Formation driving coordination method based on vehicle infrastructure cooperation

Abstract

The invention relates to a formation driving coordination method based on vehicle-road cooperation, and the method comprises the steps: a road side unit RSU is used for releasing a traffic constraint weight value containing the information of passing restriction, path guidance, risk distribution and the like into a formation driving region, and the weight value forms a traffic environment guidance domain with a potential gradient at a vehicle end; the vehicle autonomously completes formation aggregation and position arrangement based on the own state and the weight change trend in the local sensing environment; after the formation is completed, the vehicles dynamically form a multi-level influence circle structure according to relative positions, historical cooperative behaviors and influence factors provided by the RSU, a current leading vehicle exists in each layer of influence circle, the leading vehicle has local scheduling guidance capability, and when the state of the leading vehicle is unstable or the behavior is abnormal, the control right is transferred to the auxiliary control vehicle; according to the method, the traffic environment guidance domain with the potential gradient is constructed through the traffic constraint weight value issued by the RSU, and the vehicles can autonomously perform formation aggregation and array sorting based on the local weight change trend.

Description

TECHNICAL FIELD

[0001] The present application relates to a platoon driving coordination method, in particular to a platoon driving coordination method based on vehicle-road cooperation. BACKGROUND

[0002] The disclosed patent document CN107798861B is a kind of vehicle cooperative platoon driving method and system patent, although the scheme can realize the automatic platoon convergence, following, exit, parking and collaborative lane changing and other collaborative behaviors based on perception fusion, path planning and decision control between vehicles, and 5G communication and low-cost multi-source sensor are used to improve the universality of system deployment, but in the flexible platoon, heterogeneous response and multi-target dynamic cooperation under the complex vehicle-road cooperation environment, there are still many technical bottlenecks and practical drawbacks.

[0003] Firstly, the scheme still focuses on vehicle-side multi-source perception and decentralized following strategy in essence, although the vehicle's awareness of its surrounding environment and platoon adaptive ability are improved through millimeter wave radar, vision and high-precision positioning, the response to the overall traffic environment, road-level events and remote risks mainly depends on the vehicle's local perception and direct communication of information with nearby vehicles, which still has the risk of response delay, poor global consistency and information island for multi-vehicle parallel, heterogeneous platoon, road structure dynamic adjustment, remote traffic events (such as remote construction area, emergency diversion, sudden accident chain, etc.) and highly asymmetric traffic flow.

[0004] Secondly, although this scheme realizes decentralized following strategy to reduce control jitter caused by network fluctuations, its core is still based on point-to-point real-time information interaction between follower and leader vehicles, which cannot effectively coordinate new collaborative structures such as multi-leader, multi-level scheduling, and multi-influence circle fusion when the queue is expanded on a large scale. For example, in the case of multi-leader vehicle interval boundary overlap, cross formation, lane rapid switching, or temporary road topology contraction, it is difficult to support multi-objective dynamic scheduling and flexible queue autonomy in complex traffic scenarios. Thirdly, this scheme integrates perception fusion, path planning, and decision control under a single software architecture, which is highly modular at the vehicle end. Although it is easy to expand horizontally, it lacks the environmental guidance weight, risk distribution, and path induction provided by roadside infrastructure (such as RSU) system behavior field. The collaborative behavior between vehicles mainly stays in the three-stage process of perception alignment, decision synchronization, and action execution, and it is difficult to realize the induction and convergence of full-queue self-organizing behavior through spatial potential gradient, traffic constraint weight mapping, and time delay correction mechanisms. For example, in the high-flow asynchronous response scenario, if some vehicles have weak perception ability or there is a 5G communication instantaneous bandwidth degradation, the existing scheme can only make local corrections through following instruction redundancy or vehicle state prediction, and cannot dynamically adjust the behavior weight and response rhythm according to the global synchronous / asynchronous distribution of the queue. It is easy to cause local chain oscillation, delay accumulation, and a sharp decline in formation aggregation efficiency. In addition, the existing method lacks a high-dimensional historical behavior backtracking and elastic trajectory reuse mechanism. Once the queue has abnormal members, communication failures, structural disturbances, or unavailable historical states, it can only rely on the last valid perception / path planning or limited experiential templates for emergency adjustment, and cannot structurally filter, race-optimize, splice, or reconstruct historical formation fragments or adapt to the current environment through spatial elastic deformation mapping. This results in a single recovery path and limited system robustness. The existing scheme also does not involve multi-dimensional weight distribution of the leader vehicle, scheduling residual prediction, multi-circle granularity compression, boundary suppression, and dynamic transfer mechanisms, making it difficult to realize dynamic influence circle switching, automatic suppression of scheduling area overlap, and seamless collaboration of multi-level scheduling, leading to multi-leader conflicts, local autonomy failures, and degradation of formation functions when communication bandwidth is limited.

[0005] In addition, for the conflict caused by inconsistent environmental perception, the existing system can only rely on local sensor anomaly detection or passive obstacle avoidance, and does not construct a high-order conflict resolution and suppression chain diffusion mechanism based on weight dynamic adjustment, short-time scheduling priority and disturbance variable single injection, which is easy to cause continuous destruction of the formation or chain safety risk under complex working conditions. In summary, the current formation driving method based on vehicle side decentralized perception fusion and following control lacks deep integration of key technical links such as road side guidance, global potential field construction, elastic trajectory mapping, historical memory reconstruction, behavior scheduling flexible compression and multi-main circle autonomy, and is difficult to support the high safety, high efficiency and high resilience intelligent vehicle formation collaborative evolution in the future large-scale, multi-source heterogeneous, strong environmental dynamic and high complexity collaborative scene. SUMMARY

[0006] The purpose of the present application is to provide a formation driving coordination method based on vehicle-road cooperation, so as to solve some of the problems and deficiencies pointed out in the background art.

[0007] The present application solves the above-mentioned technical problems by adopting the following technical solutions:

[0008] The roadside unit RSU releases traffic constraint weight values containing traffic restrictions, path guidance, risk distribution and other information to the formation driving area, and the weight values form a traffic environment guidance domain with a potential gradient at the vehicle end. The vehicle autonomously completes formation aggregation and position arrangement based on its own state and the weight change trend in the local perception environment.

[0009] After completing the formation, the vehicle dynamically forms a multi-level influence circle structure according to the relative position, historical cooperative behavior and influence factor provided by the RSU. Each level of the influence circle has a current dominant vehicle, and the dominant vehicle has the ability of local scheduling guidance. When the state of the dominant vehicle is unstable or abnormal, the control right is transferred to the vice control vehicle. During the formation operation, if there is a behavior conflict or overlapping operation intention between member vehicles, the RSU intervenes as a third-party mediator and injects a disturbance control variable into the conflict vehicle based on the time period behavior boundary judgment.

[0010] If a formation member accidentally leaves, fails or exits, the system will continue the operation intention, control rhythm behavior trend information of the member that has not been completed within the formation. When the queue structure is disturbed, the system will backtrack to a similar stable state through the historical state trajectory recorded by the RSU.

[0011] Further, the traffic constraint weight values generated by the RSU generate individual repulsion weight offsets based on the historical path selection behavior of the vehicle, so that each vehicle forms a slightly different trend path. The potential gradient guidance domain has a time displacement characteristic, and the vehicle corrects the forward offset value on the weight map according to its own response delay, realizing synchronous aggregation under asynchronous response.

[0012] Further, the leader of the formation selects according to the space position, and the state change rate caused by the vehicle to other vehicles in the recent 5-15 seconds, forms an influence momentum factor for leader allocation; there is a mutual inhibition mechanism between the multiple influence circles, when the dispatch area boundaries of two leader vehicles coincide more than a threshold, the low stability leader will be suppressed by the boundary first.

[0013] Further, a dispatch residual evaluation is required before the leader gives up the right, the leader vehicle simulates the state difference after the instruction is issued and propagated in the formation, and then decides whether to give up; the influence circle structure supports a granularity compression mode, which merges multiple circles into one circle in a communication bandwidth limited area, and maintains the control closed loop by using a simplified scheduling method.

[0014] Further, when the RSU detects that the conflict behavior of the formation members is caused by inconsistent environmental perception, the environmental perception weight of the two vehicles is dynamically reduced, and the vehicles are guided to drive temporarily relying on scheduling information; the disturbance control variable has a diminishing inheritance, and the disturbance influence is only propagated once in the conflict core vehicle, and is not allowed to be chain conducted in the formation.

[0015] Further, the state trajectory backtracking reconstructs the stable section of the behavior memory path by running the memory fragment reconstruction strategy; if the historical state is not available, the system constructs a synthetic stable state relationship, and based on the current state, the closest historical stable structure is backtracked to replace the missing backtracking trajectory.

[0016] Further, the running memory fragment reconstruction strategy includes a reconstruction of a race track mechanism, generates a pseudo-backtracking response path for multiple candidate historical fragments in parallel, and optimizes the dynamic repair effect on the current disturbance state by simulation, to realize the stable state discrimination and selection of the backtracking path; if the time span of the fragment is not enough to cover the required backtracking period, a dynamic splicing node is extracted from the historical trajectory to construct a multi-section splicing reconstructed trajectory with logical continuity.

[0017] Further, the running memory fragment reconstruction strategy includes a path space elastic mapping relationship, which is used to map the vehicle state in the historical trajectory according to the current road topology and formation arrangement, and perform elastic coordinate mapping through a space deformation function to make the historical state executable under inconsistent structures.

[0018] The above space deformation mapping function is used to dynamically and elastically adjust the coordinate state of the vehicle in the historical trajectory according to the current road topology, formation structure and traffic boundary conditions, to construct a reusable trajectory mapping version under inconsistent structures:

[0019]

[0020] Wherein:

[0021] is the mapped historical trajectory coordinate point (corresponding to the coordinate under the current topology); x h ,y h is the single point coordinate in the historical trajectory (in the historical structure reference system); δ x is the relative density difference ratio of the current and historical road topologies in the longitudinal dimension (road length scaling difference); δ y is the structure expansion ratio of the current and historical road topologies in the transverse dimension (lane number or passing width difference); Ω(x h , ρ) is the platoon longitudinal deformation factor function, which adjusts the elastic deformation degree of x h based on the historical longitudinal queue density ρ; Λ(y h , φ) is the road transverse control function, which determines the transverse expansion or compression direction based on the boundary feasible region parameter φ; γ is the global mapping elasticity coefficient, which determines the influence degree of deformation on the final mapping, and the value range is [0, 1];

[0022] This function defines independent deformation functions in the longitudinal and transverse dimensions respectively;

[0023] The introduction of δ x , δ y makes the mapping process respond to the actual geometric structure of the current road network; Ω and Λ are design flexible custom behavior functions, which can be further evolved based on traffic state field, density field or boundary limit; the global elasticity coefficient γ is used to control the conservative or aggressive deformation level, realizing the adjustable flexible mapping strategy;

[0024] The above function derivation process includes:

[0025] The vehicle state point (x h , y h ) in the historical trajectory is established on the basis of the old road and queue geometric structure, while the current environment may have road transverse passing width changes, longitudinal passing compression or queue recombination, etc. If the old coordinates are directly reused, it will cause the trajectory point to fall into the infeasible region or out-of-control state.

[0026] Therefore, the target is to map the historical point (x h , y h ) to the deformation mapping point in the current scene. This mapping point needs to respond to the road geometric changes, traffic structure elastic changes and vehicle layout compression / expansion trends at the same time.

[0027] To achieve this, two key topological difference response coefficients are proposed: the longitudinal difference ratio δ x , which represents the current road in the longitudinal compression or stretching degree, such as the longitudinal compactness trend caused by road section length shortening, merging, queue density increasing; the transverse difference ratio δy represents the change of road width, lane number or traffic feasibility space, such as multi-lane narrowing, road deviation, etc.

[0028] Based on the two scale responses, two functions are further constructed: Omega (x h , rho) for longitudinal deformation adjustment, which describes the deformation sensitivity of the vehicle at x h under different historical queue density rho; and Lambda (y h , phi) represents the lateral structure elasticity under the current lateral passing weight phi, which allows non-uniform lateral deformation to be implemented on vehicles at different y coordinates (such as more compression of the outer lane inward). In order to ensure that the overall mapping behavior is controllable in continuous space, a global deformation adjustment coefficient gamma is introduced, which is in [0, 1], used to adjust the mapping influence weight between behavior rigidity (preserve the original coordinates) and complete elastic reconstruction. Therefore, the above elements are uniformly constructed into the mapping function The function represents: in the longitudinal direction, the historical point x h is adjusted by the topological difference coefficient delta x and the behavior function Omega to produce a structurally reasonable elastic offset; in the lateral direction, the original position y h is updated by the Lambda function after considering the road boundary and feasibility constraints; the two dimensions can be decoupled and independently changed, or can be linked through joint parameters.

[0029] Further, the synthesized stable structure has a multi-granularity decoupling layer, which divides the formation structure into a master group and a response group, and different granularity layers are synchronously advanced according to different recovery strategies: if there is a structural tension between the synthesized stable structure and the current state, the system will insert a negotiation back-off state for stage buffering to a fusible state.

[0030] The platoon driving coordination method based on vehicle-road cooperation provided by the application has the following beneficial effects by introducing a traffic constraint weight guide mechanism, a multi-layer scheduling structure of a leading vehicle, operation memory backtracking reconstruction, a spatial elastic trajectory mapping, a conflict disturbance control and a decoupling structure buffering mechanism:

[0031] The traffic constraint weight value published by the RSU constructs a traffic environment guide domain with potential gradient, and the vehicle can autonomously aggregate and array sort based on the local weight change trend, avoiding relying on centralized control, so as to maintain high freedom and stable cooperative behavior in a variable road scene; when constructing the guide weight, the individual path offset and response delay are considered, and through a time displacement correction mechanism, synchronous aggregation behavior matching is realized under asynchronous sensing conditions, and the consistency of cooperative behavior between vehicles with different computing capabilities is improved.

[0032] The evaluation mechanism of the influence of the vehicle on momentum is introduced, and residual prediction simulation is performed before scheduling, which effectively prevents scheduling error amplification, and when the dominant vehicle is unstable, it can automatically transfer to the secondary control vehicle, ensuring that the scheduling link is not interrupted; a multi-layer influence circle and particle size compression scheduling structure is constructed, and in the communication limited or complex area, the control closed loop is realized through the hierarchical merging scheduling strategy; and dynamic backtracking and reconstruction of the running memory trajectory are supported, which can quickly recover to a stable structure in the case of member disbanding, conflict, abnormal interruption, etc.

[0033] The historical trajectory is subjected to elastic coordinate transformation through a spatial deformation mapping function, which adapts to the change of the current road topology and formation mode, realizes trajectory feasibility inheritance under non-uniform structure, and significantly improves the migration and reuse efficiency of the historical model under dynamic road conditions; when the RSU detects that the behavior conflict originates from inconsistent perception, the vehicle can be guided into a short-time scheduling reliance mode, and the propagation range of the control disturbance variable has a diminishing inheritance, which can block the chain conflict spread and ensure the rapid stability of the formation structure; when the historical trajectory segment is missing or not adaptive, the system can synthesize stable structure relationship and back-propagate the target state, realize reasonable structure transition under no historical condition, and make the system have high autonomous evolution and recovery ability. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 The figure is a vehicle and road cooperative formation function relationship diagram of the present application.

[0035] Figure 2 The figure is an individual difference formation cooperative flow chart of the present application.

[0036] Figure 3 The figure is a racing trajectory reconstruction and hierarchical elastic recovery schematic diagram of the present application.

[0037] Figure 4 The figure is a high-speed formation cooperative control flow chart of embodiment 1 of the present application.

[0038] Figure 5 The figure is a high-speed formation structure disturbance adaptive recovery flow relationship diagram of embodiment 2 of the present application. DETAILED DESCRIPTION

[0039] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0040] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Figure 1As shown, by constructing a vehicle-road collaborative perception guidance mechanism, the autonomous aggregation and arrangement of vehicles in complex road environment are realized. Specifically, the system first relies on the intelligent infrastructure unit deployed on the roadside, i.e. RoadSide Unit (RSU), to continuously release a set of information containing traffic restrictions, path guidance, risk distribution, construction obstacles, temporary traffic events, etc. in the formation driving area in a broadcast or directional projection manner. The set is quantified as a traffic constraint weight value field. The weight value is not a static path selection or control command, but a dynamically changing potential field constructed in the form of a spatially continuous distribution of numerical values, which guides the vehicle to have directional perception and behavior trend judgment of the environment, forming a behavior induction mechanism called traffic environment guidance domain. After receiving the RSU broadcast information, each vehicle participating in the formation combines its own running state (such as position, speed, acceleration) and the local traffic elements detected by the perception system carried by the vehicle to reconstruct the traffic guidance domain on its local computing platform, and real-time perceive the local gradient change trend of the weight in the guidance domain. Through the guidance of the weight gradient, the vehicle can actively judge the potential aggregation path between itself and the target formation center, thereby autonomously generating the control intention of approaching the formation, and adjusting the position in real time based on the target arrangement structure, to realize the self-organizing queue generation and dynamic arrangement process of multiple vehicles without centralized control.

[0041] The formation aggregation process does not rely on fixed path planning or center instruction, but automatically induces vehicles to complete behavior focusing and control convergence through the differential change of weight values in the guidance domain in different regions, so that the entire formation realizes structure convergence and efficiency optimization under the conditions of traffic restrictions, risk perception, and traffic efficiency. This method fully utilizes the environmental insight and broadcast capabilities of RSU in the vehicle-road cooperative system, while respecting the perception differences and local decision-making capabilities of the vehicle end, and realizes low-coupling, high-robustness, and dynamically adaptive formation coordination control without master vehicles.

[0042] Each vehicle dynamically evaluates its influence level in the entire formation structure based on its real-time relative position relationship in the queue, scheduling response efficiency in historical collaborative behavior, and traffic disturbance factors, local constraint levels, risk indexes, etc. influenced by the real-time RSU, and automatically divides multiple levels of behavior influence circles accordingly. Each level of influence circle is an asymmetric control subset containing part of the queue vehicles, and a vehicle with relative behavior stability and state predictability is selected as the current leading vehicle in the circle. The leading vehicle has local scheduling guidance capability in the current influence circle, and can broadcast fine-tuning instructions or structure modification intentions based on the current traffic conditions and collaborative goals to other vehicles in the influence circle, thereby guiding the dynamic convergence of the local form of the formation.

[0043] When the leading vehicle's behavior stability decreases due to sudden interference, abnormal state, path misjudgment or control lag, the system will automatically determine the current evaluation level of the deputy control vehicle based on the preset responsibility transfer weight mechanism, and preferentially transfer the local control right to the deputy control vehicle to ensure the control continuity and behavior consistency within the influence circle. During the formation running, if the system detects that multiple member vehicles have control intention overlap, path intersection convergence or acceleration mutation conflict within a short time window, the RSU as a neutral third party mediator outside the queue will actively intervene in the mediation process. During the mediation process, the RSU does not directly issue control trajectories, but based on the overlap degree of the intersection boundary and time interval of the conflict vehicles on the behavior prediction trajectory, a set of disturbance control variables including micro-amplitude speed offset, temporary behavior freezing delay or slight path bias are injected to create a behavior off-peak window, thereby resolving the conflict situation and restoring the queue running consistency.

[0044] When a vehicle is identified as having a withdrawal trend or has been disconnected, the control rhythm, position correction behavior, coordination strategy with the leading vehicle or deputy control vehicle, and its support relationship to the overall structure of the formation exhibited by the vehicle in the recent period of time will be abstracted by the system as a set of behavior trend information segments. This segment does not simply describe the physical state trajectory of the vehicle, but rather depicts its functional positional role in multi-vehicle interaction, such as buffer node, formation edge vehicle or rhythm tuner. The inheritance mechanism will automatically filter a replacement vehicle with high behavior stability evaluation within its influence circle, and non-explicitly implant the trend information through the vehicle communication middleware, so that the replacement vehicle can absorb the rhythm and position vacancy compensation strategy of the withdrawn vehicle without changing its own control mainline, thereby maintaining the overall formation continuity and behavior coordination. At the same time, when the above structural disturbance has expanded to multiple places in the queue and affected the overall stability, the RSU dispatches the state memory module in the historical running trajectory to activate the trajectory backtracking mechanism triggered based on the disturbance level. This mechanism first performs similarity search on the current formation structure and recent historical state, and selects the feasible path segment through the running memory segment reconstruction strategy. In the reconstruction process, the trajectory segment that has shown high stability and high coordination efficiency under the same road topology and traffic conditions is preferentially selected for behavior mapping, while the path structure that has been invalidated or cannot be reused due to structural asymmetry is excluded. If the historical state cannot be directly reused, the system will construct a synthetic stable state relationship based on the current environmental state and vehicle distribution.

[0045] The accompanying drawings are incorporated in and constitute a part of this specification. Figure 2As shown, when the RSU generates the traffic constraint weight value for constructing the traffic guidance domain, it does not uniformly provide a set of weight distribution fields for all vehicles, but based on the path selection habits, local detour probability, delay response characteristics and other behavior data of each vehicle in the historical task process, calculates the individual path convergence index of each vehicle in a specific environment, and derives a perturbation vector called repulsion weight offset from it. The offset is dynamically superimposed on the standard traffic weight field, so that each vehicle generates slightly different guidance directions when constructing its local potential guidance domain due to the offset, thus naturally forming path trend micro-differences, avoiding the aggregation conflict caused by the congestion of multiple vehicles to the same trend path in the free state, and improving the distribution stability and local avoidance ability at the initial stage of queue convergence. At the same time, considering the problems of different control execution cycles, sensor processing delays, communication relay consumption and other problems in actual driving of vehicles.

[0046] The system further introduces the time displacement characteristic to the potential gradient guidance domain, that is, when constructing the local weight map of the vehicle, the system estimates the response lag between the current time and the ideal synchronous state according to the delay estimation model of the vehicle, and corrects the target attraction path on the guidance weight map according to the lag, so that the vehicles with slower response automatically avoid the future conflict area in advance, and the vehicles with faster response correct the control output based on the more timely weight trend, and thus realize the synchronous queue aggregation effect under the asynchronous information processing condition in the overall control behavior. The above mechanism does not depend on centralized control or high-precision map matching, but through local behavior heterogeneity modeling and weight gradient microstructure guidance, it constructs a cooperative fusion path regulation mode compatible with individual differences, which significantly improves the order perception, self-consistent response and control convergence performance of large-scale heterogeneous vehicle systems at the initial stage of formation aggregation.

[0047] When selecting the leading vehicle in each local influence circle, the system does not simply use the geometric center position or physical leading position of the vehicle in the queue as the priority standard, but considers two key elements: one is the spatial position factor, which is used to determine the topological centering degree and scheduling accessibility of the vehicle in the current influence circle structure; the second is the behavior influence momentum factor, which models the change rate of the vehicle to the state of other vehicles within its influence circle in the past 5 to 15 seconds, including but not limited to the speed fluctuation of neighboring vehicles caused by acceleration adjustment, the response of the formation structure caused by path micro-shift, and the convergence trend degree of the guided decision path. The above behavior influence effects are integrated by time weighting to generate an influence momentum vector, which is used as the basis for sorting the leading weight candidates, so as to select the vehicle with the strongest behavior radiation and the most optimal response consistency in the current structure as the leading vehicle, ensuring that the leading role not only has a position advantage, but also has a behavior inertia basis for scheduling dominance;

[0048] In addition, to prevent scheduling conflicts between multi-layer influence circles caused by area expansion or movement, the system constructs a boundary interaction awareness module between the influence boundaries of each lead vehicle. When detecting that the scheduling areas of any two lead vehicles are in boundary contact in the physical space and there is more than a set overlap threshold (such as more than 1.5 times the vehicle half-width or more than 25% of the influence radius), the system automatically compares the state stability indicators of the two in the last period, including their own path disturbance standard deviation, controlled vehicle response consistency coefficient, and temporary behavior failure times, etc. After comparison and calculation by the suppression function, the lead vehicle with lower stability is marked as a boundary suppression object, and its scheduling instruction weight is preferentially weakened and its expansion range is limited, so as to ensure that the boundary area is maintained by the lead vehicle with more scheduling reliability.

[0049] When the current lead vehicle triggers a state exception or a lead permission decay boundary, it does not immediately release the scheduling right, but constructs a virtual control loop of pseudo-state instruction injection-state propagation-difference feedback locally through the scheduling simulation module. This process generates a set of standard fine-tuning instructions (such as formation compression, speed normalization, or path micro-offset) and simulates their state response when propagated to each affected circle member vehicle under the current queue topology. The system records the residual error of each node state offset and expected behavior to form a scheduling residual function. When the maximum fluctuation amplitude and average coordination error of this residual function within a unit time are below the tolerable boundary (such as speed offset <0.15 m / s, path error <0.2 m), it is determined that the current lead vehicle still has the ability to stably propagate instructions in its permission decay state, and thus can safely transfer the lead right to the vice-controlled vehicle. If the evaluation residual exceeds the expectation, the switching is temporarily suspended, the original lead vehicle is maintained, and an RSU auxiliary stability mechanism is introduced to avoid behavior decoupling caused by short-time empty instruction;

[0050] At the same time, to improve the adaptability of the method in the limited communication bandwidth scenario, the system introduces a granularity compression mode to the influence circle structure. When the vehicle communication bandwidth in the RSU awareness area is lower than the threshold required for normal operation of formation control, the circle layer merging mechanism is automatically triggered, multiple high-frequency low-coupling fine-grained influence circles are compressed into a single large circle structure according to the lead vehicle behavior similarity and vehicle distance, and the fine-tuning instruction broadcast mechanism is temporarily abandoned, and a simplified scheduling mode based on rule set driving is adopted instead.

[0051] When the RSU discovers through data reporting and behavior model monitoring between queue members that there are risk characteristics such as control intention conflict, path convergence or inconsistent braking, the system first discriminates the trigger factor of the conflict behavior, and if the analysis result shows that the conflict is not caused by behavior negotiation failure or scheduling instruction conflict, but because each vehicle has inconsistent perception cognition of the current traffic environment (such as obstacle boundary, dynamic vehicle identification, passable area division, etc.), it is determined that the conflict is caused by inconsistent environment perception; At this time, the RSU will adjust the local decision weight structure of the vehicle involved, by issuing environment perception weight correction parameters, dynamically reducing the degree of decision dominance of the vehicle to the local perception result, and synchronously improving its dependence coefficient on scheduling information (such as leader vehicle fine tuning command, RSU path constraint guidance), guiding it to complete path correction and speed adjustment according to external guidance in a short time window, and then gradually restoring its perception weight after the conflict is resolved, realizing a dynamic perception trust mechanism of weakening self-consciousness and strengthening cooperation and rebalancing, and then solving the behavior divergence risk caused by perception heterogeneity;

[0052] In addition, in order to prevent the disturbance control variables introduced in the mediation process from continuously spreading in the queue, causing structural shock or feedback enhancement, the application also introduces a disturbance variable reduction inheritance mechanism, that is, after the RSU injects a disturbance control variable (such as a speed disturbance vector, a path offset factor, an instruction freeze identifier) into the conflict core vehicle, it sets the non-inheritable attribute for the disturbance marker at the same time, ensuring that the disturbance instruction only acts on the current target vehicle and cannot be forwarded or implicitly conducted to subsequent vehicles by the vehicle in path propagation; The system adds a disturbance instruction propagation identifier field in the vehicle-to-vehicle communication protocol.

[0053] The application designs a trajectory backtracking recovery mode combining running memory segment reconstruction strategy and synthetic stable state construction mechanism, wherein the state trajectory backtracking does not depend on full trajectory playback or template matching, but is intelligently recovered based on a segmented behavior memory path management mechanism, that is, the system periodically divides the behavior state sequence formed by the vehicle formation into a plurality of reusable running memory segments in the normal running state, and marks the trajectory segments with consistent behavior trend, small disturbance amplitude and stable internal interaction as stable segments through a clustering function. When the queue structure is disturbed or the overall instability trend occurs, the system searches for a high similarity matching pair between the current state characteristics and the historical stable segments from the historical memory path, and selects the trajectory segment with the highest matching degree as the recovery anchor point, which is used as the current formation target state for behavior traction, thereby guiding the queue to return to the stable mode.

[0054] If the historical trajectory is invalid or the segment is unavailable due to environmental mutations, route switching or communication interruption, etc., the system will start the synthesis stable state construction process. In this process, the system first constructs the current behavior feature map based on the position, speed, action chain relationship and behavior inertia of the remaining leading vehicles of the current platoon vehicles, then combines the key feature indicators of the historical stable structure such as vehicle spacing distribution, leading area boundary structure and behavior coordination parameter to build multi-dimensional constraint conditions, and takes the current state as the starting point to iteratively approach the virtual stable state that is most likely to be achieved through a set of reverse state reasoning functions. The synthesized state is not a reproduction of the historical real trajectory, but a substitute target that is closest to the historical stable segment in behavior structure. RSU guides each member vehicle to gradually approach this target structure.

[0055] In combination with the accompanying Figure 3 As shown, the system embeds a race trajectory reconstruction mechanism with parallel discrimination and dynamic repair optimization capability in the running memory segment reconstruction strategy. Instead of relying solely on single trajectory matching or static optimal path selection, the mechanism calls candidate historical stable segments within multiple similarity thresholds to generate a set of pseudo-backtracking response paths in a parallel simulation manner. The disturbed state of the current vehicle platoon is injected into each backtracking path as the initial input. During the simulation process, the contribution value of the platoon core structure recovery, the dissipation time of the control disturbance, the path convergence gradient trend and the recovery integrity of the leading behavior are evaluated in real time, and a set of stability scoring functions are constructed. Finally, based on the scoring function, the most efficient target path is determined as the backtracking main track under the current disturbance context, which significantly improves the discrimination accuracy of trajectory reconstruction in atypical disturbance scenarios.

[0056] In addition, considering that some historical segments may have insufficient coverage length due to limited recording time span, short event period or high-frequency micro-disturbance cutting, the system further introduces a dynamic splicing node extraction mechanism. Under this mechanism, the system first analyzes the time coverage interval of the selected candidate segment and the time demand difference of the current disturbance recovery period. If it is found that there is a coverage deficiency, it automatically finds a trajectory node in the historical trajectory library that has state compatibility, behavior trend continuity and consistent scheduling rhythm with the current segment end state, and uses this node as a connection anchor to append matching trajectory segments from its starting position, realizing multi-segment trajectory splicing reconstruction with logical continuity. The splicing process is phase-aligned and acceleration curvature-smoothed through the behavior stable transition function between adjacent segments, ensuring that the spliced path has consistency and physical executability in control output.

[0057] In the vehicle-road cooperation-based platoon driving coordination method, in order to solve the problem of insufficient reusability of historical trajectories under the condition of changes of current road structure or queue configuration, a path space elastic mapping relationship is introduced in the running memory segment reconstruction strategy, the core of which is to realize the elastic reconstruction of the coordinates of the vehicles in the historical trajectories by constructing a kind of space deformation function which can dynamically respond to the differences in road topology, changes in vehicle density and adjustment of lateral passing boundary, so that the control feasibility and path constraint consistency are maintained under the condition of inconsistent traffic structure.

[0058] Specifically, the system adopts the following form of space elastic mapping function:

[0059]

[0060] Wherein, represents the mapped trajectory point, that is, the coordinate correction result of the historical state (x h ,y h ) after considering the road topology and structure constraints; δ x is the compression rate or stretching rate of the current road longitudinal length or queue tightness relative to the historical state, δ y is the expansion coefficient of the current road lateral passing structure (such as the number of lanes, passing bandwidth) compared with the historical state; Ω (x h , ρ) is a longitudinal behavior deformation function, the value of which is dynamically calculated according to the queue density ρ at the historical time and the x-direction position change characteristics, and is used to adjust the coordinate stretching amplitude of the longitudinal position under the condition of dense platoon; Λ (y h , φ) is a transverse deformation function, which is used to express the adjustment priority direction and deformation level of the vehicle transverse position under the current road boundary condition φ (such as the available channel width or lane compression strategy);

[0061] In addition, in order to avoid the deformation process being too aggressive or completely rigid, a global elastic adjustment coefficient γ ∈ [0, 1] is introduced, which determines the influence degree of the deformation operation on the final mapping result. When γ tends to 0, the mapping behavior tends to be conservative and retains the historical state, and when γ tends to 1, the mapping process has the maximum elasticity and fully responds to the current structure adjustment demand, so that the mapping strength can be flexibly selected according to different traffic strategies; in actual application, the deformation mapping function can be independently controlled in the longitudinal dimension and the transverse dimension, so that for complex situations such as the platoon entering a narrow curve, the queue reconstruction before fusion, or the mutation of the passing topology, the system can still map the historical running trajectory to the feasible coordinate space under the new geometric conditions and maintain the consistency of the behavior intention and the historical trajectory trend; especially when multiple trajectory segments are spliced or the behavior of the failed member is inherited, the elastic mapping mechanism enables the trajectory segment to have the ability to adapt to the current road environment in structure. ​

[0062] When the system constructs the target stable structure through trajectory backtracking or state reasoning, the system does not directly project it as a whole rigid target to the current actual state, but rather performs granular hierarchical analysis to construct a multi-level structure mapping framework composed of a master group and a response group. The master group includes core vehicle units that currently have dominant scheduling ability, high structural stability, or significant behavior inertia, and mainly undertakes the tasks of reconstruction anchoring and behavior guidance of the structure framework. The response group includes flexible vehicle groups that are attached to the master group structure changes, and their behavior targets are derived from the master group evolution behavior and adapted for execution. Each granular decoupling layer independently advances according to different recovery strategies, with the upper master structure being prioritized to achieve stability, and the subsequent response layers being dynamically coordinated for recovery according to their structural inertia and control time lag. Thus, a stable propagation strategy from the core to the edge is formed. In addition, when there is structural tension between the current actual state of the formation and the synthesized stable structure, i.e., there is significant inconsistency or fusion risk between the current vehicle distribution, speed field, or interaction relationship and the target structure, the system does not directly force structure adjustment, but actively inserts a negotiation back-off state in the structure evolution path. This back-off state is not part of the ultimate stable target, but a stage intermediate state constructed by the system according to the tension degree. The design purpose is to buffer structural tension and guide vehicles to tend towards the target structure within a limited time domain. This state has transitionality, adjustability, and local behavior consistency. During this phase, the master group will weaken rigid scheduling, and the response group will gradually narrow the difference with the target structure in a soft coordination manner. Only when the structural tension converges to the system's allowable threshold will the next decoupling layer switching be started, so that the synthesized structure has a flexible recovery rhythm in time sequence, maximally avoiding chain control conflicts and stability collapse caused by drastic adjustment, and ensuring that the formation system still has self-adaptive reconstruction ability and high fault-tolerant coordination performance in scenarios of structural discontinuity, incomplete information, or strong vehicle heterogeneity.

[0063] Embodiment 1:

[0064] In combination with the accompanying Figure 4 As shown, a certain highway section is located in the east section of the G95 intelligent test section, and there is a vehicle-road cooperation area controlled by RSU-23. The current system detects that there are 6 autonomous vehicles (A1 to A6) gradually approaching the same section in different lanes, and the system expects to form a three-layer parallel dynamic formation (two columns and three rows) to adapt to the formation compression requirements of the front split area. RSU-23 first constructs a path tendency model according to the historical path selection behavior of each vehicle in the last 20 seconds, and calculates the individual repulsion offset value through the behavior consistency coefficient η. The basic constraint weight graph is set as The post-tendency path weight graph of each vehicle is where Δ iReflecting the vehicle's deviation intentions based on past path overlap and trajectory differences, such as the high overlap of historical paths A1 and A2, the RSU introduces rejection offsets of +0.05 and -0.03 respectively to achieve slight path diversion while maintaining macro-level aggregation. This effectively prevents traffic bottlenecks caused by high-density path overlap. Subsequently, RSU-23 constructs a potential gradient guidance domain Φ(x,y,t), which forms a guidance weight fluctuation value from low to high on the map, and vehicles will naturally tend to the low weight trough area. However, due to the difference in vehicle execution response delays (e.g., A3 control delay is 100ms, A5 delay reaches 350ms), different degrees of forward deviation occur in the actual execution path. To address this, the system introduces a time displacement correction mechanism, which predicts the delay time τ of each vehicle. i The forward displacement δ of Φ(x,y,t) is pre-emptively increased. i =v i ·τ i (where v) i (For vehicle speed), a potential reprojection region is formed, so that even though the vehicle responds asynchronously, it can complete synchronous aggregation in the corrected guidance map; for example, at a vehicle speed of 21m / s and a delay of 100ms, δ3=2.1m, the actual guidance target is pushed forward by 2.1m, so that its response behavior is aligned with the ideal convergence rhythm.

[0065] After initial aggregation, the system enters the dominant vehicle selection phase. The RSU calculates in real time the rate of change of formation state caused by each vehicle within the past 10 seconds (i.e., the number of changes in acceleration, deceleration, or lane-changing behavior of other vehicles), and constructs the influencing momentum factor. Where κ is the dominance adjustment factor (empirically set to 0.8). If A4 triggers two active follow adjustments of A3 and A5 within this cycle without any braking conflict, its μ4 is the largest, and therefore it is set as the first-level dominant vehicle.

[0066] Subsequently, the RSU constructs a three-tiered influence circle: the first-level circle is centered on A4 with an influence radius of 15m; the second-level circle covers A2, A3, and A5, with a radius extending to 30m; the third-level circle is set as a follow-up buffer layer, including A1 and A6. During system advancement, if another segment merges into another micro-formation (dominated by A7) and overlaps with the scheduling boundary of the A4 formation, the system determines the area of ​​the intersection of the two dominant circles. The ratio of the total influence area to the total influence area exceeds a set threshold (e.g.) When the boundary suppression mechanism is triggered, the system assesses that the momentum factor of A7 is μ7 = 0.43, which is lower than μ4 = 0.71. Therefore, the dominant behavior of A7 is suppressed first, and the RSU uniformly dispatches all vehicles in the area to prevent path oscillation caused by dual dominant conflict.

[0067] In the aforementioned G95 high-speed platoon scenario, vehicle A4 assumes the primary leading role and controls A2, A3, and A5 within a 15-m influence circle. RSU-23 detects that A4 has frequent longitudinal control command failures within the next 4 seconds, including two delayed speed commands and one path jitter. The system determines that the stability of A4 has fallen below the safety threshold, triggering the scheduling residual evaluation mechanism. RSU issues a simulation command: assuming that A4 issues a speed synchronization command v = 25 m / s, the system predicts the offset AS of each vehicle after execution based on the inertia, position deviation, and delay coefficient of each vehicle as follows:

[0068] A2: 24.7 m / s, AS = +0.3 m;

[0069] A3: 25.1 m / s, AS = -0.2 m;

[0070] A5: 23.9 m / s, AS = +1.1 m.

[0071] According to the standard residual threshold AS_max set to 0.8 m, any vehicle response deviation exceeding the threshold is considered to have excessive residuals and is not suitable for continuing to lead. A5's response deviation exceeds 1.0 m in the simulation execution, so the system rejects A4 from continuing to assume the leading role, and the leading role is transferred to the secondary control vehicle A2.

[0072] At the same time, since RSU detects that the current area is in the 5G base station switching band, the communication bandwidth is reduced from the normal 20 Mbps to 4 Mbps. To avoid congestion caused by too fine scheduling information granularity, RSU automatically triggers the granularity compression mode, which compresses the original three-layer influence circle into a primary macro circle, adopts a normalized scheduling strategy (e.g., a unified target acceleration of 0.8 m / s 2 , path offset limit ±1 m), maintains the scheduling closed loop, and avoids control out-of-sync.

[0073] Subsequently, when traveling to the front curve section, vehicles A3 and A5 have a control intention conflict during lane changing. A3 detects a right lane change, while A5 determines to go straight, blocking each other's path. RSU discovers through lateral radar feedback that the two vehicles' front sensor information is inconsistent, determining that the conflict is caused by inconsistent environmental perception. At this time, the system executes the environmental perception weight weakening strategy: reducing the environmental perception factor w e of A3 and A5 from 0.9 to 0.3, making the two vehicles temporarily rely on RSU scheduling information to get rid of self-perception path paranoia and complete dynamic avoidance. To quickly break the conflict, RSU injects a disturbance variable ε = -1.2 m / s 2 as a braking disturbance into A3. The system identifies that this disturbance is only applicable to the conflict-causing vehicle and does not propagate to the rear A6 or the front A2, thereby executing the disturbance control variable inheritance reduction strategy: the disturbance only acts once and does not chain spread, avoiding the spread of a single vehicle conflict to the entire platoon and causing platoon oscillation.

[0074] Example 2:

[0075] Combined with appendix Figure 5 As shown, on a certain section of the G15 Hangzhou-Ningbo Expressway at kilometer S57, the formation Alpha deviated from its preset path due to a road closure ahead. Vehicle B3, misled by a LiDAR warning that the exit was passable, performed a unilateral deceleration maneuver, causing the distance between following vehicles B4 and B5 to drastically decrease to 3.2 meters (the original stable distance was 8 meters), triggering a local structural disturbance warning. RSU-42 took over the area dispatching task and initiated the state trajectory backtracking mechanism. First, it called the runtime memory management module to retrieve the behavior segments of the formation containing B3 over the past 60 seconds, identifying three stable segments of the historical trajectory path: P1, P2, and P3. These correspond to the previous ramp merging segment (P1), the steady-state cruise segment (P2), and the obstacle avoidance response segment (P3), respectively. Each segment lasted approximately 15 seconds, with an average speed fluctuation σ. v <0.4m / s, average acceleration drift Δa<0.3m / s 2 The system uses the residual simulation module to perform parallel playback simulations of these three historical segments, taking states B3-B5 as the initial disturbance conditions, and simulates the collaborative adjustment cost function J(·) after trajectory access, obtaining the following backtracking adaptation score:

[0076] P1: J1 = 3.2, an initial offset of +0.9m occurs, and the tail disturbance is transmitted to the 6th car;

[0077] P2: J2 = 1.4, initial offset controlled within ±0.4m, convergence within 1.2 seconds;

[0078] P3: J3 = 2.6, indicating a forward compression anomaly that requires temporary braking command correction.

[0079] According to J min The system selects a strategy and determines P2 as the optimal backtracking path, mapping the dominant vehicle's control rhythm and relative vehicle positions in this trajectory to the current formation structure. Since P2's historical trajectory only covers 15 seconds, while the current estimated recovery window is 22 seconds, the system executes dynamic stitching logic, stitching the subsequent 5 seconds of P2's behavior structure with the preceding 2 seconds of P1's behavior, at dynamic node D. p The splicing is completed at the point where the trajectory continuity error does not exceed 0.6m and the phase difference of the dominant rhythm is less than 0.3 seconds.

[0080] If the system cannot retrieve a matching trajectory with sufficient structure from the historical segment library (e.g. due to a construction section changing the road topology from two lanes to one lane, causing the historical trajectory structure to mismatch), the system enters the synthetic stable structure reasoning process. Based on the current vehicle position set X = {x1 = 15m, x2 = 21m, x3 = 29m} and velocity set V = {v1 = 24.3, v2 = 24.0, v3 = 25.1}, the RSU computes the behavior tension function T through the behavior coordination index matrix M c (defined as the coupling index of the relative velocity difference ratio between vehicles and the distance compression ratio) :

[0081] T = max(|v i -v j | / d ij ), where d ij is the inter-vehicle distance. The current T = 0.83 is in a highly disturbed state (reference threshold T stable = 0.25), the system constructs the closest historical stable configuration in a three-vehicle platoon from the stable structure template database, targeting {Δx = 7.8m, Δv < 0.3m / s}, and constructs a synthetic stable structure state S based on this. The RSU schedules each vehicle to fine-tune and gradually converge to the position sequence defined by S at the target velocity gradient Δv g = {-0.1, 0, +0.1}, and finally completes the transition to the approximate stable structure in 3.6 seconds, with a path convergence residual less than 0.4m.

[0082] In this embodiment, the high-speed section, the platoon Beta historically cruises in a two-lane structure, with a three-vehicle queue A1, A2, A3 arranged in a line, with a longitudinal distance of 10m between the three vehicles under the historical structure, and a lane width of 3.75m. The historical trajectory point of A1 is (x h = 130m, yh = 1.2m), the historical trajectory point of A2 is (x h = 120m, y h = 1.2m), and the historical trajectory point of A3 is (x h = 110m, y h = 1.2m). Currently, due to construction compression, the road section has changed to a single lane, with an allowable width of 2.9m, and the platoon needs to be compressed from parallel to vertical and overall to the right side. At this time, the system calls the historical trajectory and prepares to perform elastic mapping through the spatial deformation function .

[0083] First, the longitudinal density changes: the historical two-lane length of 300m is arranged in parallel with three vehicles, and now the same length needs to be arranged with three vehicles in a single lane, with a theoretical longitudinal compression ratio δ x = (original distance) / (current target distance) = 10 / 7 = 1.428, and the target distance is adjusted to 7m. The lateral structure changes: δ y= (current width) / (original width) = 2.9 / 3.75 = 0.773. The original longitudinal density of the queue p is 0.03 vehicles / m (3 vehicles / 100m), and after adjustment, it is 0.043 vehicles / m. Define Ω(x h , p) = 2 * log (1 + p) * (x h - 100) (100 is the reference point, and after unitization, it is anti-overflow), as A1, Ω(130, 0.043) = 2 * log (1 + 0.043) * 30 ≈ 2 * 0.042 * 30 ≈ 2.52; define the lateral elasticity function Λ(y h , φ) = -0.8y h * (1 - φ), where φ = δ y .

[0084] Λ(1.2, 0.773) of A1 = -0.8 * 1.2 * (1 - 0.773) = -0.8 * 1.2 * 0.227 ≈ -0.218. The global elasticity coefficient γ is set to 0.8 (because the structure changes dramatically, high flexibility is required), and the longitudinal and lateral coordinate mapping of A1 is:

[0085] x A1,new = 130 + 0.8 * 1.428 * 2.52 ≈ 130 + 2.88 ≈ 132.88 (unit: m)

[0086] y A1,new = 1.2 + 0.8 * 0.773 * (-0.218) ≈ 1.2 - 0.135 ≈ 1.065 (unit: m)

[0087] Similarly, A2 and A3 are substituted respectively, and the target position coordinates of the three vehicles under the new topology are obtained, realizing the continuous mapping of the historical trajectory to the new road conditions. In this way, the historical formation is smoothly compressed and moved to the right, which can be directly used as the transition control target of the formation, solving the problem of unfeasible trajectory caused by environmental structure change.

[0088] To ensure that the structure convergence does not cause local impact, the system also uses multi-granularity decoupling recovery. In the above lane change compression case, if the state of A1 (master control group) coincides with the target error less than 0.3m, the master control group locks the recovery strategy first, and A2 and A3 (response groups) adjust to the target position according to the master control group trajectory. If it is found that the distance between A2 and the target is greater than 1m (structure tension zone), the system determines that there is fusion tension, and at this time, the target trajectory is not directly pushed, but a negotiation retreat state is inserted, i.e. A2 is first transitioned to an intermediate state 0.5m away from the target point, and then A3 is converged before being pushed synchronously, so as to reduce the overall tension by using two-step buffering, and eliminate the acceleration peak value induced by compression, so as to ensure that the multi-vehicle structure is compressed without shaking. Simulation shows that under this strategy, the maximum braking peak value of the three vehicles in the compression process is reduced from 2.5m / s 2 to 1.1m / s2 The convergence time is reduced from 8.2 seconds to 6.9 seconds, and the maximum lateral offset error between vehicles is less than 0.18 m, fully meeting the requirements of platoon safety and efficiency.

[0089] In summary, the recovery control process using the spatial deformation mapping function combined with the multi-granularity decoupling layer can realize high-quality elastic reuse of historical trajectories and dynamic buffering of platoon structure tension under complex working conditions such as road topological mutation and queue structure adjustment.

[0090] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above examples, and the above examples and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A platooning coordination method based on vehicle-road cooperation, characterized in that... include: The roadside unit (RSU) releases traffic constraint weight values ​​containing information such as traffic restrictions, path guidance, and risk distribution to the platooning area. These weight values ​​form a traffic environment guidance domain with a potential gradient at the vehicle end. Based on its own state and the weight change trend in the local perception environment, the vehicle autonomously completes platooning aggregation and position arrangement. After the formation is completed, the vehicles dynamically form a multi-level influence circle structure based on their relative position, historical cooperative behavior and the influence factors provided by the RSU. Each layer of influence circle contains the current leading vehicle. The leading vehicle has local scheduling and guidance capabilities, and when the leading vehicle is unstable or behaves abnormally, control is transferred to the secondary vehicle. During platooning operations, if there are behavioral conflicts or overlapping operational intentions among member vehicles, the RSU will intervene as a third-party mediator and inject disturbance control variables into the conflicting vehicles based on the time period behavior boundary. If a member of the formation unexpectedly leaves, malfunctions, or exits, the system will continue to inherit the member's unfinished operational intentions and control rhythm behavior trends within the formation; and when the queue structure is disturbed, it will backtrack to a similar stable state through the historical state trajectory recorded by RSU.

2. The platooning coordination method based on vehicle-road cooperation according to claim 1, characterized in that... The traffic constraint weight values ​​generated by the RSU are based on the vehicle's historical path selection behavior to generate individual rejection weight offsets, so that each vehicle forms a slightly different trend path; the potential gradient guidance domain has time displacement characteristics, and the vehicle corrects the forward offset value on the weight map according to its own response delay to achieve synchronous aggregation under asynchronous response.

3. The platooning coordination method based on vehicle-road cooperation according to claim 1, characterized in that... The selection of the leading vehicle in the formation is based on its spatial location and the rate of state change it causes to other vehicles in the past 5-15 seconds, forming an influence momentum factor for the allocation of dominance. There is a mutual inhibition mechanism between the multiple layers of influence circles. When the boundaries of the scheduling areas of two leading vehicles overlap by more than a threshold, the low-stability dominance will be suppressed by the boundary first.

4. The platooning coordination method based on vehicle-road cooperation according to claim 3, characterized in that... Before the transfer of control, a scheduling residual assessment is required. The state difference after the command issued by the leading vehicle is propagated in the formation is simulated, and then a decision is made on whether to transfer control. The influence circle structure supports a granular compression mode, which merges multiple circles into one circle in areas with limited communication bandwidth and adopts a simplified scheduling method to maintain the control closed loop.

5. The platooning coordination method based on vehicle-road cooperation according to claim 1, characterized in that... When the RSU detects that the conflict behavior of the formation members is caused by inconsistent environmental perception, it dynamically reduces the environmental perception weight of the two vehicles and guides them to drive with a short-term reliance on scheduling information; the disturbance control variable has the property of reduction inheritance, and the disturbance effect only propagates once to the core vehicle of the conflict and is not allowed to be chained by the formation.

6. The platooning coordination method based on vehicle-road cooperation according to claim 1, characterized in that... The state trajectory backtracking selects a stable segment reconstruction and recovery mode in the behavioral memory path through a memory fragment reconstruction strategy; if the historical state is unavailable, the system constructs a synthetic stable state relationship, and backtracks based on the current state to find the closest historical stable structure to replace the missing backtracking trajectory.

7. The platooning coordination method based on vehicle-road cooperation according to claim 6, characterized in that... The operational memory fragment reconstruction strategy includes a reconstructed competitive trajectory mechanism, which generates a pseudo-backtracking response path in parallel for multiple candidate historical fragments, and optimizes the dynamic repair effect of the current disturbance state through simulation to achieve steady-state discrimination and selection of the backtracking path; if the fragment time span is insufficient to cover the required backtracking period, dynamic splicing nodes are extracted from the historical trajectory to construct a logically continuous multi-segment spliced ​​reconstruction trajectory.

8. The platooning coordination method based on vehicle-road cooperation according to claim 7, characterized in that... The runtime memory fragment reconstruction strategy includes a path space elastic mapping relationship, which is used to perform elastic coordinate mapping of vehicle states in historical trajectories according to the current road topology and platooning arrangement through a spatial deformation function, so as to make historical states executable under non-uniform structures.

9. The platooning coordination method based on vehicle-road cooperation according to claim 8, characterized in that... The synthesized stable structure has a multi-granularity decoupling layer, which divides the formation structure into a master control group and a response group. Different granularity layers advance synchronously according to different recovery strategies: if there is structural tension between the synthesized stable structure and the current state, the system will insert a negotiated backoff state to buffer the process to a mergeable state in stages.

Images