Unmanned vehicle group cooperation method based on trajectory prediction

Abstract

The invention belongs to the field of unmanned driving, and provides an unmanned vehicle group cooperation method based on trajectory prediction. The invention provides an unmanned vehicle group stability dynamic evolution method, so that an unmanned vehicle group is kept stable in four dynamic evolution events of node joining, node leaving, vehicle group merging and vehicle group splitting; on the basis, modeling is carried out on the state in the vehicle group evolution process, an unmanned vehicle group cooperation model is constructed, and an unmanned vehicle group cooperation method is provided. According to the method, the influence of the change of the internal structure of the vehicle group caused by surrounding frequently-changing interference factors can be reduced, so that the stability and continuity of the vehicle group are improved, and a model and a method are provided for the intelligent cooperation of the movement behaviors of the unmanned vehicle group; therefore, the problems of acceleration, lane changing, decision making, traffic jam and the like of unmanned driving in an unmanned driving scene at present are effectively solved, and the method has certain theoretical significance and application value for promoting healthy and rapid development of unmanned vehicle groups.

Description

Technical Field

[0001] The present invention relates to the field of unmanned driving, and in particular to a method for coordinating unmanned driving vehicle groups based on trajectory prediction. Background Art

[0002] Unmanned vehicle swarms can effectively solve the perception limitation problem of a single intelligent agent, thereby improving the safety of decision-making. At present, there are two main types of unmanned vehicle swarm collaboration methods: (1) traditional unmanned vehicle swarm collaboration; (2) unmanned vehicle swarm collaboration based on swarm intelligence.

[0003] (1) Traditional autonomous vehicle swarm collaboration

[0004] Traditional unmanned vehicle group coordination is based on the dynamic relationship between vehicles to form a formation, optimize member movement to reduce energy consumption and increase traffic throughput. Jalil et al. formed an unmanned vehicle group to enter the roundabout, and improved the efficiency of the roundabout by adjusting the speed of members. Bakibillah et al. divided the roundabout area into an unmanned vehicle group formation area and an unmanned vehicle group dynamic evolution area, and then judged whether a vehicle group could be formed or evolved in the area based on the control center to improve the efficiency of the roundabout. Kamal et al. proposed a formation strategy for unmanned vehicle groups to pass through the roundabout based on the traditional method. This strategy can effectively reduce the time spent passing the roundabout by forming scattered unmanned vehicles into a regular unmanned vehicle group to pass the roundabout. Maiti et al. explored two vehicle group formation methods based on greed and destination. The simulation experimental results show that the greedy strategy-based method has a higher vehicle group formation efficiency, but requires more frequent reconstruction, while the destination-based method requires a relatively long formation time, but does not need to frequently reorganize the vehicle group after the formation of the vehicle group.

[0005] (2) Unmanned vehicle swarm collaboration based on swarm intelligence

[0006] Unmanned vehicle swarm based on swarm intelligence refers to a group of autonomous and cooperative unmanned vehicles, which achieve common goals through collective wisdom and collaborative strategies, such as improving the ability of interconnection, traffic efficiency, reducing energy consumption or enhancing the safety of the group. Existing research on the collaboration of unmanned vehicle swarms mainly focuses on clustering vehicles in the road network in specific scenarios to form unmanned vehicle swarms. Due to the interference of external environmental factors, the network topology of the vehicle swarm may change, resulting in dynamic evolution within the vehicle swarm. Some studies have completed mathematical modeling of vehicle swarm formation, dynamic evolution, and vehicle swarm collaboration, aiming to better guide the unmanned vehicle swarm to achieve interconnection and collaboration in specific scenarios. In view of the limitations of existing unmanned vehicle swarm research, Ju et al. proposed a vehicle swarm dynamic evolution method for highway scenarios. The method first defines the vehicle state and proposes a vehicle swarm formation method based on the vehicle state. Then, the unmanned vehicle swarm model is established, and the meta-vehicle swarm sequence is introduced for timing management. Finally, a dynamic evolution detection algorithm is proposed. The simulation results show that the algorithm can effectively predict the dynamic evolution of unmanned vehicle swarms. Chen et al. proposed a leadership degree for electing a leading node in a closed scenario, and then formed an unmanned vehicle group based on vehicle-to-vehicle distance, location similarity, speed, direction, etc., and then vectorized the unmanned vehicle nodes and vehicle groups respectively through graph embedding, so that the unmanned vehicle group retains the relevant information of the vehicle group nodes during the dynamic evolution process. On this basis, combined with the characteristics of the closed scenario, the evolution events within and between unmanned vehicle groups are analyzed from a microscopic perspective, and a vehicle group self-cooperation model is given in combination with finite state automata. Yu et al. proposed a method for forming an unmanned vehicle group based on edge computing. This method first gave an unmanned vehicle group model based on self-cooperation and real-time performance, and then studied the dynamic evolution law of the unmanned vehicle group based on the changes in the computing relationship between nodes in the vehicle group, and finally gave a vehicle group self-cooperation model. Hou et al. proposed a method for building a safe unmanned vehicle group by introducing a new method for detecting threats and vulnerabilities. This method integrates risk assessment scores to evaluate passenger safety and cargo value, forming a hierarchical vehicle group structure. Simulation results show that compared with the existing unmanned vehicle group model, this method is superior in terms of node survival time, average number of changes in master nodes, and average risk assessment scores. Sun et al. studied the dynamic evolution of unmanned vehicle groups in highway scenarios and proposed a distributed unmanned vehicle group dynamic evolution model. They introduced a distributed vehicle group structure and maintenance method. The model defined 8 vehicle states and transitions and proposed a collaborative unmanned vehicle group dynamic evolution method. Simulation experiments show that compared with the centralized method, this method can maintain vehicle group consistency and enhance the stability of the vehicle group structure.Cheng et al. addressed the challenge of forming a stable swarm of autonomous vehicles in a highway scenario with multiple entrances and exits, and proposed a decentralized formation method based on sidechain consensus to overcome the problems existing in centralized swarms. A distributed formation and maintenance method was given, which showed better performance in terms of computing task balance, information symmetry and stability compared with existing methods. Yuan et al. proposed a model of autonomous vehicle swarms in urban scenarios considering environmental factors. A prediction method was introduced to analyze the influencing factors, and an autonomous vehicle swarm model based on connectivity, coupling and real-time was established. Based on the autonomous vehicle swarm model, Yuan et al. solved the challenge of accurately handling the dynamic evolution events of autonomous vehicle groups in urban scenarios, considering factors such as manned vehicles, obstacles, traffic lights and pedestrians, and proposed a dynamic evolution method, which abstracted into five events according to the dynamic evolution causes, enhanced connectivity, coupling, real-time and evolution in real urban scenarios, and its performance was better than the method designed for highway scenarios. Yuan et al. established a collaborative model of autonomous vehicle swarms for urban scenarios considering complex road networks. The model is based on the non-collision aggregation motion criterion and enhances key properties such as connectivity, coupling, real-time, evolution and adaptability.

[0007] In summary, although there have been a lot of research and methods on the existing unmanned vehicle group collaboration, it still has the following problems:

[0008] 1) The future trajectory of the unmanned vehicle nodes and the neighboring vehicle nodes is not considered: The existing unmanned vehicle group only considers the current motion characteristics such as motion similarity and communication connectivity between vehicle nodes, but lacks the future trajectory information of the vehicle, making it difficult to maintain the continuity of the vehicle group.

[0009] 2) Frequent changes in the surrounding environment can easily cause changes in the internal structure of the swarm: Unmanned vehicle swarms will be affected by external interference, causing changes in the topological structure of the swarm and the relationship between its members, making it difficult to maintain the continuous stability of the swarm’s internal coordination. Summary of the invention

[0010] In response to the above problems, this application provides a collaborative method for unmanned vehicle swarms based on trajectory prediction. First, a dynamic evolution method for the stability of an unmanned vehicle swarm is provided, so that the unmanned vehicle swarm remains stable in four dynamic evolution events: node joining, node leaving, vehicle swarm merging, and vehicle swarm splitting. Then, the state of the vehicle swarm during its evolution is modeled, and a collaborative model for an unmanned vehicle swarm (Trajectory Predictionbased Cooperation Model, TPCM) is provided, and an unmanned vehicle swarm collaborative algorithm is provided. Finally, the stability and sustainability of the vehicle swarm are verified through simulation experiments.

[0011] Compared with the prior art, the present invention has the following beneficial effects:

[0012] The unmanned vehicle group coordination method based on trajectory prediction of the present invention can effectively reduce the impact of external interference of the unmanned vehicle group and maintain the continuous stability of the internal coordination of the unmanned vehicle group, thereby effectively solving the unmanned driving problems such as acceleration, lane changing, decision-making and traffic congestion in the current unmanned driving scenes, and has good theoretical significance and application value for promoting the healthy and rapid development of unmanned vehicle groups. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 Schematic diagram of the node joining process.

[0014] Figure 2 Schematic diagram of the node leaving process.

[0015] Figure 3 Schematic diagram of the merging process of two semi-centralized vehicle groups.

[0016] Figure 4 Schematic diagram of the merging process of two centralized vehicle groups.

[0017] Figure 5 Schematic diagram of the process of merging a semi-centralized vehicle group into a centralized vehicle group.

[0018] Figure 6 Schematic diagram of the collaborative node leaving split process.

[0019] Figure 7 Schematic diagram of the process of leading nodes to leave the split.

[0020] Figure 8 Schematic diagram of state transition of the vehicle group cooperation model.

[0021] Fig. 9 Schematic diagram of data packet sending, receiving and processing flow of vehicle group cooperation method.

[0022] Fig.10 Node survival time diagram

[0023] Fig.11 Node stabilization time diagram

[0024] Fig.12 The curve of average vehicle group trajectory separation change at different speeds.

[0025] Fig.13 The curve of vehicle group trajectory separation changing with time when the maximum vehicle speed is 25km / h.

[0026] Fig.14 The curve of vehicle group trajectory separation changing with time when the maximum vehicle speed is 30km / h.

[0027] Fig.15 The curve of vehicle group trajectory separation changing with time when the maximum vehicle speed is 35km / h.

[0028] Fig.16 The curve of vehicle group trajectory separation changing with time when the maximum vehicle speed is 40km / h.

[0029] Fig.17 The curve of vehicle group trajectory separation changing with time when the maximum vehicle speed is 45km / h.

[0030] Fig.18 The curve of vehicle group trajectory separation changing with time when the maximum vehicle speed is 50km / h.

[0031] Fig.19 The curve of vehicle group trajectory consistency changing with time when the maximum speed of the vehicle is 25km / h.

[0032] Fig. 20 The curve of vehicle group trajectory consistency changing with time when the maximum speed of the vehicle is 30km / h.

[0033] Fig.21 The curve of vehicle group trajectory consistency changing with time when the maximum speed of the vehicle is 35km / h.

[0034] Fig. 22 The curve of vehicle group trajectory consistency changing with time when the maximum speed of the vehicle is 40km / h.

[0035] Fig.23 The curve of vehicle group trajectory consistency changing with time when the maximum speed of the vehicle is 45km / h.

[0036] Fig.24 The curve of vehicle group trajectory consistency changing with time when the maximum speed of the vehicle is 50km / h.

[0037] Fig.25 Schematic diagram of the process flow of the invention. DETAILED DESCRIPTION

[0038] The technical solution provided by the present application will be further described below in conjunction with specific embodiments and accompanying drawings. The advantages and features of the present application will become more apparent with the following description.

[0039] The "Method for Forming Unmanned Vehicle Groups Based on Trajectory Prediction" (applicant: Tongji University, patent application number: 2024100627793) applied by Cheng Jiujun and other inventors on January 16, 2024 is a preliminary achievement. Furthermore, the team has been working on the current problems of frequent changes in the surrounding environment, which easily cause changes in the internal structure of the vehicle group, and the failure to consider the changes in the future trajectory of the vehicle as the basis for dynamic evolution. The present invention discloses a technical solution for the collaborative method of unmanned vehicle groups, which is a further new achievement.

[0040] The present invention first proposes a dynamic evolution method for the stability of unmanned vehicle swarms, so that the unmanned vehicle swarms remain stable in four dynamic evolution events: node joining, node leaving, vehicle swarm merging, and vehicle swarm splitting; on this basis, the state of the vehicle swarm in the evolution process is modeled, and the unmanned vehicle swarm coordination model and the proof of the stability of the coordination model are given, and the unmanned vehicle swarm coordination method is given. Then, evaluation indicators such as vehicle swarm trajectory similarity, average vehicle swarm trajectory separation, node stability time, and node survival time are given, and simulation experiments are constructed to verify the stability and sustainability of the proposed unmanned vehicle swarm. Selecting unmanned vehicle nodes with similar future trajectories to form a vehicle swarm can reduce the impact of changes in the internal structure of the vehicle swarm caused by frequently changing interference factors around, thereby improving the stability and sustainability of the vehicle swarm, and providing a model and method for the intelligent coordination of the motion behavior of unmanned vehicle swarms, thereby effectively solving the problems of unmanned driving such as acceleration, lane changing, decision-making, and traffic congestion in current unmanned driving scenarios, and has good theoretical significance and application value for promoting the healthy and rapid development of unmanned vehicle swarms.

[0041] The present invention specifically includes the following three aspects:

[0042] 1. A method for dynamic evolution of unmanned vehicle swarm based on trajectory prediction, including:

[0043] 1.1 Node joining event;

[0044] 1.2 Node leaving event;

[0045] 1.3 Vehicle group merging event;

[0046] 1.4 Car group split incident;

[0047] 1.5 Dynamic evolution algorithm of driverless vehicle swarm.

[0048] 2 Unmanned vehicle group collaboration model, including:

[0049] 2.1 Relevant definitions;

[0050] 2.2 Unmanned vehicle group collaboration model;

[0051] 2.3 Unmanned vehicle swarm collaboration algorithm.

[0052] 3Simulation experiments and result analysis, including:

[0053] 3.1 Simulation experiment data and methods;

[0054] 3.2 Simulation experiment results and analysis.

[0055] Details are as follows:

[0056] The symbols used in this application are shown in Table 1.

[0057] Table 1

[0058]

[0059] Step 1: Dynamic evolution method of unmanned vehicle swarm based on trajectory prediction

[0060] 1.1 Node Joining Event

[0061] A node joining event refers to an event in which a new edge node appears in the autonomous driving scene or a node that leaves the original autonomous driving group joins the autonomous driving group. According to the group structure, there are mainly three types of node joining events:

[0062] 1) The edge node joins the vehicle group through the leading node, is upgraded to the collaborative node of the leading node, and a collaborative node joining event occurs;

[0063] 2) The edge node joins the vehicle group through the collaborative node, is upgraded to a follower node of the collaborative node, and a follower node joining event occurs;

[0064] 3) If there is no unmanned vehicle group in the neighborhood set of the edge node, it will elect a leader node with the edge nodes in the neighborhood set, and a leader node joining event will occur.

[0065] To become a collaborative node or a follower node, an edge node must meet the following requirements:

[0066]

[0067] Among them, v j Indicates the edge node to be added, v i Indicates v j The leading node or collaborative node in the neighborhood set, represents the group of vehicles at time t A set of leading nodes and collaborative nodes. represents the group of vehicles at time t After the node join event occurs, the vehicle group size limit N is not exceeded max , Indicates that the node joins the rear vehicle group G i The trajectory separation is less than the trajectory separation threshold.

[0068] The process of a node joining a swarm is as follows: Figure 1 As shown, the specific processing steps are as follows:

[0069] 1) After the unmanned vehicle group is formed, the leading node and the collaborative node periodically broadcast HELLO data packets containing their own node status information.

[0070] 2) When the leading or collaborative node P receives the HELLO data packet from the edge node Q, it updates the status information of the node Q in its neighborhood set.

[0071] 3) When the edge node Q receives the HELLO data packet from the leading or collaborative node P, it replies with a REQ_JOIN message requesting to join the vehicle group, while carrying its own status information; when the leading or collaborative node P receives the REQ_JOIN message from the node Q, it checks whether the node Q is in its neighborhood set, and determines whether the current vehicle group trajectory separation degree TSEP is less than the threshold TSEP th If both conditions are met, the server will reply with APROVE_JOIN to agree to join and update its subordinate node list subordinateNodes; otherwise, it will reply with DENY_JOIN to refuse to join.

[0072] 4) If node Q receives the APROVE_JOIN message from node P, node Q is upgraded to a collaborative or follower node according to the role of node P, and updates the leaderNode information of its superior node. After successfully joining the swarm, it no longer receives messages from other swarms.

[0073] The specific process of a node joining a group of driverless cars is shown in Algorithm 1:

[0074]

[0075] 1.2 Node leaving event

[0076] A node departure event refers to an event in which a node leaves the unmanned vehicle group due to a node failure caused by internal factors, a node disconnected from its parent node due to external factors, or a node leaves the environment. According to the definition of vehicle groups and vehicle roles, there are three main types of node departure events:

[0077] 1) The leading node leaves or fails, which has a global impact on the swarm topology and causes a swarm split event;

[0078] 2) The departure or failure of a coordinated node has a local impact on the vehicle group topology, resulting in a sub-vehicle group split event;

[0079] 3) The following node leaves or fails.

[0080] The group splitting events caused by the departure of the leading or coordinating node in situations (1) and (2) will be discussed in 1.3 Group Merging Events. Here we will first discuss situation (3) where the following node leaves, which satisfies:

[0081]

[0082] Among them, vj Indicates car group The follower node to leave, Indicates following node v j Not in the coordinated node v at time t i In the neighborhood set of i ,v j )∈E i Indicates following node v j By v i Join the car group

[0083] The process of a node leaving the vehicle group is as follows: Figure 2 As shown, the specific processing steps are as follows:

[0084] 1) After the formation of a swarm of driverless vehicles, the leading and collaborative nodes send point-to-point HEARTBEAT packets to their subordinate nodes for periodic heartbeat detection.

[0085] 2) When a node in the swarm receives a HEARTBEAT packet from its superior node leaderNode or subordinate node subordinateNodes, it updates the status information of the corresponding node.

[0086] 3) If the collaborative node P does not receive the heartbeat of the following node Q within the Timeout_Period period, it is considered that the node Q has left the vehicle group and the node Q is deleted from its subordinate node list subordinateNodes.

[0087] 4) If all the following nodes of the cooperative node leave the vehicle group, the identity of the cooperative node is still maintained.

[0088] 5) If the follower node Q does not receive the heartbeat of the collaborative node P within the Timeout_Period period, the node Q is downgraded to an edge node and the node joining algorithm 1 is executed.

[0089] The specific process of a node leaving the unmanned vehicle group is shown in Algorithm 2:

[0090]

[0091]

[0092] 1.3 Car group merging event

[0093] A merge event refers to the overlap of vehicle groups in an unmanned driving scenario. When the leading node or nodes between vehicle groups maintain a relatively stable motion state, a vehicle group merge event occurs. According to the node departure event defined in 1.2 Node departure event, it can be seen that in its life cycle, there are semi-centralized vehicle groups with leading nodes as the core, and centralized vehicle groups with collaborative nodes as the core.

[0094] When two swarms with a leading node as the core merge, due to the complexity of the swarm topology, some sub-swarms will split and nodes will leave, increasing the communication overhead of the swarm. Figure 3 As shown in the figure, if the right car group meets the conditions for merging into the left car group, according to the car group structure, the leading node Q of the right car group is downgraded to the collaborative node of the leading node P of the left car group, and the collaborative node of node Q is downgraded to its follower node, and all the follower nodes of the right car group are downgraded to edge nodes, resulting in a high degree of change in the topological structure of the car group. To address this issue, this application only considers the following two car group merging events, that is, satisfying:

[0095] 1) Two vehicle groups with collaborative nodes as the core merge to form a vehicle group with the leading node as the core;

[0096] 2) The vehicle group with the collaborative node as the core is merged into the vehicle group with the leading node as the core.

[0097]

[0098] in, Indicates the receiving vehicle group. represents the vehicle group to be merged, v i To receive the car group The leading node or collaborative node of v j For the group of vehicles to be merged The collaborative node Indicates v j In v i The neighborhood set of With car group Maintain a relatively stable state of motion. Indicates that the maximum size of the vehicle group has not been exceeded after the vehicle group merges. Indicates that the trajectory separation threshold is met after the vehicle group merges.

[0099] The merging process of concentrated vehicle groups into concentrated vehicle groups or semi-concentrated vehicle groups is as follows: Figure 4 and Figure 5 As shown, the specific processing steps are as follows:

[0100] 1) If Figure 4 As shown, if the collaborative node v i In the collaborative node vj In the neighborhood set of j Calculate the value of its leading probability function like Then v j To v i Send REQ_MERGE request to merge the message.

[0101] 2) If v i Received j REQ_MERGE request merge message, node v i Judgement j Is it in its neighborhood set, and v i calculate The trajectory separation degree TSEP after joining, if it meets the threshold VTSep th and the maximum size limit of the vehicle group, then v i To v j Send APROVE_MERGE to agree to merge, v i It is upgraded to a leading node, and its follower nodes are upgraded to collaborative nodes.

[0102] 3) If v j Receive APROVE_MERGE message to agree to merge, v j Update the leaderNode information of the upper node. Otherwise, node v j Continue broadcasting REQ_MERGE messages.

[0103] 4) If Figure 5 As shown, if the leading node v i In the collaborative node v j If the node v is in the neighborhood set of j To v i Send REQ_MERGE message to request the vehicle group to merge.

[0104] 5) If v i Received j REQ_MERGE request merge message, node v i Determine whether there is node v in its neighborhood set j , while v i calculate The trajectory separation degree TSEP after joining, if it meets the threshold VTSep th And the group of vehicles does not exceed the size limit, then v i To v j Send APROVE_MERGE to agree to merge, and lead node v i Update the subordinate node list subordinateNodes.

[0105] 6) If vj Receive APROVE_MERGE message to agree to merge, v j Update the leaderNode information of the upper node. Otherwise, node v j Continue broadcasting REQ_MERGE messages.

[0106] The specific process of merging a group of driverless vehicles is shown in Algorithm 3:

[0107]

[0108]

[0109] 1.4 Car group split incident

[0110] From the discussion of node departure of different vehicle roles in 1.2 Node departure event, we can know that when key nodes such as leading nodes or collaborative nodes leave or fail, it will have a serious impact on the topological structure of the vehicle group. At this time, a vehicle group split event occurs, which satisfies:

[0111]

[0112] Among them, v i Indicates car group The node to be left. Indicates that the node is a leading node or a collaborative node. Indicates that when node v i When a node leaves or fails, its neighbor set is empty.

[0113] The specific processing steps for the group splitting process caused by the departure of the leading node or the cooperative node are as follows:

[0114] 1) If Figure 6 As shown, if the car group The collaborative node v i The vehicle leaves the group due to a fault or leaves the environment, and its following node does not receive the node v within the Timeout_Preriod period. i After receiving the HEARTBEAT heartbeat packet, it is downgraded to an edge node and executes the node joining event in Algorithm 1. The scale is

[0115] 2) If Figure 7 As shown, if the car group The leading node v i When a vehicle leaves the group due to a fault or leaves the environment, its collaborative node does not receive a message from node v within the Timeout_Preriod period. i When the heartbeat of Split into |subordinateNodes i |-1 sub-car group, cooperative node v j ∈subordinateNoodes i Execute the vehicle group merging events of Algorithm 3 respectively.

[0116] The specific process of splitting the driverless car group is shown in Algorithm 4:

[0117]

[0118]

[0119] 1.5 Dynamic Evolution Algorithm of Autonomous Vehicle Swarm

[0120] Based on the swarm formation algorithm, combined with the node joining and leaving algorithms, namely Algorithms 1 and 2, and the swarm merging and splitting algorithms, namely Algorithms 3 and 4, a complete unmanned vehicle swarm dynamic evolution algorithm can be obtained. As shown in Algorithm 5, the specific processing process of the dynamic evolution algorithm is as follows:

[0121] 1) First, unmanned vehicles form a swarm through a swarm formation algorithm, and various dynamic evolution events occur at all times during the entire life cycle of the swarm (row 1).

[0122] 2) According to the dynamic evolution event type at the current moment, execute the corresponding dynamic evolution method and update the unmanned vehicle set and the vehicle group set (lines 2-13).

[0123] 3) Return the unmanned vehicle set and vehicle group set after dynamic evolution at the current moment (lines 14-15).

[0124]

[0125] Step 2: Unmanned vehicle group collaboration model

[0126] 2.1 Relevant definitions

[0127] Definition 1: GC: The attribute set of the unmanned vehicle group consists of the relevant attributes of the unmanned vehicle group in the collaborative model, which can be described as the following five-tuple:

[0128] GC=<V,G,S,H,P> (5)

[0129] in, Represents the set of all unmanned vehicle nodes; Represents the set of all vehicle groups; It represents the state set of the vehicle group, which consists of seven states: initial state S0, cooperative state S1, growth state S2, merging state S3, contraction state S4, split state S5, and extinction state S6; Represents a set of dynamic evolution events of a swarm, including four evolution events: node joining h1, node leaving h2, swarm merging h3, and swarm splitting h4; Represents the set of vehicle group trajectory separation degrees.

[0130] Definition 2 Initial state S0: The initial state is the state where the leader node of the unmanned vehicle group is successfully elected, that is, at time t, the edge node v i The state where the leader node election conditions are met and the election is successful. At this time, the leader node periodically processes the edge node's request to join the swarm. The initial state is a temporary state of the unmanned vehicle swarm collaboration state. At this time, the vehicle swarm trajectory separation degree TSEP i is 0, the status of the vehicle group attribute set is updated as follows:

[0131]

[0132] Definition 3 Collaborative state S1: At time t, when edge node v l The election of the leading node is successful and the edge node joins the swarm request to form swarm G. i , or car group G i During the dynamic evolution process, the nodes in the vehicle group maintain a relatively stable motion state, and the vehicle group trajectory separation degree TSEP i Maintain a stable change and do not exceed the threshold TSEP th , at this time, it is considered that the car group G i Reaching the collaborative state S1, which is also the trajectory separation TSEP i of the stable point.

[0133] Definition 4 Growth state S2: At time t, when the edge node v e Broadcast a request to join the car group and receive the car group G i Leading node or collaborative node The message of agreeing to join the car group is received. At this time, the car group has a node joining event h1, and the status of the car group attribute set is updated as follows:

[0134]

[0135] Definition 5: Merge state S3: At time t, if the vehicle group G with the cooperative node as the core j With driverless car group G i When the vehicle group merge event h3 occurs, the status of the vehicle group attribute set is updated as follows:

[0136]

[0137] Definition 6 Contraction state S4: At time t, if the vehicle group G i The following node v f After disconnecting from the collaborative node, the vehicle group experiences a node departure event h2. At this time, the state of the vehicle group attribute set is updated as follows:

[0138]

[0139] Definition 7 Split state S5: At time t, if the vehicle group G i Leading node or collaborative node Leave, causing the car group G i Split into two or more car groups, the state of the car group attribute set is updated as follows:

[0140]

[0141] Definition 8: Extinction state S6: With the car group G i The cooperative nodes and the following nodes leave one after another. If the vehicle group G i Only the leading node remains. At this time, the vehicle group dissipates and the leading node is downgraded to an edge node.

[0142] 2.2 Unmanned Vehicle Group Collaboration Model

[0143] After the unmanned vehicle group has a dynamic evolution event, it will eventually reach a coordinated state through the corresponding dynamic evolution method, and the vehicle group trajectory separation will reach a stable point. According to the vehicle group dynamic evolution method given in 1.5 Unmanned vehicle group dynamic evolution algorithm and the vehicle group attribute set and vehicle group state definition given in 2.1 Related definitions, the unmanned vehicle group collaborative model TPCM can be obtained, as follows:

[0144]

[0145] Among them, TSEP t-1 represents the separation degree of vehicle group trajectories at time t-1, h t ∈H represents the dynamic evolution event that triggers the vehicle group at time t, S t-1 ∈S represents the state of the vehicle group at time t-1.

[0146] In response to the evolution of the vehicle group, the unmanned vehicle group dynamically adjusts the vehicle group based on the collaborative model TPCM, and finally achieves the collaborative state of the vehicle group. The specific process is as follows: Figure 8 shown.

[0147] Proposition 1: The autonomous vehicle swarm collaboration model can ensure that the swarm is in a collaborative state.

[0148] prove:

[0149] 1) Before the unmanned vehicle group is formed, a leading node is elected. At this time, the vehicle group trajectory separation is 0, and the vehicle group reaches the initial state S0; after the unmanned vehicle group is formed, the nodes maintain the vehicle group by sending HEARTBEAT heartbeat packets to ensure that the vehicle group trajectory separation is not higher than the threshold and maintains a stable change. At this time, the vehicle group reaches the collaborative state S1.

[0150] 2) When a node joining event h1 occurs in the vehicle group, the vehicle group will correspondingly switch to the growth state S2 = S1 (h1). From Definition 7, it can be seen that the node joining event will only occur when the vehicle group trajectory separation threshold is met. At this time, the node joining evolution method is executed to adjust the state and finally reach the collaborative state S1.

[0151] 3) When a node leaving event h2 occurs in the vehicle group, the vehicle group will correspondingly switch to the contraction state S4 = S1 (h2) and execute the node leaving evolution method to adjust the state. If the vehicle group trajectory separation is lower than the threshold, the coordinated state S1 is reached. Otherwise, the nodes continue to leave until the coordinated state S1 is reached.

[0152] 4) When the vehicle group merges into a group event h3, the vehicle group will correspondingly switch to the merged state S3 = S1 (h3). From Definition 8, it can be seen that the vehicle group merge event will only occur when the vehicle group trajectory separation threshold is met. At this time, the vehicle group merge evolution method is executed to adjust the state and finally reach the collaborative state S1.

[0153] 5) When the vehicle group splits when the vehicle group splits h4, the vehicle group will correspondingly switch to the split state S5 = S1 (h4), and execute the vehicle group merging evolution method to adjust the state. If the vehicle group trajectory separation degree is lower than the threshold, the coordinated state S1 is reached, otherwise the vehicle group continues to split until it reaches the coordinated state S1.

[0154] 6) After the unmanned vehicle group reaches the coordinated state S1 from the initial state S0, under the action of limited dynamic evolution behavior, the vehicle group will correspondingly transfer to a limited vehicle group state and eventually reach a coordinated state.

[0155]

[0156] In summary, the unmanned vehicle group coordination model can ensure that the vehicle group is in a coordinated state S1 and achieve a stable state of the vehicle group.

[0157] 2.3 Autonomous Vehicle Group Collaboration Algorithm

[0158] The unmanned vehicle swarm collaboration algorithm realizes the self-organization and collaborative driving of the unmanned vehicle swarm in its life cycle, including guiding the sending and receiving of data packets between nodes in the unmanned vehicle swarm, and processing methods for the dynamic evolution events represented by the data packets. The specific data packet sending and receiving process and processing of the unmanned vehicle swarm collaboration algorithm are as follows: Fig. 9 As shown in Algorithm 6, the specific description is as follows:

[0159] 1) The leading node periodically broadcasts HELLO packets and sends point-to-point HEARTBEAT packets to its subordinate collaborative nodes; after receiving the HELLO packets, it updates its own neighborhood set information, and after receiving the HEARTBEAT packets, it updates its subordinate node list subordinateNodes information, processes the vehicle group merge after receiving REQ_MERGE, processes the node joining after receiving the REQ_JOIN message, and updates the subordinate node list subordinateNodes (lines 1-17).

[0160] 2) The collaborative node periodically broadcasts HELLO data packets and sends point-to-point HEARTBEAT heartbeat packets to its superior leader node and subordinate follower node; after receiving the HELLO data packet, it updates its own neighborhood set information, and after receiving the HEARTBEAT heartbeat packet, it updates its superior node leaderNode and subordinate node list subordinateNodes, and after receiving the REQ_JOIN message, it updates the subordinate node list subordinateNodes. If the leader node leaves and the superior node leaderNode of the collaborative node does not exist, the REQ_MERGE message is broadcast again periodically. If the REQ_MERGE message is received at this time, it is upgraded to the leader node and handles the vehicle group merge event. When the follower node leaves the vehicle group, the subordinate node list subordinateNodes is updated (lines 18-30).

[0161] 3) The follower node periodically broadcasts HELLO packets, sends point-to-point HEARTBEAT packets to its superior collaborative node, updates its own neighborhood set information after receiving the HELLO packets, and updates its superior node leaderNode after receiving the HEARTBEAT packets. If the collaborative node leaves and the superior node leaderNode of the follower node does not exist, the follower node periodically broadcasts REQ_JOIN messages (lines 31-34).

[0162]

[0163]

[0164]

[0165] Step 3: Simulation experiment and result analysis

[0166] This application conducts simulation experiments by simulating experimental scenarios and vehicle communication behaviors through SUMO and NS3 to verify the effectiveness of the TPCM model.

[0167] (1) Simulation experiment data and methods

[0168] In this application, SUMO is used to construct a "field"-shaped road with a size of 2 km * 1 km, and a total of 200 vehicles are put into the simulation. In the simulation experiment parameters, the trajectory separation threshold TSEP th is set to 60, the maximum speed of the driverless vehicle is limited within the range of 25 km / h - 50 km / h, and the simulation time is set to 200 s. The specific simulation experiment parameter settings are shown in Table 2 below.

[0169] Table 2

[0170]

[0171]

[0172] To verify the effectiveness of the TPCM model, this application uses the following four indicators to conduct a comparative experiment on the TPCM model and the SCCDEM model. SCCDEM is the latest collaborative model for driverless vehicle swarms, and this model has the optimal performance compared to other existing models.

[0173] 1) Swarm trajectory consistency: It represents the average trajectory similarity difference degree between the nodes of the driverless vehicle swarm. The smaller the average trajectory similarity difference between nodes, the higher the swarm trajectory consistency and the higher the overall stability of the swarm. Specifically as follows:

[0174]

[0175] Among them, represents the average trajectory similarity of a node relative to all nodes in the swarm, which is defined by formula (14); |V| represents the number of nodes in the driverless vehicle swarm.

[0176]

[0177] 2) Average swarm trajectory separation: It represents the average swarm trajectory separation size per unit time during the life cycle of all swarms. This indicator can reflect the overall persistence of the swarm.

[0178] 3) Node stable time: It represents the average survival time of the driverless vehicle in each driverless vehicle swarm. The longer the node stable time, the longer the duration of the node in each swarm, and the better the effectiveness of the swarm evolution method.

[0179] 4) Node survival time: It represents the time when the driverless vehicle participates in the swarm activities during the entire simulation experiment. The longer the node survival time, the more nodes are covered by the swarm, the higher the efficiency of the nodes joining the swarm, and the better the effectiveness of the dynamic evolution method.

[0180] (2) Analysis of simulation and experimental results

[0181] In order to evaluate the performance of unmanned vehicles in various motion speed scenarios, the maximum speed of unmanned vehicles was gradually increased from 25km / h to 50km / h at intervals of 5km / h, and a total of 6 representative speed scenarios were selected. Simulation experiments were conducted in each speed scenario, and the node participation rate, node survival time, vehicle group trajectory consistency and average vehicle group trajectory separation of the vehicle group were statistically analyzed to evaluate the effectiveness of the TPCM model and the SCCDEM model in the dynamic evolution method and the stability of the vehicle group.

[0182] 1) Effectiveness of the autonomous vehicle swarm evolution method

[0183] Fig.10 and Fig.11 These are the relationships between the node survival time and node stability time and the maximum speed of the unmanned vehicle in the TPCM model and the SCCDEM model. The node survival time reflects the time that the node is in the swarm during the simulation cycle. A longer node survival time means that the node is more efficient in joining the swarm and the effectiveness of the swarm evolution method is better. Fig.10 As shown in the figure, the node survival time of the TPCM model represented by the red line remains around 166.2s with the increase of the maximum speed of the unmanned vehicle, with a small fluctuation range, and reaches a minimum value of 163.54s when the unmanned vehicle is 40km / h. The node survival time of the SCCDEM model represented by the green line remains stable as a whole, reaching a maximum value of 152s when the speed of the unmanned vehicle is 45km / h, which is much smaller than the minimum node survival time of 163.54s of the TPCM model. In summary, both the TPCM and SCCDEM models are suitable for different speed scenarios, but the TPCM model has better performance in different speed scenarios, and the node survival time is improved by 11.55% overall, indicating that the effectiveness of the TPCM model is better than that of the SCCDEM model.

[0184] The node stability time reflects the average survival time of the unmanned vehicle in each swarm. A higher node stability time means that the node lasts longer in the swarm and the more effective the unmanned vehicle swarm dynamic evolution method is. Fig.11It can be seen that the node stabilization time of the TPCM model represented by the red line is maintained at around 63.1s as a whole, and shows a small fluctuation, reaching a maximum value of 69.7s and a minimum value of 58.34s when the maximum speed of the unmanned vehicle is 25km / h and 35km / h, respectively. The node stabilization time of the SCCDEM model represented by the green line reaches a minimum value of 32.4s when the maximum speed of the unmanned vehicle is 50km / h, and can be maintained above 33.7s in other speed scenarios. Among them, the maximum value reaches 39.94s when the maximum speed of the unmanned vehicle is 45km / h, which is much smaller than the minimum value of 58.34s of the node stabilization time of the TPCM model. In summary, the node stabilization time of the TPCM model is improved by 71.78% compared with the SCCDEM model, indicating that the TPCM collaborative model is more effective in the dynamic evolution of the vehicle group.

[0185] 2) Sustainability of driverless fleets

[0186] The average vehicle group trajectory separation reflects the overall sustainability of the vehicle group model. The lower the average vehicle group trajectory separation, the more similar the future trajectories between the vehicle group nodes, the more stable the communication link, and the higher the overall sustainability of the vehicle group collaboration model. In this simulation experiment, the maximum speed of the unmanned vehicle is set in the range of 25km / h to 50km / h, increasing by 5km / h intervals. Six motion scenes covering different speeds are selected to simulate the TPCM and SCCDEM models. First, the average vehicle group trajectory separation of the unmanned vehicle group in different speed scenes is statistically analyzed, and then for each speed scene, the vehicle group trajectory separation of the vehicle group during the entire simulation experiment is analyzed.

[0187] Fig.12 The relationship between the average vehicle group trajectory separation and the maximum speed of the driverless vehicle in the TPCM model and the SCCDEM model is shown, where the red line is the TPCM model and the green line is the SCCDEM model. Fig.12 It can be seen that the average vehicle group trajectory separation of the TPCM model shows an overall trend of first rising and then falling, reaching a maximum value of 26.56 when the maximum speed of the driverless vehicle is 40km / h, and maintaining at around 22.96 at other times. The average vehicle group trajectory separation of the SCCDEM model is the same as the TPCM model, showing a trend of first rising and then falling, reaching a minimum value of 30.87 and a maximum value of 39.8 when the maximum speed of the driverless vehicle is 25km / h and 40km / h, respectively, which is much larger than the maximum value of 25.56 of the TPCM model. In summary, the TPCM and SCCDEM models perform well in both low-speed and high-speed scenarios. The average trajectory separation of the TPCM model has increased by 32.52% overall, indicating that the TPCM model has a better overall effect in different speed scenarios, indicating that the TPCM model has higher sustainability.

[0188] like Fig.13 As shown in the figure, the red line and the green line represent the curves of the TPCM model and the SCCDEM model respectively, when the maximum speed of the unmanned vehicle is 25km / h, of the vehicle group trajectory separation degree as a function of simulation time. As can be seen from the figure, in the time periods of 316s-320s, 349s-350s and 372s-374s in the middle and late stages of the simulation, the vehicle group trajectory separation degree of the TPCM model is not much different from that of the SCCDEM model, and is lower than that of the SCCDEM model at other times. The vehicle group trajectory separation degree of the TPCM model is reduced by 31.25% overall. The simulation experiment results show that in low-speed scenarios, the TPCM model can enable the vehicle group to maintain a high degree of sustainability as a whole.

[0189] like Fig.14 As shown in the figure, the red line and the green line represent the curves of the TPCM model and the SCCDEM model respectively, when the maximum speed of the unmanned vehicle is 30km / h, where the vehicle group trajectory separation changes with the simulation time. As can be seen from the figure, the vehicle group trajectory separation of the TPCM model maintains a stable change with the simulation time, while the SCCDEM model shows a significant upward trend. The vehicle group trajectory separation of the TPCM model is much lower than that of the SCCDEM model in the middle and late stages of the simulation, and the vehicle group trajectory separation is reduced by 35.04% overall. The simulation experiment results show that the TPCM model can maintain a high level of vehicle group continuity in medium and low speed scenarios.

[0190] like Fig.15 As shown in the figure, the red line and the green line represent the curves of the TPCM model and the SCCDEM model respectively, when the maximum speed of the unmanned vehicle is 35km / h, of the vehicle group trajectory separation changing with the simulation time. As can be seen from the figure, in the 390s-412s period of the later simulation, the vehicle group trajectory separation of the TPCM model is slightly higher than that of the SCCDEM model, but at most times the TPCM model achieves a lower vehicle group trajectory separation, and the vehicle group trajectory separation of the TPCM model is reduced by 29.68% overall. The simulation experiment results show that in the medium-speed scenario, the TPCM model can ensure the continuity of the vehicle group.

[0191] like Fig.16As shown in the figure, the red line and the green line represent the curves of the TPCM model and the SCCDEM model respectively, when the maximum speed of the unmanned vehicle is 40km / h, of the vehicle group trajectory separation degree as a function of simulation time. As can be seen from the figure, in the 261s-325s period of the early and middle stages of the simulation, the vehicle group trajectory separation degree of the TPCM model is not much different from that of the SCCDEM model. After that, the vehicle group trajectory separation degree of the TPCM model is much lower than that of the SCCDEM model, and the vehicle group trajectory separation degree of the TPCM model is reduced by 33.29% overall. The simulation experiment results show that in the medium-speed scenario, the TPCM model can maintain a high degree of continuity of the vehicle group.

[0192] like Fig.17 As shown in the figure, the red line and the green line represent the curves of the vehicle group trajectory separation of the TPCM model and the SCCDEM model respectively when the maximum speed of the unmanned vehicle is 45km / h. It can be seen from the figure that in the 281s-310s period at the beginning of the simulation, the vehicle group trajectory separation of the TPCM model is slightly lower than that of the SCCDEM model. After that, the vehicle group trajectory separation of the TPCM model is lower than that of the SCCDEM model. The vehicle group trajectory separation of the TPCM model is reduced by 30.03% overall. The simulation experiment results show that in the medium and high speed scenario, the TPCM model can achieve a lower vehicle group trajectory separation.

[0193] like Fig.18 As shown in the figure, the red line and the green line represent the curves of the TPCM model and the SCCDEM model respectively, when the maximum speed of the unmanned vehicle is 50km / h, of the vehicle group trajectory separation degree as a function of simulation time. As can be seen from the figure, in the early and late stages of the simulation, the vehicle group trajectory separation degree of the TPCM model is not much different from that of the SCCDEM model. At other times, the vehicle group trajectory separation degree of the TPCM model is much lower than that of the SCCDEM model. The vehicle group trajectory separation degree is reduced by 35.56% overall, and the SCCDEM model reaches a maximum value of 54.5 at 409s of simulation. The simulation experiment results show that in high-speed scenarios, the TPCM model can achieve a higher vehicle group persistence.

[0194] In summary, under different speed scenarios, the TPCM model can achieve a lower vehicle group trajectory separation, and the overall sustainability of the vehicle group is higher.

[0195] 3) Stability of driverless vehicle swarms

[0196] The consistency of the vehicle group trajectory reflects the overall stability of the vehicle group. A higher consistency of the vehicle group trajectory means that the higher the average trajectory similarity between the vehicle group nodes, the higher the stability of the vehicle group. This simulation experiment was conducted in the range of 25km / h-50km / h, with an increase of 5km / h. Six motion scenes covering different speeds were selected to simulate the TPCM and SCCDEM models, and statistical analysis was performed on the vehicle group trajectory consistency of the unmanned vehicle group in different speed scenes.

[0197] like Fig.19 As shown in the figure, when the maximum speed of the unmanned vehicle is 25km / h, the vehicle group trajectory consistency of the TPCM model represented by the red line is generally stable at around 0.986, and fluctuates in a small range. The vehicle group trajectory consistency of the SCCDEM model represented by the green line is maintained at around 0.979 in the 221s-365s period of the early simulation, and then begins to decline rapidly, and is much lower than the TPCM model during the entire simulation process. The vehicle group trajectory consistency of the TPCM model has been improved by 2.36% overall. The simulation experiment results show that the TPCM model has obvious advantages in low-speed scenarios.

[0198] like Fig. 20 As shown in the figure, when the maximum speed of the unmanned vehicle is 30km / h, the vehicle group trajectory consistency of the TPCM model represented by the red line is higher than that of the SCCDEM model represented by the green line during the entire simulation time, and maintains a steady upward trend. The vehicle group trajectory consistency of the TPCM model is improved by 2.15% overall. The simulation experiment results show that in medium and low speed scenarios, the TPCM model can maintain a high stability of the vehicle group.

[0199] like Fig.21 As shown in the figure, when the maximum speed of the unmanned vehicle is 35km / h, the vehicle group trajectory consistency of the TPCM model represented by the red line is not much different from that of the SCCDEM model represented by the green line in the 241s-252s period at the beginning of the simulation, and is much higher than the SCCDEM model at other times thereafter. The vehicle group trajectory consistency of the TPCM model has been improved by 2.01% overall. The simulation experiment results show that in the medium-speed scenario, the TPCM model can maintain a high stability of the vehicle group.

[0200] like Fig. 22 As shown in the figure, when the maximum speed of the unmanned vehicle is 40km / h, the vehicle group trajectory consistency of the TPCM model represented by the red line is not much different from that of the SCCDEM model represented by the green line in the 241s-249s period at the beginning of the simulation, and is higher than the SCCDEM model at other times thereafter. The SCCDEM model shows a larger range of fluctuations, and the vehicle group trajectory consistency of the TPCM model is improved by 2.18% overall. The simulation experiment results show that in the medium-speed scenario, the TPCM model can ensure the stability of the vehicle group.

[0201] like Fig.23 As shown in the figure, when the maximum speed of the unmanned vehicle is 45km / h, the vehicle group trajectory consistency of the TPCM model represented by the red line is higher than that of the SCCDEM model represented by the green line, and both show a stable upward trend. The vehicle group trajectory consistency of the TPCM model is improved by 2.15% overall. The simulation experiment results show that in high-speed scenarios, the TPCM model can maintain a high vehicle group stability.

[0202] like Fig.24 As shown in the figure, when the maximum speed of the driverless vehicle is 50km / h, the vehicle group trajectory consistency of the TPCM model represented by the red line remains at around 0.986, and shows a stable upward trend. The vehicle group trajectory consistency of the SCCDEM model represented by the green line remains at around 0.965, showing a relatively large range of fluctuations. The vehicle group trajectory consistency of the TPCM model has increased by 2.17% overall. The simulation experiment results show that in high-speed scenarios, the overall stability of the TPCM model vehicle group is higher.

[0203] In summary, under different speed scenarios, the TPCM model can achieve a high consistency of vehicle group trajectories, and the overall stability of the vehicle group is higher.

[0204] In summary, compared with the SCCDEM model, the vehicle group trajectory consistency of the TPCM model is improved by 2.17%, the average vehicle group trajectory separation is reduced by 32.52%, and the node survival and stability time are increased by 11.55% and 71.78% respectively, indicating that the TPCM model improves the effectiveness of the dynamic evolution method of the unmanned vehicle group and the stability of the vehicle group.

[0205] The above description is only a description of the preferred embodiments of the present application, and is not intended to limit the scope of the present application. Any changes or modifications made by any person skilled in the art based on the above disclosed technical contents shall be deemed as equivalent effective embodiments and shall fall within the scope of protection of the technical solution of the present application.

Claims

1. A collaborative method for unmanned vehicle groups based on trajectory prediction, characterized in that: It consists of two parts: Part 1: Design a dynamic evolution method for the stability of unmanned vehicle swarms, so that the unmanned vehicle swarms can remain stable in four dynamic evolution events: node joining, node leaving, swarm merging, and swarm splitting; Part 2: Based on the first part, the state of the vehicle group in the evolution process is modeled, the unmanned vehicle group coordination model is constructed, and the unmanned vehicle group coordination algorithm is given.

2. The unmanned vehicle group collaboration method based on trajectory prediction according to claim 1, characterized in that: In the first part, the four dynamic evolution events are as follows: Node joining event: A node joining event refers to an event in which a new edge node appears in the autonomous driving scene or a node that leaves the original autonomous driving group joins the autonomous driving group. There are three situations: (1) The edge node joins the swarm through the leading node and is upgraded to a collaborative node of the leading node, and a collaborative node joining event occurs; (2) The edge node joins the vehicle group through the collaborative node, is upgraded to a follower node of the collaborative node, and a follower node joining event occurs; (3) If there is no unmanned vehicle group in the neighborhood set of the edge node, it will elect a leader node with the edge nodes in the neighborhood set, and a leader node joining event will occur; To become a collaborative node or a follower node, an edge node must meet the following requirements: Among them, v j Indicates the edge node to be added, v i Indicates v j The leading node or collaborative node in the neighborhood set, represents the group of vehicles at time t A set of leading nodes and collaborative nodes. represents the group of vehicles at time t After the node join event occurs, the vehicle group size limit N is not exceeded max , Indicates that the node joins the rear vehicle group G i The trajectory separation degree is less than the trajectory separation degree threshold; Algorithm 1 is the process of a node joining a vehicle group. The specific processing steps are as follows: (1) After the autonomous vehicle group is formed, the leading node and the cooperative node periodically broadcast HELLO packets containing their own node status information; (2) When the leading or collaborative node P receives the HELLO data packet from the edge node Q, it updates the status information of the node Q in its neighborhood set; (3) When the edge node Q receives the HELLO data packet from the leading or collaborative node P, it replies with a REQ_JOIN message requesting to join the vehicle group, while carrying its own status information; when the leading or collaborative node P receives the REQ_JOIN message from node Q, it checks whether node Q is in its neighborhood set and determines whether the current vehicle group trajectory separation degree TSEP is less than the threshold TSEP th If both conditions are met, the server will reply with APROVE JOIN to agree to join and update its subordinate node list subordinateNodes; otherwise, the server will reply with DENY JOIN to refuse to join. (4) If node Q receives the APROVE_JOIN message from node P, node Q is upgraded to a collaborative or follower node according to the role of node P, and updates the leaderNode information of its superior node. After successfully joining the swarm, it no longer receives messages from other swarms. Node leaving event: A node departure event refers to an event in which a node leaves the unmanned vehicle group due to a node failure caused by internal factors, a node disconnected from a parent node due to external factors, or a node leaves the environment. According to the definition of vehicle groups and vehicle roles, there are three types of node departure events: (1) The leader node leaves or fails, which has a global impact on the swarm topology and causes a swarm split event; (2) The departure or failure of a coordinated node has a local impact on the swarm topology, resulting in a sub-swarm split event; (3) The following node leaves or fails; Among them, case (3) follows the node leaving event, that is, it satisfies: Among them, v j Indicates car group The follower node to leave, Indicates following node v j Not in the coordinated node v at time t i In the neighborhood set of i , v j )∈E i Indicates following node v j By v i Join the car group Algorithm 2 is the process of a node leaving the vehicle group. The specific processing steps are as follows: (1) After the formation of a swarm of unmanned vehicles, the leading and collaborative nodes send point-to-point HEARTBEAT packets to their subordinate nodes for periodic heartbeat detection; (2) When a node in the swarm receives a HEARTBEAT packet from its superior node leaderNode or subordinate node subordinateNodes, it updates the status information of the corresponding node; (3) If the collaborative node P does not receive the heartbeat of the following node Q within the Timeout_Period period, it is considered that node Q has left the vehicle group and node Q is deleted from its subordinate node list subordinateNodes; (4) If all the following nodes of the cooperative node leave the group, the identity of the cooperative node is still maintained; (5) If the follower node Q does not receive the heartbeat of the collaborative node P within the Timeout_Period period, the node Q is downgraded to an edge node and the node joining algorithm l is executed; Vehicle group merging event: A merge event refers to the overlap of vehicle groups in an unmanned driving scenario. When the leading node or nodes between vehicle groups maintain a relatively stable motion state, a vehicle group merge event occurs. According to the node departure event, it can be seen that in the life cycle of a vehicle group, there are semi-centralized vehicle groups with leading nodes as the core, and centralized vehicle groups with collaborative nodes as the core. When two vehicle groups with the leading node as the core merge, the following conditions are met: (1) Two vehicle groups with collaborative nodes as the core merge to form a vehicle group with the leading node as the core; (2) The vehicle group with the collaborative node as the core is merged into the vehicle group with the leading node as the core; in, Indicates the receiving vehicle group. represents the vehicle group to be merged, v i To receive the car group The leading node or collaborative node of v j For the group of vehicles to be merged The collaborative node Indicates v j In v i The neighborhood set of car group With car group Maintain a relatively stable state of motion. Indicates that the maximum size of the vehicle group has not been exceeded after the vehicle group merges. Indicates that the trajectory separation threshold is met after the vehicle group merges; Algorithm 3 is the process of merging a centralized vehicle group into a centralized vehicle group or a semi-centralized vehicle group. The specific processing steps are as follows: (1) If the collaborative node v i In the collaborative node v j In the neighborhood set of j Calculate the value of its leading probability function like Then v j To v i Send REQ_MERGE request to merge the message; (2) If v i Received j REQ_MERGE request merge message, node v i Judgement j Is it in its neighborhood set, and v i calculate The trajectory separation degree TSEP after joining, if it meets the threshold VTSep th and the maximum size limit of the vehicle group, then v i To v j Send APROVE_MERGE to agree to merge, v i Upgrade to a leading node, and its follower nodes are upgraded to collaborative nodes; (3) If v j Receive APROVE_MERGE message to agree to merge, v j Update the leaderNode information of the upper node; otherwise, node v j Continue broadcasting REQ_MERGE messages; (4) If the leading node v i In the collaborative node v j If the node v is in the neighborhood set of j To v i Send REQ_MERGE message to request the vehicle group to merge; (5) If v i Received j REQ_MERGE request merge message, node v i Determine whether there is node v in its neighborhood set j , while v i calculate The trajectory separation degree TSEP after joining, if it meets the threshold VTSep th And the group of vehicles does not exceed the size limit, then v i To v j Send APROVE_MERGE to agree to merge, and lead node v i Update subordinate node list subordinateNodes; (6) If v j Receive APROVE_MERGE message to agree to merge, v j Update the leaderNode information of the upper node; otherwise, node v j Continue broadcasting REQ_MERGE messages; Car group split incident: From the node departure situation of different vehicle roles in the node departure event, it can be seen that when key nodes such as leading nodes or collaborative nodes leave or fail, it will have a serious impact on the vehicle group topology structure. At this time, a vehicle group split event occurs, which satisfies: Among them, v i Indicates car group The node to be left, Indicates that the node is a leading node or a collaborative node. Indicates that when node v i When leaving or failing, its neighborhood set is empty; Algorithm 4 is the process of splitting the vehicle group caused by the departure of the leading node or the cooperative node. The specific processing steps are as follows: (1) If the vehicle group The collaborative node v i The vehicle leaves the group due to a fault or leaves the environment, and its following node does not receive the node v within the Timeout_Preriod period. i After receiving the HEARTBEAT heartbeat packet, it is downgraded to an edge node and executes the node joining event in Algorithm 1. The scale is (2) If the vehicle group The leading node v i When a vehicle leaves the group due to a fault or leaves the environment, its collaborative node does not receive a message from node v within the Timeout_Preriod period. i When the heartbeat of Split into |subordinateNodes i |-1 sub-car group, cooperative node v j ∈subordinateNodes i Execute the vehicle group merging events of Algorithm 3 respectively.

3. The unmanned vehicle group collaboration method based on trajectory prediction according to claim 2, characterized in that: In the first part, the dynamic evolution algorithm of the unmanned vehicle group: Based on the swarm formation algorithm, combined with the node joining and leaving algorithms, namely Algorithms 1 and 2, and the swarm merging and splitting algorithms, namely Algorithms 3 and 4, a complete unmanned vehicle swarm dynamic evolution algorithm, namely Algorithm 5, is obtained; the specific processing process of the dynamic evolution algorithm is as follows: (1) First, unmanned vehicles form a swarm through a swarm formation algorithm, and various dynamic evolution events occur at all times during the entire life cycle of the swarm; (2) According to the dynamic evolution event type at the current moment, execute the corresponding dynamic evolution method and update the unmanned vehicle set and the vehicle group set; (3) Return the unmanned vehicle set and vehicle group set after dynamic evolution at the current moment.

4. The unmanned vehicle group collaboration method based on trajectory prediction according to claim 1, characterized in that: The second part: Related definitions Definition 1: Vehicle group attribute set GC: The unmanned vehicle group attribute set is composed of the relevant attributes of the unmanned vehicle group in the collaborative model, which is described as the following five-tuple: GC=<V,G,S,H,P> (5) in, Represents the set of all unmanned vehicle nodes; Represents the set of all vehicle groups; It represents the state set of the vehicle group, which consists of seven states: initial state S0, cooperative state S1, growth state S2, merging state S3, contraction state S4, split state S5, and extinction state S6; Represents a set of dynamic evolution events of a swarm, including four evolution events: node joining h1, node leaving h2, swarm merging h3, and swarm splitting h4; represents the set of vehicle group trajectory separation degrees; Definition 2 Initial state S0: The initial state is the state where the leader node of the unmanned vehicle group is successfully elected, that is, at time t, the edge node v i The state where the leader node election conditions are met and the election is successful. At this time, the leader node periodically processes the edge node's request to join the swarm. The initial state is a temporary state of the unmanned vehicle swarm collaboration state. At this time, the vehicle swarm trajectory separation degree TSEP i is 0, the status of the vehicle group attribute set is updated as follows: Definition 3 Collaborative state S1: At time t, when edge node v l The election of the leading node is successful and the edge node joins the swarm request to form swarm G. i , or car group G i During the dynamic evolution process, the nodes in the vehicle group maintain a relatively stable motion state, and the vehicle group trajectory separation degree TSEP i Maintain a stable change and do not exceed the threshold TSEP th , at this time, it is considered that the car group G i Reaching the collaborative state S1, which is also the trajectory separation TSEP i The stable point; Definition 4 Growth state S2: At time t, when the edge node v e Broadcast a request to join the car group and receive the car group G i Leading node or collaborative node The message of agreeing to join the car group is received. At this time, the car group has a node joining event h1, and the status of the car group attribute set is updated as follows: Definition 5: Merge state S3: At time t, if the vehicle group G with the cooperative node as the core j With driverless car group G i When the vehicle group merge event h3 occurs, the status of the vehicle group attribute set is updated as follows: Definition 6 Contraction state S4: At time t, if the vehicle group G i The following node v f After disconnecting from the collaborative node, the vehicle group experiences a node departure event h2. At this time, the state of the vehicle group attribute set is updated as follows: Definition 7 Split state S5: At time t, if the vehicle group G i Leading node or collaborative node Leave, causing the car group G i Split into two or more car groups, the state of the car group attribute set is updated as follows: Definition 8: Extinction state S6: With the car group G i The cooperative nodes and the following nodes leave one after another. If the vehicle group G i Only the leading node remains. At this time, the vehicle group dissipates and the leading node is downgraded to an edge node. Unmanned vehicle group collaboration model: After the unmanned vehicle group has a dynamic evolution event, it will eventually reach a coordinated state through the corresponding dynamic evolution method. At this time, the vehicle group trajectory separation reaches a stable point. According to the unmanned vehicle group dynamic evolution algorithm and the vehicle group attribute set and vehicle group state definition given in the above related definitions, the unmanned vehicle group coordination model TPCM is obtained, which is as follows: Among them, TSEP t-1 represents the separation degree of vehicle group trajectories at time t-1, h t ∈H represents the dynamic evolution event that triggers the vehicle group at time t, S t-1 ∈S represents the state of the vehicle group at time t-1; In response to the evolving behavior of the vehicle group, the unmanned vehicle group dynamically adjusts the vehicle group based on the collaborative model TPCM to achieve the collaborative state of the vehicle group.

5. The unmanned vehicle group collaboration method based on trajectory prediction according to claim 4, characterized in that: In the second part, the unmanned vehicle group collaboration algorithm: The unmanned vehicle group cooperative algorithm realizes the self-organization and cooperative driving of the unmanned vehicle group in its life cycle, including guiding the sending and receiving of data packets between nodes of the unmanned vehicle group, and processing methods of the dynamic evolution events represented by the data packets; the specific data packet sending and receiving process and processing of the unmanned vehicle group cooperative algorithm are respectively Algorithm 6, which is described as follows: (1) The leading node periodically broadcasts HELLO packets and sends point-to-point HEARTBEAT packets to its subordinate collaborative nodes; after receiving the HELLO packets, it updates its own neighborhood set information, and after receiving the HEARTBEAT packets, it updates its subordinate node list subordinateNodes information, processes the vehicle group merge after receiving the REQ_MERGE message, processes the node joining after receiving the REQ_JOIN message, and updates the subordinate node list subordinateNodes; (2) The collaborative node periodically broadcasts HELLO packets and sends point-to-point HEARTBEAT packets to its superior leader node and subordinate follower node. After receiving the HELLO packet, it updates its own neighborhood set information. After receiving the HEARTBEAT packet, it updates its superior node leaderNode and subordinate node list subordinateNodes. After receiving the REQ_JOIN message, it updates the subordinate node list subordinateNodes. If the leader node leaves and the superior node leaderNode of the collaborative node does not exist, it periodically broadcasts the REQ_MERGE message. If it receives the REQ_MERGE message, it is upgraded to the leader node and handles the vehicle group merge event. When the follower node leaves the vehicle group, it updates the subordinate node list subordinateNodes. (3) The follower node periodically broadcasts HELLO packets and sends point-to-point HEARTBEAT packets to its superior collaborative node. After receiving the HELLO packets, it updates its own neighborhood set information. After receiving the HEARTBEAT packets, it updates its superior node leaderNode. If the collaborative node leaves and the superior node leaderNode of the follower node does not exist, the follower node periodically broadcasts REQ_JOIN messages.

Images