculations and conflict detection based on a unified map. After receiving the global information, each vehicle enters a continuous "perception-decision-action" cycle.

[0078] S2: Distributed communication based on state broadcasting

[0079] During operation, each vehicle obtains its real-time position (X, Y, θ) and velocity vector (v, ω) through a positioning module. The positioning module employs one of the following two methods:

[0080] Option 1: Encoder (1024 lines) and MPU6050IMU are used for fusion positioning. The fusion algorithm is extended Kalman filter (EKF), the sampling frequency is 100Hz, the positioning accuracy is ≤±2cm, and the attitude angle error is ≤±0.5°.

[0081] Option 2: DW1000 chip UWB positioning module, positioning refresh rate 20Hz, static positioning accuracy ≤±10cm, dynamic positioning accuracy ≤±15cm.

[0082] The vehicle uses a Kalman filter algorithm to fuse positioning data, reducing noise interference and improving data accuracy. The Kalman filter data fusion parameters are designed as follows: the process noise covariance Q is set based on the inherent noise characteristics of the positioning module. The IMU's angular velocity noise is approximately 1e-6 rad² / s², and the encoder's position noise is approximately 1e-4 m², therefore the Q matrix is ​​set to diag([1e-4,1e-4,1e-6,1e-6]); the measurement noise covariance R is obtained statistically from actual test data. The positioning module's position measurement error is approximately 1e-3 m², and the attitude angle measurement error is approximately 1e-5 rad², therefore the R matrix is ​​set to diag([1e-3,1e-3,1e-5]).

[0083] The car is based on a differential driving kinematics model ( , , The system calculates trajectory points within the next 1-3 seconds, with a sampling interval of 35-60ms, preferably 50ms. These trajectory points are then mapped onto a virtual grid map, generating a set of "pre-occupied grids" (i.e., the grids the car will pass through within the next 1-3 seconds). For example, a car traveling at a speed of 0.5m / s that travels 1m in 2 seconds corresponds to 5 consecutive 20cm×20cm grids; the pre-occupied grids cover these 5 grids.

[0084] The vehicle broadcasts its status data packets to all other vehicles in the system via a wireless communication module (ESP32-WROOM-32). The broadcast period is 200-500ms (default 300ms), and a TDMA (Time Division Multiple Access) mechanism is used to avoid communication conflicts. The status data packets use the LZ77 data compression algorithm (sliding window size 4KB, history buffer size 16KB) to reduce the size (compression ratio ≥3:1), and a CRC-32 checksum is added to ensure transmission accuracy. The data packet format (total length ≤256 bytes) is shown in Table 1 below:

[0085] Table 1 Data Packet Format

[0086] The task urgency is calculated as: task deadline (minutes) / current progress × correction factor. Current progress is calculated as: distance traveled / total path distance (rounded to two decimal places). The correction factor for urgent tasks is 1.5, and the correction factor for ordinary tasks is 1.0.

[0087] S3: Conflict Prediction and Decision-Making Based on Pre-determined Occupancy

[0088] After receiving status data packets from other vehicles, each vehicle first verifies the integrity of the data using a CRC-32 checksum. If the data is incorrect, it is discarded. If the data is correct, it is merged with its own status data and a real-time updated global traffic situation map (containing the position, speed, reserved grid space, and task urgency of all vehicles) is maintained in local memory. The situation map update cycle is consistent with the broadcast cycle.

[0089] The car uses a grid intersection detection algorithm (time complexity O(n+m)) to traverse its own pre-assigned grid set and the pre-assigned grid sets of other cars to determine if there are any intersecting grids. Simultaneously, based on the current speed and position of both cars, the estimated conflict time is calculated (t=d / v_relative, where d is the current distance between the two cars and v_relative is the relative speed), and the conflict level is determined according to the following rules:

[0090] No conflict: No intersecting grids, or the number of intersecting grids = 0;

[0091] Low-risk collisions: Number of intersecting grids = 1 and expected collision time > 1 second;

[0092] Medium-risk conflict: 1-2 intersecting grids with an expected conflict time of 0.5-1s, or 3-5 intersecting grids with an expected conflict time >1s;

[0093] High-risk conflict: The number of intersecting grids is ≥3 and the expected conflict time is ≤1s.

[0094] If there is no conflict, the vehicle continues to travel along the original planned path and speed, and updates the status data packet in the next broadcast cycle. If there is a low / medium / high risk conflict, a local cooperative obstacle avoidance decision is immediately triggered, and a high-risk conflict additionally triggers an emergency broadcast (the data packet flag is set to 0xFF).

[0095] S4: Local Cooperative Obstacle Avoidance Decision

[0096] When a potential conflict is detected, the relevant vehicles autonomously negotiate and make decisions based on preset multi-dimensional dynamic priority rules, without relying on the central control unit. The rules are executed in the following hierarchy (priority from high to low):

[0097] (1) Main Road Priority Rule: Cars on main roads have priority over cars on side roads; where a main road is defined as a passage with a width of ≥3 virtual grids and marked as "main road" on the global map, and a side road is defined as a passage with a width of <3 virtual grids or marked as "side road". If both cars are on the main road or both are on the side road, then proceed to the next level of judgment.

[0098] (2) Task urgency priority rule: The car with the higher task urgency value takes priority. If the difference in urgency between two cars is ≤10, then proceed to the next level of judgment.

[0099] (3) Distance Priority Rule for Conflict Points: Cars with smaller Manhattan distances to conflicting grids (intersecting grids) are given priority. Manhattan distance d = |x1-x2| + |y1-y2|, where the distance unit is the number of grids; where x1 and y1 are the current grid coordinates of the car, and x2 and y2 are the coordinates of the conflicting grid. If the distance difference between the two cars is ≤ 1 grid, then proceed to the next level of judgment.

[0100] (4) ID priority rule: The car with the smaller ID number has priority, which is the final arbitration rule.

[0101] In one specific embodiment, based on the decision result, the vehicle with lower priority performs an active avoidance action, and the action type is selected according to the conflict level as follows:

[0102] Low-risk conflict: pre-deceleration (deceleration a∈[-0.5,-0.2]m / s², target velocity reduced to 50% of original velocity);

[0103] Medium-risk conflict: Pre-deceleration (target speed reduced to 30% of original speed) or temporary stop;

[0104] High-risk conflict: Temporary stop or switching to an alternative local route.

[0105] Temporary parking must meet the following requirements: parking must be within the safe zone of two grids before the conflict point (this grid and the adjacent grid are not covered by any pre-occupied grid of any vehicle, and there are no static obstacles), the parking attitude angle error must be ≤ ±1°, and the parking brake response time must be ≤ 100ms.

[0106] The triggering conditions for switching to an alternative local path are: if pre-deceleration or stopping cannot avoid conflict (i.e., there are still intersecting grids after execution and the expected conflict time is ≤0.3s), the vehicle sends an alternative path request to the central control unit. The request data packet includes its own ID, current position, conflicting grid information, and original path information. Within 300ms, the central control unit calculates and generates an alternative path using an improved A* algorithm. The alternative path must meet the following requirements: it does not conflict with grids pre-occupied by other vehicles, its total length increase is ≤20%, and its deviation from the original path is ≤5 grids. The central control unit issues the alternative path to the requesting vehicle. The vehicle uses a B-spline curve interpolation algorithm (number of control points ≥4, order 3) to achieve seamless switching between the original path and the alternative path. During the switching process, the speed fluctuation is ≤±0.1m / s, and the attitude angle change rate is ≤0.5rad / s.

[0107] S5: Action Execution and Information Synchronization

[0108] The vehicle performing the avoidance maneuver updates its "current position, speed, reserved grid occupancy, and avoidance status (pre-deceleration / stop / path switching)" information in the status data packet of its next broadcast cycle, and reports its status change to the central control unit.

[0109] The central control unit receives real-time status reports from all vehicles and calculates the virtual grid occupancy rate for each path (occupancy rate = number of pre-assigned grids / total number of grids on the path × 100%), with a calculation period of 1 second. When the occupancy rate of a path exceeds 70% (the preset threshold can be configured within the range of 50%-80%), it is determined to be congested, and the following load balancing measures are automatically implemented:

[0110] Plan alternative routes for the new task vehicle, prioritizing routes with an occupancy rate of less than 50%;

[0111] Send route adjustment suggestions to vehicles currently on congested routes that have not entered conflict areas. After receiving the suggestions, the vehicles can decide whether to switch routes (switching condition: the increase in the length of the new route is ≤15%).

[0112] Adjust the expected speed of cars on congested routes, reducing the speed limit by 20%-30% to avoid further congestion.

[0113] The central control unit displays the vehicle's location, path occupancy rate, conflict warning information, and system operating status in real time through a visual monitoring interface, and supports manual intervention for path adjustment and emergency stop control; the emergency stop command response time is ≤50ms.

[0114] In this invention, the optimization of the load balancing threshold is based on the following: by testing the system throughput under different path occupancy rates, it was found that when the occupancy rate exceeds 70%, the growth rate of system throughput slows down significantly, and the collision risk begins to rise rapidly; when the occupancy rate is below 50%, path resources are idle. Therefore, the congestion judgment threshold is set to 70%, and backup paths are preferentially selected from those with an occupancy rate of less than 50%.

[0115] like Figure 2 The diagram shown is a schematic diagram of a high-precision virtual grid map in a preferred embodiment of the present invention; it displays the current position, global path, and predetermined grid space occupied by vehicles A and B. Figure 2 The high-precision virtual grid map established in step S1 and the state perception and broadcasting mechanism based on this map in step S2 are demonstrated. Specifically, the entire driving area is divided into multiple uniformly sized virtual grids, each grid having a unique coordinate identifier (e.g., A-1, B-2).

[0116] Car and Global Path: Car A is traveling from west to east (from left to right) along the main road, while Car B is traveling from south to north (from bottom to top) along the side road. Their paths intersect in grids D-4 and E-4, forming a conflict point.

[0117] Pre-defined occupancy grids: The pre-defined occupancy grid for car A covers grids {C-4, D-4, E-4, F-4}. This is the area that car A will predict it will travel through in the near future (e.g., 2 seconds) based on its current speed and position. The pre-defined occupancy grid for car B covers grids {D-3, D-4, D-5, E-4}. This is the future trajectory area predicted by car B based on its own state.

[0118] Conflict Prediction: As shown in the figure, the "pre-arranged occupied grids" of car A and car B intersect at grids D-4 and E-4. This intersection serves as a warning to the system of potential conflict risk before a physical collision occurs. According to the method of the present invention, the two vehicles will immediately trigger a local cooperative obstacle avoidance decision (e.g., according to the rules, car A on the main road has priority, and car B on the side road should slow down or stop to yield at grid D-3 or D-5).

[0119] like Figure 3The diagram shown is a schematic representation of a conflict scenario where the pre-defined occupied grids of vehicles A and B intersect in a preferred embodiment of the present invention. The diagram visually illustrates the scenario in S3 where the "pre-defined occupied grid of the vehicle itself" intersects (conflicts) with the "pre-defined occupied grids of other vehicles," providing a visual representation of the conflict prediction algorithm. Specifically, the diagram clearly demonstrates the four core stages of the method of the present invention:

[0120] Phase 1: Normal Driving and Status Broadcast

[0121] During their journey, cars A and B periodically broadcast their status data packets via a wireless network. The core content of these status data packets is their "reserved grid space".

[0122] Phase Two: Conflict Prediction

[0123] After receiving data packets from the other vehicle, each vehicle performs data fusion and conflict prediction calculations locally. Both vehicles independently discover that their "pre-booked occupancy grids" intersect in the D-4 grid, thus predicting potential collision risks in advance.

[0124] Phase 3: Triggering Local Cooperative Obstacle Avoidance Decisions

[0125] Upon anticipating a conflict, both vehicles simultaneously trigger their local decision-making processes. They make independent, distributed decisions based on a pre-defined, consistent dynamic priority rule ("main road priority" in this example).

[0126] The decision-making results are distributed and consistent: car A determines that it has priority; car B determines that it needs to yield.

[0127] Phase Four: Execution of Actions and Rebroadcast

[0128] Car B performs an avoidance maneuver (decelerates or stops on grid D-3) and immediately broadcasts a new status, informing other vehicles of its new intention (such as the predetermined grid becoming empty or containing only the current position).

[0129] Car A continues to travel at its original speed and updates its scheduled grid to continue broadcasting.

[0130] Figure 3 This powerfully demonstrates that the present invention, through "pre-occupied grid" and "distributed negotiation decision-making," achieves a leap from "passive response" to "proactive prediction and collaboration," effectively avoiding collisions and deadlocks, and ensuring the safety and efficiency of the system.

[0131] like Figure 4The diagram shown is a schematic representation of obstacle avoidance decision-making based on dynamic priority rules (such as main roads taking precedence) in a preferred embodiment of the present invention. It illustrates how two cars in scenario S4 negotiate and make decisions based on rules such as "road attribute priority" (main roads take precedence over side roads) at an intersection. This decision flowchart demonstrates a distributed, self-negotiating obstacle avoidance logic. Its core lies in a consistent set of rules shared by all cars, ensuring that each car arrives at a unified conclusion after making independent decisions. Specifically:

[0132] Orderly judgment of rules: The decision-making process acts like a filter, starting with the most specific and reasonable rules.

[0133] First, determine the road relationship: Are they on the same road? Are they going in the same direction or opposite directions? This solves the most basic traffic scenario.

[0134] Secondly, determine the road classification: the "main road priority rule" is the most commonly used and efficient rule for resolving intersection conflicts.

[0135] Then consider task attributes: the "task urgency priority rule" ensures that the system can handle special tasks.

[0136] Finally, the arbitration rule is used: when none of the above rules can determine the outcome, "distance priority" and "ID priority" are used as the final arbitration to ensure that the decision can always be made and is unambiguous.

[0137] Clear output action: The final output of a decision is not a complex instruction, but a choice between two options: keep the right or give up the right.

[0138] High priority (right of way): The car confirms that it has the right of way and maintains its original state of travel, which is the key to achieving smoothness.

[0139] Low priority (give way): The car understands that it has the obligation to give way and takes the initiative to perform actions such as pre-deceleration, stopping or changing lanes.

[0140] Closed loop and feedback: Regardless of the action performed, the vehicle will update its own status and broadcast it immediately, so that the traffic situation can be updated. Other vehicles can make predictions and decisions based on the new information, forming a complete closed loop control.

[0141] Figure 4 It is clearly shown that the obstacle avoidance decision-making mechanism of the present invention is rule-driven, distributed, and ultimately consistent, which perfectly solves the core problem of how to achieve collaborative decision-making in decentralized systems.

[0142] Example 1

[0143] In this embodiment, the scenario is set as follows: "30 AGVs collaboratively transport goods in a warehouse logistics workshop (20m long × 15m wide). There are 3 crossroads. The main road is east-west (3 grids wide, 60cm), and the branch road is north-south (2 grids wide, 40cm). The AGV load capacity is ≤50kg, the maximum speed is 1.0m / s, and the AGV density during peak operation is ≥2 vehicles / m²."

[0144] 1. System Initialization and Global Path Planning (S1)

[0145] (1) The central control unit (Intel Core i7-12700H industrial PC) assigns IDs (1-30) to 30 AGVs and obtains the workshop map (including main road / branch road markings, shelf obstacle distribution (coordinates known), task point location) and task information:

[0146] AGV1 (ID=1): Start point A1 (grid A1, X=0m, Y=0m) → End point B1 (grid E10, X=9.8m, Y=1.8m), task type: normal, deadline: 30 minutes, total path distance: 12m;

[0147] AGV2 (ID=2): Starting point A2 (grid C20, X=0.4m, Y=3.8m) → End point B2 (grid E5, X=9.8m, Y=0.8m), task type urgent, deadline 20 minutes, total path distance 10m;

[0148] The other AGV task information is similar, including ordinary tasks and emergency tasks (accounting for 10%).

[0149] (2) The global path is planned using the improved A* algorithm (path smoothing coefficient 0.2, Gaussian blur kernel 3×3):

[0150] AGV1 path: A1→B1→C1→...→E10 (travels on the main road, 12m in total, expected speed 0.8m / s);

[0151] AGV2 path: C20→C19→...→C5→D5→E5 (branch road travel, total distance 10m, expected speed 0.6m / s);

[0152] The algorithm automatically avoids shelf obstacles during path planning, reducing the number of turns by 45% compared to the traditional A* algorithm.

[0153] (3) Generate a 20cm×20cm virtual grid map (100 columns×75 rows in total) and distribute it to all AGVs; the grid coding rule is “column number (AJ, corresponding to X=0-19.8m) + row number (1-75, corresponding to Y=0-14.8m)”, such as the C5 grid coordinates corresponding to the global coordinate system (X=0.4m, Y=0.8m).

[0154] 2. Status Broadcast (S2)

[0155] (1) AGV1 adopts "encoder + MPU6050 IMU fusion positioning" with a sampling frequency of 100Hz to obtain real-time position (grid D8, X=0.6m, Y=1.4m) and speed (0.8m / s, eastward). After optimizing the data through Kalman filtering (Q=diag ([1e-4,1e-4,1e-6,1e-6]), R=diag ([1e-3,1e-3,1e-5])), the trajectory for the next 2 seconds is calculated based on the differential driving model (sampling interval 50ms, a total of 40 trajectory points), and mapped to the grid to obtain the predetermined occupied grid {D8, D9, D10, E8, E9, E10}; the current completion progress = distance traveled 3m / total distance 12m = 0.25, and the task urgency = 30 / 0.25×1.0 = 120.

[0156] (2) The AGV2 uses a UWB positioning module (DW1000) with a positioning refresh rate of 20Hz to obtain the real-time position (grid C6, X=0.4m, Y=1.0m) and speed (0.6m / s, eastward). It calculates the pre-occupied grid {C6, C7, C8, D6, D7, D8} corresponding to the trajectory in the next 2 seconds. The current progress = distance traveled 4m / total distance 10m = 0.4, and the task urgency = 20 / 0.4 × 1.5 = 75.

[0157] (3) AGV1, AGV2 and other AGVs broadcast status data packets through the ESP32-WROOM-32 module with a period of 300ms. LZ77 compression (compression ratio 3.2:1) and CRC-32 verification are used. The data transmission success rate is 99.95% and the communication delay is ≤50ms.

[0158] 3. Conflict Prediction (S3)

[0159] (1) After receiving the data packet from AGV2, AGV1 passes the CRC check and merges the data to generate a global traffic situation map. Through the grid intersection detection algorithm, it is found that the grids it plans to occupy {D8, D9, D10, E8, E9, E10} intersect with AGV2's {C6, C7, C8, D6, D7, D8} at grid D8. The estimated conflict time is calculated as follows: the current distance between the two vehicles is 0.5m, the relative speed is 0.2m / s, t=0.5 / 0.2=2.5s, which is determined to be a low-risk conflict, triggering an obstacle avoidance decision.

[0160] (2) Similarly, AGV2 detects the grid intersection with AGV1, determines it as a low-risk conflict, and triggers a decision.

[0161] 4. Local Cooperative Obstacle Avoidance Decision-Making (S4)

[0162] (1) AGV1 and AGV2 negotiate based on multi-dimensional dynamic priority rules:

[0163] Road attribute determination: AGV1 is on the main road, AGV2 is on the side road. AGV1 has higher priority than AGV2 and does not need to be judged at a later level.

[0164] Final priority ranking: AGV1 > AGV2.

[0165] (2) Decision result: AGV1 maintains the original speed (0.8m / s) and moves; AGV2 performs pre-deceleration, with a deceleration acceleration of -0.3m / s², and the target speed is reduced to 0.3m / s (50% of the original speed). The grid to be occupied in the next 2 seconds is updated to {C6, C7, C8, C9} to avoid entering the D8 grid at the same time as AGV1.

[0166] 5. Action execution and information synchronization (S5)

[0167] (1) AGV2 decelerates uniformly according to the preset acceleration. After 300ms, the speed drops to 0.3m / s. In the next broadcast cycle, the status data packet is updated: current position C7, speed 0.3m / s, pre-occupied grid {C6, C7, C8, C9}, avoidance status "pre-deceleration", and reported to the central control unit.

[0168] (2) AGV1 is driving normally and broadcasts status data packets: current position D9, speed 0.8m / s, reserved grid {D9, D10, E8, E9, E10, E11}.

[0169] (3) The central control unit receives all AGV status reports and calculates the occupancy rate of each path: the occupancy rate of the main road (AGV1 path) is 55%, and the occupancy rate of the branch road (AGV2 path) is 62%, both of which do not exceed the 70% threshold, so there is no need to adjust the global path.

[0170] (4) When AGV1 passes through the D8 grid, AGV2 detects that the conflict has been resolved, and recovers to its original speed of 0.6m / s with an acceleration of 0.2m / s², updates the status data packet and broadcasts it.

[0171] 6. System hardware parameter verification

[0172] In this embodiment, the core hardware parameters of the multi-vehicle cooperative tracking system are as follows:

[0173] Central control unit: Intel Core i7-12700H, 16GB RAM, 512GB SSD, Ubuntu 20.04+ROSNoetic, path planning response time ≤200ms;

[0174] AGV main control unit: STM32F407ZGT6, 168MHz main frequency, collision detection and decision calculation time ≤100ms;

[0175] Wireless communication module: ESP32-WROOM-32, 2.4GHz band, transmission rate 20Mbps, communication distance 60m (unobstructed).

[0176] Positioning module: encoder + MPU6050 fusion positioning, static positioning accuracy ±1.5cm, dynamic positioning accuracy ±2cm;

[0177] Motion control module: L298N motor drive, PID control (Kp=5.0, Ki=0.1, Kd=0.5), attitude stability error ±0.2°.

[0178] 7. Technical Performance Test Data

[0179] In this embodiment, the system was tested continuously for 8 hours. The test results are shown in Table 2.

[0180] Table 2 Comparison of Test Results

[0181] The test results above show that the system of the present invention is significantly superior to the existing technology in terms of safety, efficiency and robustness, and fully meets the needs of multi-vehicle collaborative operation in high-density and dynamic environments in warehousing and logistics workshops.

[0182] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A communication and obstacle avoidance method for multi-vehicle cooperative tracking, characterized in that, The method is executed by a multi-vehicle cooperative tracking system and includes the following steps: S1. System Initialization and Global Path Planning: The central control unit assigns a unique ID to each vehicle and plans an initial global path for each vehicle based on a preset global map and global map information; wherein, the global map is divided into several virtual grids to form a high-precision virtual grid map, and the map is distributed to all vehicles; S2. Distributed State Awareness and Broadcasting: During operation, each vehicle periodically broadcasts its own state data packets through the wireless communication module. The state data packets include at least the vehicle ID, current position, velocity vector, and the set of virtual grids that will be occupied in the next 1-3 seconds based on the current motion state, i.e., the pre-occupied grid set. S3. Proactive Conflict Detection: Each vehicle receives and processes status data packets from other vehicles locally. It compares its own pre-defined grid set with that of other vehicles using a grid intersection detection algorithm to determine if there are any intersecting grids. Simultaneously, it calculates the estimated conflict time based on the current speed and position of both vehicles and determines the conflict level according to set rules. If there is no conflict, the vehicle continues to run according to the planned path and speed and updates its status data packet in the next broadcast cycle. If a risky conflict exists, a local cooperative obstacle avoidance decision is immediately triggered, and when the conflict level is a high-risk conflict, an emergency state broadcast is also triggered. S4. Local Cooperative Obstacle Avoidance Decision: When a risk conflict is detected, the relevant vehicles autonomously negotiate based on preset multi-dimensional dynamic priority rules to determine the passage priority. S5. Autonomous Avoidance and System Coordination: Based on the negotiation results, low-priority vehicles autonomously perform any avoidance action, such as pre-deceleration, temporary stopping, or switching to a local backup path; after performing the action, they update their own status and broadcast it, while the central control unit monitors the overall traffic status and performs system-level path load balancing adjustments.

2. The method according to claim 1, characterized in that, In step S1, the global map information includes obstacle distribution, road attributes, and task point locations; the global path planning is performed using an improved A* algorithm that introduces a path smoothing coefficient, the path smoothing coefficient being in the range of 0.1-0.3, and the driving trajectory is optimized by Gaussian blurring of the path inflection points to reduce the steering energy consumption of the car.

3. The method according to claim 1, characterized in that, In step S2, the prediction method for the predetermined occupied grid is as follows: First, each vehicle obtains its current motion state and velocity vector through the positioning module, and then uses a Kalman filter to fuse the obtained data. Then, based on the differential driving kinematics model, the trajectory points in the next 1-3 seconds are calculated, and the sampling interval of the trajectory points is 35-60ms; then the trajectory points are mapped onto the high-precision virtual grid map to obtain the corresponding grid set; the broadcast period of the status data packet is 200-500ms.

4. The method according to claim 1, characterized in that, In step S3, the conflict level determination rules specifically include: No conflict: No intersecting grids, or the number of intersecting grids = 0; Low-risk collisions: Number of intersecting grids = 1 and expected collision time > 1 second; Medium-risk conflict: 1-2 intersecting grids with an expected conflict time of 0.5-1s, or 3-5 intersecting grids with an expected conflict time >1s; High-risk conflict: The number of intersecting grids is ≥3 and the expected conflict time is ≤1s.

5. The method according to claim 1, characterized in that, In step S4, the multi-dimensional dynamic priority rules are executed in descending order of hierarchy; the hierarchy from highest to lowest is as follows: ① Road attribute priority: Travel on main roads takes priority over travel on secondary roads. Main roads are defined as passages with a width of no less than 3 virtual grids and marked as main roads. ② Task urgency priority: Task urgency = Task deadline / Current progress × Weighting factor, the higher the value, the higher the priority; Progress completed = Distance traveled / Total distance traveled; ③ Distance priority to conflict points: Distance is calculated using Manhattan distance. The smaller the distance, the higher the priority. The unit of distance is the number of grid cells. ④ Car ID Priority: Cars with smaller ID numbers have priority.

6. The method according to claim 5, characterized in that, When the task type is an emergency task, the correction factor is 1.5; when the task type is a normal task, the correction factor is 1.0; the task type is marked by the central control unit when assigning tasks.

7. The method according to claim 1, characterized in that, In step S5, based on the decision result, the car with lower priority performs an active avoidance action, and the action type is selected according to the conflict level: Low-risk conflict: Pre-deceleration, deceleration acceleration a∈[-0.5,-0.2] m / s², target velocity reduced to 50% of original velocity; Medium-risk conflict: Pre-deceleration or temporary stop, with the target speed reduced to 30% of the original speed; High-risk conflict: temporary stop or switching to an alternative local route; Temporary parking must meet the following requirements: parking must be done in the safe area two grids before the conflict point, parking attitude angle error ≤ ±1°, and parking brake response time ≤ 100ms. The switching of the local backup path is generated by the central control unit in response to the vehicle's request, using an improved A* algorithm. During path switching, a smooth transition of motion posture is achieved through a B-spline curve interpolation algorithm, and the speed fluctuation during path switching is ≤ ±0.1m / s. The system-level path load balancing adjustment specifically involves the central control unit calculating the virtual grid occupancy rate of each path in real time. When the occupancy rate of a certain path exceeds the pre-approval threshold, the global path is proactively adjusted for newly planned tasks or some vehicles in transit. During path adjustment, the deviation between the new path and the original path is ensured to be ≤5 grids, and the increase in the total path length is ≤20%.

8. A multi-vehicle cooperative tracking system for use in the method described in any one of claims 1-7, characterized in that, include: The central control unit is used to perform system initialization, global path planning, high-precision virtual grid map management, and system-level load balancing scheduling. Multiple intelligent vehicles are provided, each including at least a main control unit, a wireless communication module, a positioning module, and a motion control module. The main control unit is used for data fusion, collision detection, and decision calculation. The wireless communication module is used to receive the status of other vehicles. The positioning module is used to obtain the pose of its own vehicle, calculate, and broadcast a status data packet containing a predetermined occupied grid. The motion control module includes a motor drive unit and an attitude adjustment unit. The attitude adjustment unit uses a PID control algorithm to stabilize the vehicle's driving attitude during obstacle avoidance maneuvers, with an attitude stability error ≤ ±0.3°.

9. The system according to claim 8, characterized in that, The intelligent vehicle also includes an environmental perception module for assisting in positioning and static obstacle detection. The environmental perception module includes an ultrasonic sensor and a color sensor for recognizing path marking lines.

10. The system according to claim 8, characterized in that, The central control unit is equipped with a visual monitoring interface that displays the vehicle's location, path occupancy rate, conflict warning information, and system operating status in real time, and supports manual intervention for path adjustment and emergency stop control.

Images